Ausgewählte Beiträge

Gier, Nadine R.

doi:10.1007/978-3-658-38206-3_4

Nadine R. Gier²

3727 Accesses

Zusammenfassung

In ausgewählten Beiträgen werden Käufer- und Konsumentenentscheidungsprozesse anhand verschiedener methodischer, neurowissenschaftlich fundierter Herangehensweisen empirisch untersucht, um die Entscheidungsprozesse umfassend beschreiben, effektiver unterstützen und erfolgreich vorhersagen zu können.

You have full access to this open access chapter, Download chapter PDF

4.1 Beitrag 1: Affecting consumers: A fMRI study on regulatory focus framed information in the field of animal welfare

Dieser Beitrag wurde in abgewandelter Form bei der „North American Conference of the Association for Consumer Research“ 2017 im Rahmen einer Poster Session vorgestellt und ist als „Short Abstract“ veröffentlicht worden: Gier, N., Krampe, C. & Kenning, P. (2017). 5-A: Affecting Consumers: A fMRI study on regulatory focus framed information in the field of animal welfare. In Gneezy, A., Griskevicius, V. & Williams, P. (Hrsg.), NA – Advances in Consumer Research (45. Aufl., S. 1028). Association for Consumer Research.

4.1.1 Abstract

Applying regulatory focus theory to animal welfare information, we show that promotion focus framed information elicits greater subjective liking, indicated by increased neural activity in the vmPFC. Moreover, framed information influences neural processing of subsequent information demonstrated by a greater activity in ACC, an effect not seen on behavioural level.

4.1.2 Extended abstract

To better understand human decision-making, regulatory focus theory assumes that two social-cognitive motivational systems drive intention, evaluation and behaviour of consumers (Crowe & Higgins, 1997; Motyka et al., 2014). Since theoretical constructs such as ‘intention’ can only be hypothetically inferred and are neither observable nor measurable (Kenning & Plassmann, 2005), neuroscience has the potential to contribute to consumer research by specifying underlying neural processes associated with these constructs (Plassmann, Venkatraman, Huettel, & Yoon, 2015, Kosslyn, 1999). Furthermore, by means of neuroscientific techniques changed processing mechanisms can probably be detected on a pre-actional level (Riedl, Mohr, Kenning, Davis, & Heekeren, 2014). For example, by applying individual adjusted stimuli in combination with the inherent regulatory focus orientation of participants, pioneer studies utilise fMRI to reveal brain regions associated with emotion and motivation processes (Cunningham, Raye, & Johnson, 2005; Eddington, Dolcos, Cabeza, Krishnan, & Strauman, 2007).

In contrast to the mentioned studies, our work does not focus on personal constructed regulatory focus primes and its fit with the individual inherent orientation, but rather applies the regulatory focus theory in the context of goal-framing (Levin, Gaeth, Schreiber, & Lauriola, 2002; Levin et al., 1998). Against this background, the current study aims to identify the neural mechanism and processing changes of the valuation of regulatory focus framed information. The topic of animal welfare was chosen, since both framing types can be reliably formulated (Franz, Meyer, & Spiller, 2010). As a result, our research not only contributes neuroscientific evidence to goal-framing and the regulatory focus theory but also reveals potential for effective marketing communication strategies in the context of animal welfare. Following these deliberations, two hypotheses are derived:

H₁: Promotion compared to prevention regulatory focus framed animal welfare information elicits greater neural activity in valuation areas, such as the vmPFC or associated brain regions.
H₂: The regulatory focus framed information affects processing of subsequent non-framed information, indicated by an increased activity in the ACC after compared to the same information prior the regulatory focus manipulation.

4.1.2.1 Method

To test our hypotheses, a fMRI study was conducted using different regulatory focus framed animal welfare information. Information regarding animal welfare were formulated in goal-frames, either stressing the promotion or prevention regulatory focus, serving as information stimuli in our study. Stimuli that were pre-tested on reliability and valuation, whereby promotion information was significantly more positive rated than prevention information (t(83) = 8.697, p < 0.001), suggesting an influence of valuation on the regulatory focus framing effect that has to be kept in mind when interpreting the given results.

For the fMRI experiment, a total of 29 participants (14 female) was used for data analysis (M = 41.45 years of age, SD = 10.83). During scanning, the participants performed an information-based judgement task. The task consisted of three blocks, whereby the first and last block only included 24 basic information trials. The intermediate regulatory focus block included 48 trials with 24 trials per regulatory focus frame. At the beginning of each trial, participants had a time interval of 10 seconds (s) to carefully read the predefined framed animal welfare information. Subsequently, a picture related to the prior given animal welfare information was presented for three seconds. At the end of each trial, participants rated each picture according to their subjective liking on a 7-point Likert scale (5 s). After completing the fMRI task, participants were asked to answer a questionnaire including control variables and demographics.

4.1.2.2 Data

Functional and anatomical brain images were obtained by means of fMRI. Using a general linear model (GLM), the effects of the regulatory focus framed animal welfare information on neural information processing were estimated. Information type, associated picture and evaluation events were modelled separately and convolved with a canonical hemodynamic response function. Whole-brain contrast maps between prevention and promotion framed information and first and third block information were generated for each participant. These individual contrast maps were used in a second-level analysis and analysed using a significance threshold of p < 0.05 with whole-brain false discovery rate (FDR; Logan & Rowe, 2004).

4.1.2.3 Results

The aim of our study was to identify the neural underpinning of the valuation of regulatory focus framed animal welfare information (Gifford & Bernard, 2004; Zhang & Buda, 1999). Confirming H₁, results provide evidence that promotion regulatory focus framed animal welfare information, which stresses accomplishments and advancements, significantly increased activation in the vmPFC (peak voxel: Montreal Neurological Institute (MNI) coordinates −3 44 −10, FDR corrected threshold p ≤ 0.05; Abbildung 4.1). This effect was also seen on the subjective liking of the pictures, where pictures after promotion focus animal welfare information was significantly greater (t(28) = 5.932, p < 0.001). Supporting H₂, basic information increased activity in the ACC compared to the same information given after and prior to the regulatory focus frames (peak voxel: MNI coordinates 0 35 20, FDR corrected threshold p ≤ 0.05). Surprisingly, this effect was not found on a behavioural level, demonstrating the added value of collecting and analysing neural data in information processing.

4.1.2.4 Conclusion

Confirming our hypotheses, the results indicate that the vmPFC is activated to a greater extent for promotion compared to prevention focus framed animal welfare information. Furthermore, it influences the neural processing of subsequent information, associated with activity in the ACC, an effect which is not indicated on behavioural level. There is mounting evidence that the vmPFC is associated with subjective valuation especially during decisions and the emotional comprehension of sentences (Bartra et al., 2013; Hervé, Razafimandimby, Jobard, & Tzourio-Mazoyer, 2013). Hence, goal-framed information formulated with regard to promotion regulatory focus, is able to induce increased activity in the vmPFC and thus, positively influence consumer’s (emotional) reaction on a neural level and extend its influence to other information. Taking into account the role of emotions in consumers’ decision-making, our work might help to understand why respective marketing communication strategies are able to affect consumer behaviour not only in the field of animal welfare.

4.2 Beitrag 2: Wahrnehmung der Nutztierhaltung – alles eine Frage der Kommunikation?

Der Beitrag entspricht der folgenden Publikation: Gier, N., Krampe, C., & Kenning, P. (2018). Wahrnehmung der Nutztierhaltung – alles eine Frage der Kommunikation? Journal of Consumer Protection and Food Safety, 13(2), 177–182. https://doi.org/10.1007/s00003-0171144-7

4.2.1 Einführung

Ziel der betrieblichen Kommunikationspolitik im Rahmen der marktorientierten Unternehmensführung ist es unter anderem, Kunden und Verbraucher über die Eigenschaften der angebotenen Produkte oder Services zu informieren (Verbeke & Ward, 2006). Im Kontext der in die Lebensmittelwirtschaft eingebundenen Nutztierhaltung haben dabei neben allgemeinen Angaben – zum Beispiel zur Art, der Herkunft, den ernährungsphysiologischen Werten oder den Verarbeitungseigenschaften – insbesondere Informationen über die mit der Tierhaltung verbundenen Praktiken eine hohe Bedeutung (Roininen, Arvola, & Lähteenmäki, 2006; Wille, Ermann, & Spiller, 2016). So spielen, vor dem Hintergrund der aktuellen gesellschaftlichen Diskussion um artgerechte Tierhaltung und der zunehmenden Kritik an der Massentierhaltung, Aspekte der Prozessqualität neben Kriterien der Produktqualität eine übergeordnete Rolle bei Kaufentscheidungen. Um den damit verbundenen Informationsbedarf der Verbraucher zu decken und am Markt höhere Prozessqualitäten zu signalisieren, werden von den Anbietern und Herstellern regelmäßig Kombinationen von Bild- und Textelementen verwendet, deren Wirkungsweise die Wahrnehmung der Verbraucher auf kognitiver und emotionaler Ebene beeinflussen kann (Levin, 1987). Es spielt also nicht nur die bewusste Wahrnehmung und Verarbeitung der Information eine Rolle, sondern ebenso die unbewusste Wirkung der Darstellungsweisen, sogenannten Frames (Levin, 1987; Levin et al., 1998).

Framing Effekte können auf unterschiedliche Weise hervorgerufen werden. Ein Beispiel wären veränderte Umgebungselemente, wie beispielsweise Bildelemente am PoS, die unterschiedliche Assoziationen der Verbraucher beim Kauf von tierischen Produkten aktivieren. Diese können einen positiven oder negativen Effekt auf das Kaufempfinden bzw. -verhalten ausüben (Chowdhury, Olsen, & Pracejus, 2008). Lange Zeit jedoch konnten diese, dem Verbraucher meist unbewussten Einflüsse, kaum empirisch erfasst werden. Demzufolge konnte der zugrundeliegende Mechanismus innerhalb der sogenannten ,,Black Box‘‘ oft nur theoretisch beschrieben werden (Kenning & Plassmann, 2005). Mithilfe von neuroökonomischen Analysen erscheint es nun aber möglich, die mit dem hier interessierenden Verbraucherverhalten verbundenen, oft unbewussten neuralen Prozesse direkt, das heißt in vivo zu beobachten und hinsichtlich ihrer Verhaltensrelevanz einzuordnen.

Die damit angesprochene Consumer Neuroscience (Kenning, 2014) hat in einigen Fällen dazu beigetragen, eine erhöhte Varianzaufklärung des Verbraucherverhaltens zu erlangen (Hubert & Kenning, 2008; Kenning, Oehler, Reisch, & Grugel, 2017; Kenning & Plassmann, 2005; Kosslyn, 1999; Plassmann et al., 2015). Darauf aufbauend könnten zukünftige Informationsstrategien und -maßnahmen entwickelt werden, die eine effektivere Kommunikation und Verbraucherinformation im Kontext der Nutztierhaltung ermöglichen. Der vorliegende Beitrag stellt die Ergebnisse einer SocialLab-Studie vor, bei der zwei Verfahren der Consumer Neuroscience eingesetzt wurden, um zu untersuchen, wie kommunikative Maßnahmen direkt am PoS zum Thema Tierwohl wahrgenommen und damit gegebenenfalls kaufentscheidend werden können.

4.2.2 Methoden und Ergebnisse

Um die Wirkungsweise verschiedener Kommunikationsmaßnahmen auf die Verbraucherwahrnehmung im Bereich der Nutztierhaltung zu untersuchen, wurden im Rahmen des SocialLab-Projektes 2 sogenannte ,,bildgebende‘‘ Methoden verwendet. Zum einen handelte es hierbei um die fMRT (Dimoka, 2010), zum anderen um die in vielerlei Hinsicht innovative fNIRS (Kopton & Kenning, 2014; Krampe, Strelow, & Kenning, 2016). Die Studien und die Ergebnisse sollen im Folgenden kurz skizziert werden.

4.2.2.1 fMRT Studie

Ziel der fMRT Studie war es zu untersuchen, ob es möglich ist, mit Hilfe der Theorie des regulatorischen Fokus (Crowe & Higgins, 1997) neurophysiologische und oftmals unbewusst ablaufende Wahrnehmungsprozesse im Kontext verschiedener Haltungsmethoden der Nutztierhaltung zu interferieren. Diese Frage ist deswegen bedeutsam, weil es bis dato unklar ist, ob kommunikative Maßnahmen überhaupt einen Einfluss auf die Einstellung der Verbraucher in diesem Bereich haben können oder ob die entsprechenden Reaktionen nicht vielmehr quasi reflexhaft ablaufen und im Gehirn mehr oder weniger ,,fest verdrahtet‘‘ sind. Um eine Antwort auf diese Frage zu finden, wurde nach Freigabe des Studiendesigns durch eine Ethikkommission eine fMRT-Analyse mit 29 Probandinnen und Probanden (Alter M = 41,45 Jahre, SD = 10,83; 14 weiblich) durchgeführt.

Im Vorfeld der MRT-Messungen wurden verschiedene Haltungsmethoden der Nutztierhaltung entsprechend der regulatorischen Fokus Theorie als Informationen differenziert formuliert. Dabei wurden zwei Foki unterschieden: Der Promotionsfokus ist dadurch gekennzeichnet, dass er Erfolge einer Maßnahme in den Vordergrund stellt. Im Gegensatz dazu ist es Merkmal des Präventionsfokus’, dass dieser Sicherheits- und Schutzaspekte thematisiert. Aufgabe der Probanden war es, sich die entsprechend formulierte Information aufmerksam durchzulesen und im Nachgang eine bildlich dargestellte Haltungsmethode zu betrachten und anschließend zu bewerten (Abbildung 4.2). In den Analysen wurde die Gehinaktivität während des Informationszeitraums und der Bildwahrnehmung zwischen den beiden Formulierungsarten analysiert.

Die ersten Ergebnisse der neuralen fMRT-Analyse zeigten, dass insbesondere der vmPFC immer dann eine stärkere Aktivität aufwies, wenn die Information im Promotionsfokus im Kontrast zum Präventionsfokus formuliert war (Abbildung 4.3). Demzufolge erzeugt eine Information, welche sich auf Erfolge, Errungenschaften und Verbesserungen fokussiert, eine stärkere neurale Reaktion in Hirnregionen, welche mit emotionalen Bewertungsprozessen assoziiert sind. Frühere Studien der Consumer Neuroscience haben zudem gezeigt, dass der vmPFC eine zentrale Rolle für das Bewertungssystem sowie das Kaufverhalten spielt (Bartra et al., 2013; Enax, Krapp, Piehl, & Weber, 2015; Plassmann & Weber, 2015). Dies bedeutet im Hinblick auf die Forschungsfrage zum einen, dass eine Beeinflussung neuraler Prozesse durch eine zieladäquate Darstellungsform der jeweiligen Information offenbar möglich ist. Zum anderen zeigt sich, dass Informationen beziehungsweise Darstellungsformen, die Verbesserungen beziehungsweise Erfolge einer Maßnahme in den Vordergrund stellen, eine erhöhte neurale, subjektiv-emotionale Wertung erfahren als sicherheits- und schutz-orientierte Informationen. Betrachtet man zudem die zentrale Rolle von Emotionen im Konsumverhalten insgesamt, so scheinen entsprechend modifizierte Informationen die Fähigkeit zu besitzen, das Konsumverhalten auf neuraler Ebene signifikant zu beeinflussen. Somit kann die Art und Weise der Kommunikation die Wahrscheinlichkeit erhöhen, dass Informationen zur Tierhaltung handlungsrelevante Implikationen für die Verbraucher haben.

Betrachtet man diese Ergebnisse bezugnehmend auf das übergeordnete Ziel der Wirkung unterschiedlicher Darstellungsvarianten der Tierhaltungsverfahren auf die (implizite) Wahrnehmung und die möglicherweise daraus resultierende gesellschaftliche Akzeptanzgewinnung, so lässt sich feststellen, dass die untersuchten Darstellungsweisen die neurale emotional-subjektive Wertung signifikant beeinflussen. In weiteren tiefergehenden Analysen wird nun nach Tierhaltungsmethoden differenziert, da Werte und a priori Einstellungen einen Einfluss auf den Effekt von Kommunikation auf die Akzeptanz haben können. So kann diese Studie dazu beitragen, differenzierte, akzeptanzfördernde Kommunikation für Verbraucher zu entwickeln.

4.2.2.2 fNIRS Studie

Aufbauend auf diesen Ergebnissen wurden ergänzende fNIRS Studien durchgeführt. Die mobile fNIRS bietet hierbei einen innovativen Ansatz die neuralen Prozesse, ähnlich dem Prinzip der fMRT, mittels Lichtimpulsen zu quantifizieren (Kopton & Kenning, 2014). Durch die mobile Einsetzbarkeit können neurale Prozesse und assoziiertes Konsumentenverhalten in einem naturalistischen Umfeld bemessen werden. Somit bestand das Ziel dieser Studien darin, zu prüfen, welche neuralen Prozesse am PoS ablaufen, wenn Verbraucher eine ‚echte‘ nutztierhaltungsrelevante Kaufentscheidung unter realitätsnahen Bedingungen treffen. Parallel hierzu sollte zudem die Validität der mobil einsetzbaren fNIRS untersucht werden. Hierbei zeigte sich zunächst, dass die mobile fNIRS eine valide neurowissenschaftliche Methodik im Forschungsfeld der Consumer Neuroscience darstellt (Krampe et al., 2016). Aufbauend auf diesen ersten Forschungsergebnissen wurde eine zweite Studie konzipiert, welche auf die Datenerhebung am PoS fokussierte und somit die mobile Einsetzbarkeit der fNIRS in einer innovativen Feldstudie prüfen sollte. Hierzu wurden außerhalb der Öffnungszeiten in einem Lebensmittelmarkt über der Selbstbedienungstheke für Fleischwaren entweder biologisch- oder konventionell-orientierte Tierhaltungskommunikationsmaßnahmen (TKM), die als situative Frames fungierten, platziert. Um die neurale Reaktion der Probandinnen (n = 18; Alter durchschnittlich 41 Jahre, SD = 7,96) auf die dementsprechend veränderten Marktaufbauten zu erfassen, wurden diese mit einen mobilen fNIRS- und Eye-Tracking-Gerät ausgestattet. Danach wurden sie gebeten, einem vorgegebenen Einkaufsweg zu folgen und einen Einkauf zu tätigen.

Im Rahmen der Datenanalyse wurde die Gehirnaktivität der beiden TKM (biologisch- und konventionell-orientierte TKM) kontrastiert (Abbildung 4.4). Die Ergebnisse der fNIRS Datenanalyse zeigten bei einer statistisch-liberalen Auswertung (p < 0,1), dass insbesondere Regionen des OFC sensitiv für Veränderungen der TKM sind. So ist die neurale OFC-Hirnaktivität im Kontrast zu konventioneller TKM bei biologisch-orientierter TKM erhöht. Ergänzend zu den fNIRS Daten wurden über einen Zeitraum von 6 Wochen die mit der TKM assoziierten Abverkaufszahlen pseudo-randomisiert ermittelt. Hierbei zeigte sich für die biologisch-orientierte TKM ein signifikant höherer durchschnittlicher Fleischwaren-Wochenumsatz pro Kunde (Wochenumsatz Fleischware/Anzahl Kunden; t(5) = 2,65, p < 0,05). Betrachtet man beide Befunde im Zusammenhang, so lässt sich schlussfolgern, dass biologisch-orientierte TKM zu einer höheren neuralen Reaktion im OFC führt. Dies wiederum weist auf eine erhöht-aktivierte, subjektive Bewertung der Probanden hin, welche offenbar verhaltensrelevante Auswirkungen auf das Einkaufsverhalten der Kunden hat.

Des Weiteren wurden die TKM mit Hilfe der fMRT im Labor untersucht, um Befunde zur umfassenderen Lokalisierung der neuralen Aktivität zu ergänzen. Die Resultate zeigen jedoch überraschenderweise keine signifikanten Aktivitätsunterschiede zwischen den beiden verwendeten TKM (biologisch- und konventionell-orientiert). Dies lässt vermuten, dass die situative Präsentation der TKM im Markt, also am PoS, den entscheidenden Einfluss auf die neurale Reaktion der Kunden und den damit verbundenen oft unbewussten Kaufentscheidungsprozess hat. Somit scheint das Entscheidungsumfeld (also der situative Frame), in welche die TKM platziert wird, eine zentrale Rolle einzunehmen. Dieser Aspekt wird in der verbraucherpolitischen Diskussion über in anderer Hinsicht optimierte Informations- und Kommunikationsinstrumente oft übersehen und unterstreicht noch einmal die zentrale Rolle des Handels, der das Entscheidungsumfeld maßgeblich gestaltet (vgl. Krampe, Gier, Römhild, Kenning (2018) „Standards, Hindernisse und Wünsche in der Nutztierhaltung – Die Perspektive des Handels“ im “Journal of Consumer Protection Food Safety”). In Bezug auf das Ziel der Identifizierung der mit dem relevanten Verhalten neuropsychologischen Prozesse zeigt die fNIRS-Technologie somit eine höhere externe ökologische Validität und gibt wichtige Hinweise, die in einer weniger biotischen Studienanlage nicht gewonnen werden konnten. Konkret zeigt sich, dass kaufentscheidende Phänomene oft erst am PoS entstehen – ein Resultat, dass die verhaltensökonomische Konsumforschung als ‚konstruierte Präferenzen‘ (constructive preferences) bezeichnet und mit der so genannten Query Theory (also „Abfrage-Theorie“) begründet (Johnson, Steffel, & Goldstein, 2005). Dies unterstreicht in methodischer Hinsicht noch einmal die besondere Bedeutung der mobilen fNIRS und verdeutlicht theoretisch die Rolle sogenannter exogener Präferenzen in den entsprechenden Entscheidungsprozessen.

4.2.3 Ausblick und Implikationen

Betrachtet man die dargestellten Forschungsergebnisse bezugnehmend auf das übergeordnete Ziel, die neuropsychologische Wirkung unterschiedlicher Darstellungsvarianten der Tierhaltungsverfahren auf die Wahrnehmung und die möglicherweise daraus folgende gesellschaftliche Akzeptanzgewinnung besser zu verstehen, so lässt sich feststellen, dass die untersuchten Darstellungsweisen die impliziten neurophysiologischen Prozesse signifikant beeinflussen. Insbesondere scheint die oftmals implizite Wirkungsweise von Darstellungsvarianten der Tierhaltungsverfahren bedeutsam zu sein. Des Weiteren spielt der Präsentationsrahmen – das Framing – eine entscheidende Rolle in der (impliziten) Kommunikationswahrnehmung und -verarbeitung der Verbraucher. Es ist daher wichtig, diese Aspekte – Darstellung und Kontext – bei der Gestaltung von Kommunikationen im Bereich der Nutztierhaltung künftig noch stärker zu beachten. Und auch wenn die Ergebnisse der vorgestellten Studien vorläufig sind und erste Hinweise geben, in welche Richtung weiter geforscht werden sollte, so wird doch erkennbar, dass die Frage nach dem ‚Wie?‘ bei der Gestaltung von Verbraucherinformationen auch in diesem Bedarfsfeld künftig von Bedeutung sein wird.

Die Ergebnisse bestärken zudem die zunehmend in der sozialwissenschaftlichen Konsumforschung gewonnene Erkenntnis, dass individuelle Faktoren wie Wissen, Einstellungen und Handlungsintentionen zwar durchaus bedeutend sind für die Kaufentscheidung, dass jedoch der unmittelbare Entscheidungskontext der letztlich ausschlaggebend ist.

4.3 Beitrag 3: Online reviews as marketing placebo? First insights from NeuroIS utilising fNIRS

Dieser Beitrag wurde bei der „European Conference on Information Systems“ 2020 im Rahmen eines Vortrags vorgestellt und ist als „Research in Progress“ veröffentlicht: Gier, N. R., Kurz, J. & Kenning, P. (2020). Online reviews as marketing placebo? First insights from Neuro-IS utilising fNIRS. ECIS 2020 Research-in-Progress Papers. 33. https://aisel.aisnet.org/ecis2020_rip/33

4.3.1 Abstract

In the digital era, costumers often preuse online reviews prior to purchasing a product or a service. These online reviews may not only influence purchase intentions, but also unintentionally shape the expectations that form later customer experiences with the product – a phenomenon defined as ‘marketing placebo effect’. Against this background this research aims to investigate a potential placebo effect of online reviews on a behavioural and neural level. While recording neural activity via functional near-infrared spectroscopy (fNIRS), participants consumed a product after reading a well- or poorly-fitting online review of the product. The results indicate that well-fitting online reviews increased the liking of the product compared to poorly-fitting online reviews, and by doing so, constitute a placebo effect. In parallel, they induce increased neural activity in regions that are associated in neuroscientific literature with processing social information (i.e. medial PFC). In contrast, poorly-fitting online reviews increased activity in regions related to conflict resolution (i.e. lateral PFC). These results demonstrate that online reviews make for an effective placebo, influencing customer experiences on both, the behavioural and the neural level.

4.3.2 Introduction

Customers often use the product’s appearance, price, and taste attributes to determine product preferences which subsequently change the perceptions and experiences of the product (Plassmann et al., 2008). This expectancy-based marketing effect is named the marketing placebo effect (MPE). Originating from the medical field, the psychological phenomenon of placebo effects has spilled-over to other disciplines, such as consumer research and marketing. In marketing, different product characteristics have the capacity to act as placebos by shaping and altering product perceptions and expectations that result in changes in customers behaviour (Shiv, Carmon, & Ariely, 2005). Many marketing activities have been corroborated as placebos, showing perturbing behavioural and neural effects of price (Plassmann et al., 2008), brand information (McClure et al., 2004), product description (Denisova & Cairns, 2015) and countless others.

Nevertheless, in the digital era, customers experience and interaction with products and other customers drastically changes, generating a multitude of new factors that could potentially operate as the MPE. For instance, until now MPE research often neglects the fit of the placebo to the target product, since the placebo information is normally directly associated with the product and its verity never questioned. However, this fit would potentially influence the believability and trustworthiness of the placebo information provided, affecting the expectations used to judge the product’s quality and experience (Banerjee, Bhattacharyya, & Bose, 2017). An example where the fit between placebo and products and its believability are especially prominent is in the case of online reviews. Online reviews are indirectly linked to the product as any individual or bot can write subjective opinions that are, unlike traditional types of reviews, anonymous and potentially flawed (Hu, Bose, Koh, & Liu, 2012; Hu, Liu, & Sambamurthy, 2011; Hu, Pavlou, & Zhang, 2006). In contrast to online reviews, it is possible to discern the trustworthiness and accuracy of product opinions when deriving it with an identifiable individual or from a brand with high reputation. However, reading a review on a website is anonymous and does not allow for this identifiability, and thus lending to the degradation of trustworthiness of the review since it is uncertain whether the review is credible and fits the product and the consumption process.

Whether poorly-fitting online reviews can have a significant influence on the customers’ product experience, known that they are potentially flawed, constitutes an unanswered question. However, its possible disadvantages on online retailing might be extensive as 88 % of customers are shown to consult reviews prior their purchase in order to evaluate a product (Duan, Gu, & Whinston, 2008; Hitt & Li, 2004; Lee, Park, & Han, 2008; Silver, Tan, & Mitchell, 2012). Consequently, the consideration of fit between online reviews and product, is still limited in placebo- and IS-research. To fill this gap, this study expands MPE literature with the examination of a novel marketing placebo of online reviews on a behavioural and neural level. For the later, the neuroimaging technique of functional near infrared-spectroscopy (fNIRS) was utilised. Since traditional measurements of customers behaviour such as questionnaires and self-reports are only capable to assess constructs that the customers are consciously aware of, often unconscious processing differences of product experiences are neglected (Brocke, Riedl, & Léger, 2013). The addition of neuroscience to traditional behavioural approaches enhances the ability to go beyond the deductions made by traditional variables and paradigms, and thus advances explanations via examination of underlying processes (Dimoka, 2010; Dimoka et al., 2012; Riedl, Hubert, & Kenning, 2010). Ascribable to its mobility, fNIRS methodology was used in this research with the aim of providing greater ecological validity to the MPE as customers often read reviews on their technological devices (i.e. smart phones) rather than on a small screen in a loud and uncomfortable, immobile MRI scanner. Consequently, this research aims to better understand the neural mechanism of customers experiences induced by interactions in the digital world through a neural measurement that allows for ecological validity.

4.3.2.1 Theorisation of the MPE

The majority of MPE theories maintain that this phenomenon stems from a priori expectations held about the product that alter consumption experiences (Enax & Weber, 2015; Plassmann & Weber, 2015; Shiv, Carmon, & Ariely, 2005). Most models subsume expectations as a moderator between salient beliefs and outcomes (Enax & Weber, 2015; Shiv, Carmon, & Ariely, 2005). However, one hindrance to these theories is that they do not incorporate the fit between placebo and product. In consequence, researchers are divided whether manipulation of expectancies is not required to elicit a placebo effect (Shiv, Carmon, & Ariely, 2005) or are observed only when one has high expectations about the placebo information (Alves, Lopes, & Hernandez, 2017) or even assume that people do not have to expect a product to work but rather require motivation for a placebo effect to occur (Irmak, Block, & Fitzsimons, 2005).

The theory of reference-dependent preferences might solve this conflict (Gneezy, Gneezy, & Lauga, 2014; Koszegi & Rabin, 2006). The theory propounds that the appearance of a price-quality link is dependent on whether consumption had fulfilled previous expectations. In essence, reaching or exceeding the reference point results in a typical placebo effect (meeting expectations), however, when consumption experience is below the reference point (failing to meet expectations), a loss sensation is felt, resulting in no effect. Transferring this reasoning to online reviews, the following hypothesis was formulated based on this theory.

H₁. Customers online reviews can act as a placebo for liking ratings, generating a greater liking for the consumed product for well-fitting reviews compared to poorly-fitting reviews.

4.3.2.2 Neuroscientific basis of the MPE

Research has displayed the significance of investigating the MPE with neuroimaging techniques, often displaying changes in prefrontal cortex (PFC) neural activity when manipulating the product information (i.e. price, brand, descriptions) while keeping physical properties constant (Denisova & Cairns, 2015; McClure et al., 2004; Plassmann et al., 2008). These studies suggest that modulation of product information can impact the representation of the product in two specific brain areas, namely the dorsolateral PFC (dlPFC) and ventromedial PFC (vmPFC). It has been asserted in MPE research that medial PFC regions, such as the vmPFC and dorsomedial PFC (dmPFC), have a particular role in encompassing expectation-related valuations (Enax & Weber, 2015). The vmPFC integrates sensory information and cognitive encoded expectations in order to direct consumption experiences (Plassmann et al., 2008). Particularly in the realm of online reviews, the dmPFC could has a role in incorporating social information in value computations (De Martino et al., 2017).

In contrast to the medial PFC regions, the neural components implicated in cognitive control and active during conflict are the more lateral regions of the PFC, such as the dlPFC and ventrolateral PFC (vlPFC; Berns, 2005; MacLeod & MacDonald, 2000). The dlPFC implements the cognitive control required to perform a task successfully through maintaining, biasing and controlling information processing in working memory, serving as a top-down regulative control device (Berns, 2005; MacLeod & MacDonald, 2000). The vlPFC, on the other hand, shows increased activity during the acceptance of evidence that conflicts with one’s own beliefs by incorporating evidence that contradicts a personal belief and inhibits this personal belief in order to develop correct and valid judgements (Hooker & Knight, 2006).

Based on the mentioned considerations regarding the medial and lateral parts of the PFC, it is expected that there will be a neural processing difference between reading and consumption phases for poorly- and well-fitting reviews, since integration of conflicting information and greater cognitive control is required for poorly-fitting reviews. Furthermore, the consumption period is assumed to be neural processed differently, since the review-induced expectancies will either confirmed or disproved. As during the reading of reviews no expectancy conflict will be realised at that stage, no hypothesis is formulated for this contrast. Consequently, the following hypotheses on neural MPE can be made:

H₂. Reading of reviews will show in contrast to the consumption of products: a) increased mPFC activity for well-fitting information, and b) increased mPFC and decreased lPFC activity for poorly-fitting information.
H₃. A lPFC deactivation will be present during the consumption of a product succeeding a well-fitting review compared to consumption following a poorly-fitting review.

4.3.3 Research methodology

4.3.3.1 Participants

In order to test the hypotheses, a sample, consisting of nineteen right-handed healthy participants, was recruited. However, two data sets had to be excluded due to incorrect data collection, leaving seventeen participants (10 females and 7 males) ages 19 to 57 (M = 31.12, SD = 13.19), whereby care was taken to avoid a convenience sample of students. All participants were not taking medication nor pregnant and were screened for any history of major psychological or neurological disorders and dietary constraints. The study was approved by the local institutional ethics committee.

4.3.3.2 Placebo

Investigating online reviews as a potential placebo for the MPE, twenty-one easily imbibed tea products were used as stimuli. The ambiguous nature of the products from not disclosing any product information (i.e. brand, flavour, packaging) ensured no issue in tasting the same product twice. Thereby, the only information presented to participants were the online reviews. All reviews were formulated with the same three attribute word structure with attributes that either fit the product well or poorly. For example, the well-fitting review for a ginger-mint tea stated ‘this natural tea is intensive and wholesome’, whereas the poorly-fitting review stated ‘this dry tea is chocolatey and salty’.

Attributes used in the reviews were derived from pre-test results. During the pre-test, participants (N = 99; M_age = 36.37, SD_age = 11.62) were shown an image of a tea package and were asked to rate how well 33 attributes fit the product on a six-point Likert scale ranging from ‘fit very poorly’ to ‘fit very well’. The top- and lowest-rated attributes for each product were subjected to paired-samples t-tests. Of those significant (p < 0.05, Bonferroni), the twenty-one products and its attributes that had the greatest difference between the best and poorest fitting attributes were chosen for the main experiment. Stimuli were presented in a form representative of an online supermarket website, specifically allyouneedfresh.de.

4.3.3.3 Experimental paradigm

The experimental paradigm was designed to examine online reviews as a potential placebo for the MPE on a neural level with the fNIRS neuroimaging technique. First, participants were welcomed, and then provided with information regarding the experiment. They were informed both in verbal and written form about the fNIRS device and the experimental procedure, explaining that they will see evaluation of tea products which they subsequently had to consume while wearing the fNIRS device on their forehead. Once participants understood the study, informed consent was obtained in accordance with the Declaration of Helsinki.

Participants were then sat down in front of the computer used to present the stimuli. Once comfortable, the fNIRS device was prepared, placed on the participant’s head, and data quality checked. Participants then underwent the experimental task. Applying an event-related design with two levels of online review information fit, the twenty-one different small tea samples (2 cl each) were placed on a tablet in the order that participants were to drink the tea in each run. During each trial (Abbildung 4.5), participants were first shown the review of one product (5 seconds (s), reading phase). Then after a randomly fixed interstimulus interval between 1–5 s, they were asked to consume the tea while the review was reshown on the top of the screen (10 s, consumption phase). Participants were instructed to reach for the cup with the number that corresponds to the consumption instructions on the screen and consume it with limited movement to minimise interferences with the neural data collection. Subsequently, participants were asked to rate the fit of information and the liking of the product with a six-point Likert scale without any time constraints. A randomly fixed intertrial interval of 4–6 s was given at the end of each trial. The experimental paradigm containing two runs with twenty-one trials per run (42 trials in total), lasting 25–30 minutes. After the experimental task was completed, the device was taken off and a questionnaire was given. Once completed participants were freecccisclosure and financial compensation of 10.00 €.

4.3.3.4 fNIRS data acquisition

Functional near infrared-spectroscopy (fNIRS) is an optical spectroscopy methodology that measures for cerebral hemodynamic responses through near-infrared light sources (Ferrari & Quaresima, 2012). More precisely, fNIRS measures the hemodynamic response of both oxygenated (oxy-) and deoxygenated (deoxy-) hemoglobin (Hb) through assessing absorption rates of certain light wavelengths (Kopton & Kenning, 2014). Fundamentally, near-infrared light has the ability to pass through and illuminate biological tissue non-invasively. Thereby, light is sent into the brain through a diode placed on the individual’s head and is absorbed by oxy-Hb and deoxy-Hb. The remaining light is then picked up by a detector placed close-by, generating a light pathway in the form of a banana (Kopton & Kenning, 2014). Absorption changes due to changes in oxy-Hb and deoxy-Hb concentrations are translatable into measurable hemodynamic responses based on mathematical formulations (Jöbsis, 1977; Kopton & Kenning, 2014).

fNIRS has been successful in measuring functional neural activations in a multitude of cognitive and physical tasks, lately also in Neuro-IS research (Balardin et al., 2017; Ferrari, Mottola, & Quaresima, 2004; Miyai et al., 2001; Yoshino, Oka, Yamamoto, Takahashi, & Kato, 2013). In particular, the regions relevant to the current experiment validated in prior fNIRS research include the medial PFC (dmPFC, vmPFC) and lateral PFC (dlPFC and vlPFC) (Çakir, Çakar, Girisken, & Yurdakul, 2018; Krampe, Gier, & Kenning, 2017; Krampe, Strelow, Haas, & Kenning, 2018b; Misawa, Shimokawa, & Hirobayashi, 2014, Abbildung 4.6).

In this study, a fNIRSport-System (NIRx Medical Technologies, Berlin, Germany) was used to record optical signals on two-wavelengths (760 and 850 mm) at a sampling rate of 7.81 Hz. 22 channels comprising of eight light emitters (or sources) and seven detectors, which were 3 mm separated, were used for the recording of cortical neural activity. The optode montage setup used in this fNIRS experiment was attained from NIRx Medical Technologies (Abbildung 4.6). Channels were classified in this research into medial (channel: 3, 4, 5, 8, 10, 11, 12, 13, 15, 18, 19, 20) and lateral (channel: 1, 2, 6, 7, 9, 14, 16, 17, 21, 22) PFC. The commonly used continuous-wave instrument was utilised in the current experiment (Boas, Elwell, Ferrari, & Taga, 2014). This fNIRS device requires participants to wear a fitted headband that covers most of the PFC. To increase comparability between the anatomical brain structures of the participants and consistency among neural data, the craniometric point of the nasion was used as an orientation point (Krampe et al., 2018b). Additionally, signal quality was calibrated for each participant for the inspection of any impediments to the signal measurement (i.e. external light; Gefen, Ayaz, & Onaral, 2014). If any poor signal quality was found, it was corrected for by moving hair away from directly under the optode with a cotton-bud. NIRS-Star software package (version 14.2) was used for checking signal quality and data collection.

4.3.3.5 Statistical analysis

4.3.3.5.1 Behavioural analysis

To test H₁, a repeated measures Analysis of Variance (ANOVA) was conducted on the liking ratings. The first repeated factor comprised of the information type: well-fitting or poorly-fitting. The second repeated factor were the twenty-one different stimuli products.

4.3.3.5.2 fNIRS data analysis

Prior the neural data analysis, fNIRS data was pre-processed using NIRx Software Package (NIRx Medical Technologies, Berlin, Germany). Channels with bad signals, showing discontinuous shifts in measurement values, were removed. As common, fNIRS data time series were smoothed and artefacts (i.e. heart-rate or drifts in the optical signal) were controlled for by a band-pass filter with a low cut-off frequency of 0.01 Hz and high cut-off frequency of 0.2 Hz, analogous to previous research to this field (Krampe et al., 2018b). Furthermore, the modified Beer-Lambert law was used to convert raw optical density signals to hemoglobin concentration changes and were used to set the parameters computed from the hemodynamic states function.

In order to evaluate H₂ and H₃, a general linear model (GLM) was set up for every participant. Four regressors were used in the GLM: [1] reading of well-fitting reviews, [2] reading of poorly-fitting reviews, [3] consumption succeeding well-fitting reviews, and [4] consumption succeeding poorly-fitting reviews. The entire duration of the reading phase was integrated in the analysis however the durations in the consumption phase started halfway through the consumption phase. This was done to assure that head movements from consumption in the first 5 s were not an impediment to measuring neural responses. However, all analyses were repeated using the whole duration of the consumption phase, resulting in comparable results.

The GLM were first calculated on a single subject individual level (within-subjects level) and after a second-level group contrasts analysis was carried out to compare neural activation differences across subjects (between-subjects level). For the GLM, a t-contrast activation map was plotted on a standardized brain and a family-wise error corrected p-value threshold was set at p < 0.05. The contrasts comparing reading and consumption phases were performed separately for the review conditions in order to evaluate H_2(a) and H_2(b) respectively. H₃ was evaluated by contrasting the consumption phases between well-fitting and poor-fitting reviews.

4.3.4 Results

4.3.4.1 Online reviews as an effective placebo

The manipulation check confirmed, using a two-way repeated ANOVA on the information fit ratings, that fit of information varied as intended across the two information conditions (F(1,16) = 221.932, p < 0.001, partial eta-squared (η_p²) = 0.933).

A two-way repeated measures ANOVA was carried out on liking ratings to assess the main behavioural research question on whether online reviews are an effective placebo. In accordance with H1, there was a significant difference between the two review conditions (F(1,16) = 17.60, p = 0.01, η_p² = 0.524) and across the twenty-one products (F(20,320) = 3.509, p < 0.001, η_p² = 0.180). There was no significant interaction between the review conditions and products (F(20,320) = 0.763, p = 0.758, η_p² = 0.045). This non-significant interaction suggests the placebo effect was not influenced by the different products as the difference between the two review conditions for the liking responses did not differ across products. These results allow for the acceptance of H₁.

4.3.4.2 Influence of online reviews fit on a neural level

Neural results demonstrated significant activation differences between the reading and consumption phases for the different review conditions. As suggested in H_2a, there was increased activity in the medial PFC (channel 3: t(16) = 2.303, p = 0.018, d = 0.559; channel 4: t(16) = 3.119, p = 0.003, d = 0.756; channel 8: t(16) = 2.268, p = 0.019, d = 0.550; t-statistics for all channels upon request) during reading phases in contrast to consumption phases while presented with well-fitting reviews (Abbildung 4.7a). In accordance with H_2b, results established an increased medial PFC (channel 4: t(16) = 2.782, p = 0.006, d = 0.675) activity and also lateral PFC (channel 21: t(16) = −2.229, p = 0.0202, d = −0.541; channel 22: t(16) = −2.028, p = 0.029, d = −0.492; t-statistics for all channels upon request) deactivation during reading phases versus consumption phases while presented with poorly-fitting reviews (Abbildung 4.7b). H₃ was however rejected as there were no significant activation differences in the consumption of the product succeeding well-fitting reviews compared to succeeding poorly-fitting reviews (t-statistics for all channels upon request).

4.3.5 Discussion

The current research aimed to investigate a placebo effect caused by digital interactions, i.e. reading online reviews, and evaluate its effect on customer experiences on a behavioural and neural level. The results of this study provide supporting evidence towards the hypotheses. They show that online reviews are able to elicit an effective marketing placebo effect. This implies that anonymous comments left on websites by customers or bots can alter product preferences and physiological customer experiences. The same product (i.e. tea) was liked better when the online review fit to the consumption experience compared to when it did not fit the consumed tea.

The influence of online reviews as a placebo for the MPE was also seen on a neural level, allowing to better understand the underlying neural mechanisms. In presenting information that had both fit well and fit poorly the consumed product, there was increased medial PFC activity during the reading of reviews, indicating that participants were indeed using the review information as social information. Furthermore, the number of significant channels in the medial PFC in the well-fitting review condition might hinting towards the direction that more medial PFC activity is present when reviews fit well compared to when reviews fit poorly. This would be in line with past research suggesting that the medial PFC regions are related to updating beliefs, activating based on the social information presented (De Martino et al., 2017). The lateral PFC deactivation in reading reviews compared to consumption while presenting poorly-fitting review information indicates that a conflict was felt in tasting a product that did not match its information, and that cognitive effort was required to resolve this conflict and to correct faulty beliefs generated from previous knowledge (Hooker & Knight, 2006). These results are in accordance with an eminent notion in research suggesting increases in cognitive demands activates lateral PFC regions and as a result reduces neural activity in regions responsible for emotional processing, particularly the medial PFC and emotional limbic structures (Dimoka, 2010). The insignificance of hypothesis three was surprising as it is quite substantiated in research that the lateral PFC controls for conflict and incongruence (Berns, 2005), and thus should have been activated during the consumption of a poorly-fitting review. One possible explanation is that since the review was untrustworthy when poorly-fitting, the participants solely focused on consumption rather than the both in conjunction during the consumption experience. It might be that other specifics, which characterise anonymous online reviews, influence the processing of online reviews. As prior literature indicates, trust and distrust seem to be two disjunct constructs associated with medial areas (Dimoka, 2010). In future studies it might be valuable to focus on the effect of trust and distrust on the occurrence of MPE.

4.3.6 Limitations

As this research is work in progress, some refinements to the methodological design could potentially increase its validity and significance. Firstly, additional laboratory studies should be conducted to facilitate the understanding of the underlying mechanism behind the effect being observed. Although a multitude of studies demonstrate the MPE on a behavioural level (Denisova & Cairns, 2015; McClure et al., 2004; Plassmann et al., 2008), it often investigates the improvement rather than a worsening of the product. Moreover, even though it is suggested that sufficient statistical power in neuroscientific and fNIRS studies can already be obtained with 20 participants (Riedl et al., 2010), the sample size of 17 in this research is rather small. Furthermore, it should be noticed that other processes are obviously included in the analysed contrasts of reading a review and consuming a product, e.g. in contrast to reading, consuming is apparently linked to some sort of movement. These could potentially interfere the neural processes and need to be considered while interpreting the results. However, still a differentiated processing of well- and poorly-fitting reviews is observable, indicating differences in underlying mechanisms. Further statistical analysis, such as temporal derivative distribution repair, a newer movement artefact reduction technique, may show beneficial for the interpretations of the results (Fishburn, Ludlum, Vaidya, & Medvedev, 2019).

4.3.7 Implications

However, the preliminary results indicate that online reviews act as placebos and affects neural mechanisms. The application of the fNIRS enhanced the conventional behavioural traditions through the explication of the underlying mechanisms that drive the MPE in digital settings in application of an economic theoretical explanation. Digitalization does not only disrupt economic processes, but also challenges traditional theories e.g. by creating new factors of MPE. Online reviews are, unlike traditional types of reviews, anonymous and due to the unidentifiability of the source potentially flawed. Reference-dependence theory is only one potential theorisation of the MPE. Other related theories from other disciplines might be useful in future research as potential conceptualizations of new types of the MPE as for example Schultz’s et.al. (1997) neurological theory of reward-prediction error and Oliver’s (1977) marketing theory of expectancy-disconfirmation. In addition, the study findings suggest that social information can modulate the customers’ perceptions held about a product and related customer experiences. In a broader sense, it indicates that digitalisation does not only change customers experience on a subjective level, but also has consequential effects on the underlying, physiological information processing. This would hold managerial implications as firms should be aware of the reviews written about their product as it could influence not only the purchase decision but as well the consumption experience and moreover the consumer’s repeat purchase behaviour. The management of online reviews might be a considerably crucial point in the innovation management process. Since it is still relatively unknown which factors determine the success of an innovation (Tong, Acikalin, Genevsky, Shiv, & Knutson, 2020), reviews as potential source of variation should be effectively administered and the integration of neural evidence seem to be especially valuable in this matter (Ariely & Berns, 2010). In order to verify the reviews and ensure the fit of the review to the product, online platform could potentially increase the social identifiability of the review source, since brain areas responsible for updating beliefs based on the social information seem to be involved in this effect. In conclusion, customers reviews communicated over online platforms are indeed an influential piece of information used to direct customers decision making and, finally, customer experiences.

4.4 Beitrag 4: Verbraucherinformationssystem

4.4.1 Beitrag 4.1: Zur Konzeption eines Verbraucherinformationssystems als Ergänzung – oder Alternative? – zum klassischen Informationslabel

4.4.1.1 Hintergrund und Zielsetzung

Eine wesentliche Problematik moderner Volkswirtschaften bilden die oft durch arbeitsteilige Prozesse und entsprechend ausdifferenzierte Wertschöpfungsketten induzierten Informationsasymmetrien zwischen den Anbietern und Nachfragern einer Leistung. Beim Thema ‚Tierwohl‘ kommt hinzu, dass dies eine Vertrauenseigenschaft (Credence Attribute) des Produktes darstellt, deren Ausprägung vom Verbraucher, beispielsweise beim Produkt ‚Fleisch‘, am PoS kaum festgestellt werden kann. Die Informationsökonomik hat auf dieses Marktversagen reagiert und verschiedene Ansätze entwickelt, solche Informationsasymmetrien zu reduzieren. Die klassische Antwort der Verbraucherpolitik ist das Signaling mit Hilfe von Labels oder Siegeln. Im Kern sollen diese dem Verbraucher auf den ersten Blick, leicht verständlich und verlässlich eine – mehr oder weniger – bestimmte Qualität signalisieren (Eberle et al., 2011; Olaizola & Corcoran, 2003; Reisch, 2003).

Mit diesem Ansatz verbinden sich mehrere Vorteile: So bieten Label zum Beispiel in der Lebensmittelwirtschaft die Möglichkeit, den Verbraucher direkt am PoS über produktspezifische Vertrauenseigenschaften, wie bspw. Tierwohl- oder Bio-Aspekte, zu informieren (Eberle et al., 2011; Janssen & Hamm, 2012; Olaizola & Corcoran, 2003; Reisch, 2003). Dabei lassen sich Label-Qualitäten in unterschiedlicher Breite und Tiefe definieren. Während beispielsweise ‚gentechnikfrei‘ ein Prozessattribut betrachtet, behandeln Bioqualitäts- oder QS-Siegel umfassendere Merkmale. Diese setzen dabei durchaus auf Halo-Effekte, also auf Qualitätsvermutungen, die über die eigentliche Qualität des Produktes hinausgehen. Recht schnell stoßen Label jedoch an ihre Grenzen und erzeugen oftmals unerwünschte Nebeneffekte (Eberle et al., 2011; Franz et al., 2010). So können Verbraucher beim alltäglichen Einkauf durch die Vielzahl an Labeln, verbunden mit einem geringen Involvement, verwirrt und überfordert werden (Roosen, Lusk, & Fox, 2003; Verbeke, 2005, 2008). Gerade im Bereich der Nutztierhaltung gibt es eine Flut an Labeln, so dass bisweilen je nach Qualitätsprüfung dasselbe Produkt mit mehreren, unterschiedlichen Labeln gekennzeichnet werden kann. So fordert unter anderem auch der Bundesverband der Verbraucherzentrale mehr Transparenz und eindeutige Kennzeichnungen im Bereich des Tierwohls in der Nutztierhaltung (VZBV, 2017). Diese Vielfalt und gelegentliche Inkonsistenz hat Auswirkungen auf die Glaubwürdigkeit der einzelnen Label und zeigt, dass das Labelsystem in seiner jetzigen Form wohl einen ‚abnehmenden Grenznutzen‘ hat (Verbraucherkommission Baden-Württemberg, 2011). Die Verfügbarkeit label-induzierter Information stellt heute somit kein Maximierungs-, sondern ein lokales Optimierungsproblem dar, das in der Praxis auf erhebliche Probleme stößt (Kenning et al., 2017). Vor diesem Hintergrund ist es Ziel des im Folgenden zu skizzierenden Forschungsprojektes, Hinweise für die zukünftige effektive und, nach Möglichkeit, effiziente Gestaltung der Kommunikation von Verbraucherinformationen zu geben. Darauf aufbauend sollen Politik- und Kommunikationsempfehlungen für die gesetzliche und privatwirtschaftliche Umsetzung von Kennzeichnungsmaßnahmen abgeleitet werden. Um dieses Ziel zu erreichen, werden verschiedene Ansätze der Verbraucherinformation diskutiert und Erkenntnisse aus bisherigen Forschungsarbeiten genutzt, um bestehende Instrumente zu optimieren bzw. zu ergänzen oder Alternativen zum bisherigen System der Informationsökonomik durch klassische Label zu entwickeln.

4.4.1.2 Methodik

Um das Themenfeld der Verbraucherinformation zunächst phänomenologisch zu durchdringen, wurden im ersten Schritt im Kontext der Nutztierhaltung der Informationsstand, die Informationsbeschaffung sowie der Informationseinfluss von und zu Verbrauchern mittels einer Literaturrecherche und einer flankierenden Fokusgruppe untersucht. Die aus der qualitativen Sozialforschung stammende Methode der Fokusgruppe (Krueger & Casey, 2014) ermöglicht es, den Informationsprozess aus Sicht der Verbraucher zu begreifen und mögliche Anschlusskriterien für die Informationsbereitstellung und -beschaffung zu identifizieren. Im konkreten Projektfall wurden mit Hilfe einer moderierten Diskussion 9 Verbraucher/-innen eingeladen, sich über die Thematik der Informationskommunikation im Bereich der Nutztierhaltung auszutauschen und diese zu diskutieren. Anhand der daraus gewonnenen Erkenntnisse konnten unterschiedliche Bedürfnisse und Motive der Verbraucher an Informationsinhalten und zur Informationsbeschaffung zum Thema Nutztierhaltung unterschieden werden. Daraus abgeleitet wurden in einer tiefergehenden Literaturrecherche Alternativen zu den heutigen Angeboten der Verbraucherinformation gesucht, welche bereits durch Studien erste Hinweise auf ihre Effektivität geben oder in ähnlicher Form in anderen Bereichen genutzt werden.

4.4.1.3 Ausgewählte Ergebnisse

Im Rahmen der Fokusgruppe zeigte sich, dass retrospektiv wahrgenommene Informationen lediglich auf das Herkunftsland sowie quantitative Kennzahlen wie Haltbarkeit, Preis und Gewicht, beschränkt waren und insgesamt eher undifferenziert und oberflächlich erinnert wurden. Informationen zur Haltungs- und Schlachtungsweise sowie zur Futter- bzw. Medikamentenzugabe wurden zwar in der Fokusgruppe als wünschenswerte Information genannt; sie scheinen jedoch in empirischen Studien bei der Kaufentscheidung kaum eine Rolle zu spielen (Andersen, 2011; Harper & Henson, 2001; Olaizola & Corcoran, 2003). Des Weiteren schienen die Verbraucher nicht grundsätzlich, sondern eher ausnahmsweise bewusst und aktiv nach ausführlicheren Informationen zu suchen und nur gewisse Angaben – je nach individuellem Involvement und persönlicher Situation – als relevant einzuordnen. Auch Label schienen hier wenig zu bewirken, da es oft keinen entsprechenden Informationsbedarf gibt. Als Konsequenz wurden Label von der Fokusgruppe überwiegend als unverständlich beschrieben und ihre Fülle und Vielfalt eher als lästig empfunden. Gleichwohl teilten die Verbraucher die Meinung, dass Labels die einzige Möglichkeit böten, sich über die Produkte und deren Eigenschaften am PoS zu informieren. So wurde die Verpackung als einzige Informationsoberfläche angesehen, die neben dem subjektiven Aussehen des eigentlichen Produktes Aufschluss über dessen Merkmale geben könne. Im Gegensatz zu diesen grundsätzlichen Aussagen schien das Informationsbedürfnis bei den Verbrauchern in der Fokusgruppe nur dann erhöht zu sein, wenn durch Skandale – wie BSE in Rindfleisch oder die kritische Berichterstattung über unzureichende Tierhaltungsverfahren – der sorgenfreie Konsum von Fleischwaren eingeschränkt wird. In diesem Fall werden aus vertrauenden Verbrauchern, die sich durch eine durchaus rationale Naivität auszeichnen, offenbar verantwortungsvolle Verbraucher, die ein entsprechend gesteigertes Informationsbedürfnis haben (Micklitz, Oehler, Piorkowsky, Reisch, & Strünck, 2010; Wobker, Lehmann-waffenschmidt, Kenning, & Gigerenzer, 2012). So gaben die Verbraucher an, während solcher Krisenzeiten, welche auch moralischer Natur sein können, einen erhöhten Informationsbedarf zu haben und vermehrt auf Label zu achten oder alternativ auf den ,,Metzger des Vertrauens‘‘ zurückzugreifen.

Im Ergebnis zeigt sich, dass Label eher eine situative Relevanz haben und je nach Kontext und Involvement selektiv durch die Verbraucher wahrgenommen werden. Zwar werden Label eher wahrgenommen und können kaufentscheidungswirksam sein, wenn sie einfach und intuitiv gestaltet und auf der Vorderseite der Verpackung angebracht sind, und durch entsprechende Kommunikationskampagnen begleitet werden (Grunert, 2002; Padilla, Villalobos, Spiller, & Henry, 2007), jedoch sind sie während eines gewöhnlichen Einkaufs eher wenig relevant. In (moralischen) Krisenzeiten hingegen gewinnen sie an Bedeutung, weisen dann aber oftmals zu wenig Informationen auf, so dass ergänzende Informationsquellen, die oft mit Personenvertrauen ausgestattet sind, hinzugezogen werden. Dieses Ergebnis stimmt mit der aktuellen Forschungslage überein, nach der ein individueller, möglichst personalisierter, zeitlich-flexibler und differenzierter Informationsfluss den Verbrauchern in ihren Entscheidungen helfen kann, ohne diese dabei zu überfordern (de Jonge, van der Lans, & van Trijp, 2015; Eberle et al., 2011; Kenning et al., 2017; Reisch, 2003; Weinrich & Spiller, 2016). Abbildung 4.8 verdeutlicht diesen Zusammenhang grafisch.

4.4.1.3.1 Multilayer statt Binarität

Das oben skizzierte Label-Dilemma hat aus Sicht der Verbraucher Konsequenzen: So wird ein hoher Preis bei unzureichender Information über die ‚wahren‘ Produkteigenschaften oftmals als Barriere gesehen (Boogaard, Oosting, & Bock, 2006; Larceneux, Benoit-Moreau, & Renaudin, 2012; Napolitano et al., 2010; Padel & Foster, 2005). Eine Differenzierung innerhalb der offenbar schwankenden preislichen Grenzen und moralischen Ansichten ist somit kaum möglich: Dem einen Verbraucher ist es zumeist zu teuer, dem anderen ist es zu wenig ‚bio‘. Angesichts dieser Heterogenität wäre es zweckmäßig, den Verbrauchern die Möglichkeit zu eröffnen, nach individuellem Involvement diejenige Produktinformation zu beziehen, welche die informierte Kaufentscheidung nach den eigenen, gegebenenfalls zeitlich instabilen, Präferenzen ermöglichen kann. Diese Möglichkeit ließe sich durch die Integration eines sogenannten multilayer Labelsystems eröffnen (de Jonge et al., 2015; Eberle et al., 2011; Weinrich & Spiller, 2016). In diesem System geht es nicht nur um die binäre Unterscheidung zwischen gelabelten und ungelabelten Produkten, sondern es wird innerhalb der Labelstruktur in weitere Stufen (Layer) unterschieden. Durch die Einführung von Differenzierungsebenen können somit psychologische Effekte wie zum Beispiel Kompromiss- oder Anziehungseffekte entstehen, welche den Verbrauchern erlauben, nach ihrem Involvement innerhalb ihrer Preisgrenzen zu entschieden. Erste Implementierungen eines noch recht einfachen angebotsseitig induzierten multilayer Label zeigen sich in den Niederlanden („Beter Leven“) bzw. in Dänemark („Bedre Dyrevelfærd“). Empirische Studien bestätigen, dass die so erreichte Ausdifferenzierung der Labelstruktur zu einem höheren Marktanteil von Tierwohlprodukten führt, die Heterogenität und individuellen Bedürfnisse der Verbraucher besser abgedeckt und auch zeitlich schwankende Zahlungsbereitschaften ‚abgegriffen‘ werden können (de Jonge et al., 2015; Weinrich & Spiller, 2016). Problematisch ist jedoch, dass nach wie vor eine Vielzahl an Information (Tierwohlhaltung, Fairtrade, Gentechnik, Inhaltsstoffe, u.v.m.) auf den Verpackungen angeboten wird, die in den allermeisten Fällen, nicht benötigt wird. So sind einzelne Informationen (z. B. Laktose-, Gluten- oder Nussanteil) nur für spezielle Käufergruppen relevant oder werden nur nach besonderen Vorkommnissen oder moralischen Krisen durch den Verbraucher aktiv nachgefragt. Die damit verbundene Logik ist informationslogistisch ineffizient und kann zudem zu der bereits erwähnten Verwirrung und Überforderung der Verbraucher am PoS führen. Die in Abbildung 4.8 skizzierte Problematik wäre somit allenfalls teilweise behoben. Eine Lösung dieser Problematik könnte darin bestehen, den Informationsfluss nach einem anderen Prinzip zu organisieren und den multilayer Ansatz um eine vertikale, nachfrageorientierte und damit zeitlich flexible Perspektive zu erweitern (Eberle et al., 2011). Im Ergebnis würde der Informationsfluss somit nicht nach einem generellen, zeitlich unflexiblen Push-Ansatz organisiert werden, wie es bei einem klassischen Labelansatz der Fall ist, sondern vielmehr nach einem Pull-Prinzip, welches nicht nur zeitlich flexibel wäre, sondern auch den verschiedenen verbrauchertypenspezifischen Informationsbedarfen/typen, die sich situativ ändern können, entsprechen würde (Micklitz et al., 2010; Wobker et al., 2012). Bei dieser Lösung könnten sich Verbraucher bspw. anhand von aufeinander aufbauenden Fragen, die der Logik sogenannter ereignisorientierter Prozessketten (EPK) entspricht (analog zur Organisation von betrieblichen Informationsflüssen im Rahmen von Managementinformationssystemen (z. B. SAP R/3), die Informationen in der Informationstiefe beschaffen, welche ihrem (situativen) Involvement bzw. Typ entsprechen. Durch bereits bekannte Technologien, wie einen QR-Code über eine Smartphone-App oder einem im Markt installierten Informationsterminal, könnten so verantwortungsvolle Verbraucher bspw. bei einer Produktneueinführung im Bereich der ‚Fleischware‘ das Produkt einscannen. Sie würden dann allgemeine Informationen zum Produkt erhalten, welche zum Beispiel das Produkt zunächst nach einer einfachen multilayer Labellogik kennzeichnen. Anschließend und anhand der integrierten EPK könnten diese Verbraucher individuell detailliertere Informationen zum Produkt auf den verschiedenen Layerebenen erlangen (Abbildung 4.9). Durch diese Konzeption würde der breitere und tiefere Informationsbedarf dieser Verbrauchertypen informationslogistisch effizient befriedigt werden und integriert in einen klar vorgegebenen politischen Rahmen könnte eine entsprechende Labelflut möglicherweise verhindert werden (Eberle et al., 2011).

4.4.1.3.2 Zur Konzeption eines Verbraucherinformationssystems im Rahmen der Nutztierhaltung

Ein dieser Konzeption entsprechender Ansatz, der eine Vielzahl an bereits vorhandenen Verbraucherinformationen integrieren könnte und darüber hinaus eine bedarfsgerechte, situative Informationsbeschaffung ermöglichen würde, bestünde in der Entwicklung eines öffentlich verfügbaren VIS am PoS. Dieses für die Nutztierhaltung durchaus innovative System soll im Folgenden kurz skizziert werden.

Ein Informationssystem ist ein Ansatz, welcher in der Betriebswirtschaft ursprünglich Informationsnachfragen effizient und effektiv in ein System integrieren sollte (Becker & Schütte, 2004; Schütte, 2011). Als VIS kann man analog hierzu ein System bezeichnen, welches den Verbrauchern ermöglicht, durch die Nutzung von Informationstechnologien und dementsprechend informierte Kaufentscheidungen die Wertschöpfungskette nach dem Pull-Prinzip weiterzuentwickeln und mitzugestalten (Tuunanen, Myers, & Cassab, 2010).

Anders als bei Informationssystemen bspw. im beruflichen Kontext, wo Nutzer auf das System für ihre Arbeitstätigkeit angewiesen sind und vor allem Effektivität und Effizienz wichtige Parameter darstellen, sollte bei dem individuellen Gebrauch durch Verbraucher eine Balance zwischen Nützlichkeit und Benutzerfreundlichkeit gefunden werden. Denn nur wenn die Verbraucher einen utilitarischen und hedonischen Nutzen erfahren, wird sich ein solches System dauerhaft etablieren können (Tuunanen et al., 2010). Ähnliche Anwendung zeigen sich bereits für Obst und Gemüse (Max Rubner-Institut, 2015) und im deutschen Bäckerhandwerk (baeckerhandwerk.de). Mit Hilfe eines Informationsterminals können sich Kunden dort über die angebotenen Waren informieren und individuelle Produktinformationen abrufen. Dadurch wird kein unverständliches Etikettierungssystem benötigt und fachkundige Beratungsgespräche werden durch eine weitere Informationsquelle ergänzt. Welche Ansätze und Herausforderungen ein solches VIS aus theoretisch-konzeptioneller Sicht integrieren müsste und welche Treiber der Verbraucher den Gebrauch ermöglichen, wurde bereits in einem ersten Rahmenkonzept zusammengefasst, welches insbesondere den folgenden Aspekten Rechnung trägt (vgl. Tuunanen et al., 2010). Ein VIS sollte die Verbraucher individuell nach situativer und persönlicher Relevanz und Involvement über die Produkte informieren, sodass diese selbst bestimmen, welche Informationen sie wann erhalten wollen. Dies könnte zum Beispiel anhand von QR-Codes in Kombination mit Smartphone-Apps oder Informationsterminals im Markt realisiert werden.

Das VIS sollte eine Schnittstelle zu sozialen Netzwerken beinhalten. Dort sollte eine unabhängige Moderation ergänzt durch Expertenmeinungen von Landwirten, Händlern und Wissenschaftlern stattfinden. Im Bereich der Nutztierhaltung könnten bspw. verifizierte Nutzer (Verbraucher) aktuelle, für sie relevante Themen (z. B. Medienberichte und Warnhinweise) untereinander und mit unabhängigen Experten diskutieren und Erfahrungen (z. B. Meinungen, Kochideen und Angebote) austauschen. Dadurch wird eine Identitätskonstruktion erzielt, welche den Nutzer an den Service bindet und auch ein Crowdsourcing ermöglicht (vgl. Enkel, 2018). Zudem sollte das VIS den Anwendungskontext (z. B. am PoS) berücksichtigen, da dieser einen Einfluss auf das Nutzungsverhalten haben wird. Verbraucher sind in diesem System ein wichtiges, zentrales Element und können die Gestaltung und Nützlichkeit des Informationssystems durch ihren Gebrauch entscheidend beeinflussen (Tuunanen et al., 2010). Der Gestaltungsprozess des VIS ist entsprechend voraussetzungsvoll: Zum einen sind der Zeitpunkt und die Art der Teilnahme an der Gestaltung des Service durch die Verbraucher festzulegen, sodass die Ziele und Ansprüche der Verbraucher an das VIS den gewünschten Nutzen erzeugen. Zum anderen müssen die Informationen, welche in dem VIS verwendet werden, effektiv und effizient aggregiert und integriert werden (vgl. Oehler & Kenning, 2013). Denkbar wäre es, dass bereits vorhandene Systeme kombiniert werden und Informationen aus vertrauensvollen und unabhängigen Quellen integriert werden. Kompatible SAP-Systeme, welche bereits vom Handel genutzt werden, könnten im VIS eine Schnittstelle bilden und so Verbrauchern Informationen zum Beispiel zur Herkunft der jeweiligen Produkte bereitstellen. So könnten bspw. handelsbezogene Daten aus den Warenwirtschaftssystemen freigegeben werden und mit weiteren Daten (z. B. aus dem Bundesinformationszentrum Landwirtschaft, BZL) im VIS zu einem multilayer Informationsansatz aufbereitet und verknüpft werden (Abbildung 4.10).

Um den Handel zu motivieren, die jeweiligen Systeme zu öffnen und zu pflegen wäre es denkbar, die entsprechenden Investitionen zu fördern. Das nötige Vertrauen in das VIS könnte durch die Unabhängigkeit und ein glaubwürdiges Monitoring der Inhalte gewährleistet werden. Die verwendeten Daten sollten von öffentlichen Institutionen wie zum Beispiel der Bundesanstalt für Landwirtschaft und Ernährung (BLE) verwaltet werden, wobei es wichtig ist, dass Informationsstandards vereinheitlicht werden und interne Qualitätskriterien, welche sich zurzeit unter anderem durch private Bio-Label äußern, sichtbar und transparent von unabhängigen Informationen getrennt werden (Beschluss der Verbraucherkommission, 2012). Angesichts dessen böte sich insgesamt ein modularer, integrativer Aufbau an (Abbildung 4.10).

4.4.1.4 Ausblick

Labels bieten mitunter die einfachste Möglichkeit, den Verbraucher am PoS zu informieren. Sie stoßen jedoch oftmals an ihre Grenzen. Aufbauend auf qualitativen und quantitativen Studien kann man rasch erkennen, dass dieser starre Ansatz der Verbraucherinformation möglicherweise durch differenziertere, horizontal und vertikal organisierte Alternativansätze zu optimieren wäre. Sinnvoller wäre es, einen systemischen Ansatz in der Form eines VIS zu verfolgen. Durch ein solches System wäre es möglich, die Informationsbedarfe der Verbraucher flexibel und informationslogistisch optimal zu bedienen.

4.4.2 Beitrag 4.2: Besser statt mehr! Vom Daten-DIY zur ‚Verbraucherinformatik‘

4.4.2.1 Abstract

Die wirksame Gestaltung von Verbraucherinformation ist ein zentrales Anliegen der Verbraucherpolitik. Die in diesem Zusammenhang übliche Anwendung klassischer informationsökonomischer Instrumente (z. B. Gütesiegel und Label) stößt jedoch zunehmend an ihre Grenzen. Es scheint notwendig, Verbraucherinformation wirksamer zu gestalten, wobei aktuelle Ansätze der Informatik hilfreich sein können. Die sogenannte Verbraucherinformatik ist dabei gut beraten, den konkreten Informationsbedarf der Verbraucher zu fokussieren. Dazu ist es wichtig, die Faktoren zu kennen, die wiederum den Informationsbedarf maßgeblich beeinflussen. Aus der Verbraucherforschung sind einige solcher Variablen – wie das Involvement – bekannt. In dieser Studie wird ein digital unterstützendes Verbraucherinformationssystem vorgestellt, das bestehende klassische Informationsinstrumente ergänzen könnte. Das Involvement der Verbraucher wird als wesentlicher Faktor für den Informationsbedarf identifiziert. Konkret wird gezeigt, dass das Involvement der Verbraucher deren Informationsbedarf beeinflusst.

4.4.2.2 Extended Abstract

Verbraucherpolitik möchte Verbraucher darin unterstützen, selbstbestimmte und informierte Entscheidungen treffen zu können. Label und Gütesiegel sind dabei die klassischen Instrumente, um Verbraucher am PoS über die Eigenschaften eines Produktes zu informieren. Das dadurch geschaffene Informationsangebot scheint jedoch den tatsächlichen Informationsbedarf nicht vollständig und vielleicht sogar immer weniger abzudecken (Frey & Pirscher, 2018). Ein Grund dafür könnte sein, dass das Verbraucherverlangen nach Information sehr individuell, im Zeitverlauf dynamisch und von der jeweiligen Situation abhängig ist (Gier, Krampe, Reisch, & Kenning, 2018).

Ein möglicher Einflussfaktor auf den Informationsbedarf könnte das jeweilige Involvement der Verbraucher sein (Frank & Brock, 2018). Da jedoch das Involvement eines Verbrauchers – bestehend aus persönlichem, situativem und objektbezogenem Involvement (Trommsdorff, 2008) – in der Praxis ex ante nicht vorhergesagt werden kann, scheint der mit einem Label oder Gütesiegel verbundene ‚informationsstarre‘ Ansatz begrenzt zweckmäßig. So muss sich der Verbraucher bei Bedarf die jeweiligen Informationen nach dem Do-it-Yourself („DIY“)-Prinzip aus unterschiedlichen Informationsquellen zusammen zu suchen. Ein Umstand der sehr zeitintensiv sein kann und eventuell erklärt, warum Verbraucher das Gefühl der Labelflut empfinden und sich bei Informationen nur auf ein paar wenige Merkmale beschränken (Gier et al., 2018). Eine Alternative hierzu könnte ein flexibles Verbraucherinformationssystem darstellen, in dem Informationen bereitgestellt werden können, die der Kunde falls nötig abrufen kann (Gier et al., 2018).

Ziel der hier skizzierten Studie ist es daher, die Annahme zu prüfen, ob das Involvement der Verbraucher tatsächlich eine wesentliche Determinante ihres Informationsbedarfs darstellt. Dazu wurde ein Online-Experiment mit einem prototypischen Informationssystem durchgeführt. Im Rahmen der Studie wurden 598 Probanden (Probanden hatten keine Einschränkungen in der Ernährung; 56.4 % weiblich, 43.6 % männlich; Durchschnittsalter = 39,58 Jahre, SD_Alter = 11,35), in einem Online-Experiment gebeten, einen hypothetischen Einkauf von Fleischprodukten (Schweine- und Hähnchenbrustfilet) zu tätigen. Während der Kaufentscheidung konnten die Probanden auf ein Verbraucherinformationssystem zur Beschaffung weiterführender Information über die jeweiligen Produkte sowie deren Prozessqualitäten zurückgreifen. Ziel war es zu testen, ob die Inanspruchnahme bzw. die Nutzungsdauer des Systems durch das situative und persönliche Involvement der Verbraucher positiv beeinflusst werden würde (vgl. Abbildung 4.11).

Um diese Annahme zu testen, wurde das situative Involvement im Rahmen des Experiments durch das Lesen verschiedener Nachrichtenartikel über die Nutztierhaltung induziert. Die verwendeten Artikel basierten auf tatsächlichen Berichten, die auf 150–160 Wörter gekürzt und von Ort-/Medienangaben befreit wurden. In einer Vorstudie wurden diese auf Glaubwürdigkeit, Seriosität und Informationsgehalt getestet und zehn Artikel ausgewählt, welche ausreichend Variabilität im situativen Involvement hervorrufen konnten. Ergänzend hierzu wurde das persönliche Involvement zu Beginn des Experiments durch eine modifizierte Form der Animal Attitude Scale erfasst, welche die generelle Einstellung gegenüber der Nutztierhaltung misst (Trommsdorff, 2008).

In der Haupterhebung wurden auf Basis der individuellen Bewertungen der Artikel für jeden Probanden drei Nachrichtenartikel ausgewählt, welche ein hohes, mittleres oder niedriges Involvement auslösten. Nachdem die Probanden einen der vorselektierten Artikel erneut gelesen hatten, wurde randomisiert ein Produkt gezeigt (Schweine- oder Hähnchenbrustfilet) zu dem dann die maximale Zahlungsbereitschaft genannt werden sollte. In dieser Phase der Erhebung hatten die Probanden die Möglichkeit, ein Verbraucherinformationssystem zu nutzen, welches dem jeweiligen Produkt hinterlegt war. So konnten sich die interessierten Probanden vor der Nennung ihrer Zahlungsbereitschaften weiterführend über das Produkt informieren. Nachdem alle Zahlungsbereitschaften genannt worden waren, sollten die Probanden abschließend die wahrgenommene Nützlichkeit des Verbraucherinformationssystems sowie die Nutzungsintention beurteilen.

Um die in Abbildung 4.11 dargestellten Wirkungszusammenhänge zu prüfen, wurden Regressionsanalysen durchgeführt. Diese unterstützen unsere Annahmen auf einem Signifikanzniveau von p < 0,05. Konkret zeigte sich, dass Probanden bei erhöhtem situativem Involvement einen erhöhten Informationsbedarf haben und sich mehr Zeit nehmen, um eine informierte Entscheidung treffen zu können (b = 0,313, t(1432,644) = 2,163, p = 0,031). Dieser Effekt zeigte sich auch für das persönliche Involvement (b = 5,643, t(605,755) = 3,879, p < 0,001), wobei der Effekt von situativem Involvement verstärkt wird bei Probanden mit hohem persönlichem Involvement (b = 0,239, t(1489,627) = 2,256, p = 0,024). Die Studienergebnisse verdeutlichen zudem, dass der Informationsbedarf einen positiven Einfluss auf die Nutzungsintention des Informationssystems hat (b = 0,004, t(593) = 8,944, p < 0,001).

Zusammenfassend lässt sich feststellen, dass im vorliegenden Experiment der Informationsbedarf je nach persönlichem und situativem Involvement variiert. Die im Markt verfügbaren Instrumente – zumeist handelt es sich um ‚informationsstarre‘ Labels – können diese Variabilität kaum abdecken (Herzog, Betchart, & Pittman, 1991). Sie lösen mithin nicht das Problem, dass es inter- und intrapersonal unterschiedliche Informationsbedarfe gibt. Ein flexiblerer Ansatz, der Informationen aufbereitet und im Bedarfsfalle bereitstellt, wäre somit eine sinnvolle Alternative oder auch Ergänzung zu den bisherigen Informationsinstrumenten (Gier et al., 2018). Sie könnten auch eine erste Entwicklungsstufe einer umfassenderen, noch zu entwickelnden Verbraucherinformatik darstellen, mit dem Ziel Verbraucher durch Informationssysteme in ihrem Einkaufs- und Konsumverhalten zu unterstützen (Stevens & Boden, 2014).

4.5 Beitrag 5: Measuring dlPFC signals to predict the success of merchandising elements at the Point-of-Sale – A fNIRS approach

Der Beitrag entspricht der folgenden Publikation: Gier, N. R., Strelow, E. & Krampe, C. (2020). Measuring dlPFC signals to predict the success of merchandising elements at the Point-of-Sale – A fNIRS approach. Frontiers in Neuroscience. https://doi.org/10.3389/fnins.2020.575494

4.5.1 Abstract

The (re-)launch of products is frequently accompanied by PoS marketing campaigns in order to foster sales. Predicting the success of these merchandising elements at the PoS on sales is of interest to research and practice, as the misinvestments that are based on the fragmented PoS literature are tremendous. Likewise, the predictive power of neuropsychological methods has been demonstrated in various research work. Nevertheless, the practical application of these neuropsychological methods is still limited. In order to foster the application of neuropsychological methods in research and practice, the current research work aims to explore, whether mobile fNIRS – as a portable neuroimaging method – has the potential to predict the success of PoS merchandising elements by rendering significant neural signatures of brain regions of the dlPFC, highlighting its potential to forecast shoppers’ behaviour aka sales at the PoS. Building on previous research findings, the results of the given research work indicate that the neural signal of brain regions of the dlPFC, measured with mobile fNIRS, is able to predict actual sales associated with PoS merchandising elements, relying on the cortical relief effect. More precisely, the research findings support the hypothesis that the reduced neural activity of brain regions associated with the dlPFC can predict sales at the PoS, emphasising another crucial neural signature to predict shoppers’ purchase behaviour, next to the frequently cited reward association system. The research findings offer an innovative perspective on how to design and evaluate PoS merchandising elements, indicating fruitful theoretical and practical implications.

4.5.2 Introduction

The (re-)launch of products is frequently accompanied by PoS marketing campaigns, given that effective PoS merchandising elements have been shown to significantly increase sales of advertised products (Sinha & Verma, 2017). Predicting the success of these PoS marketing campaigns in terms of the company’s objectives, for example forecasting the sales before its launch, is of substantial economic importance. An aspect that is reflected in the multibillion-dollar investments companies spend on advertising and merchandising each year (Guttman, 2019). Consequently, a significant amount of research investigated the PoS and its effective design. In this regard, previous PoS research examined in particular the assortment size, the in-store design and the PoS atmosphere. The assortment size and the associated choice overload effects have been investigated most frequently, identifying the circumstances and operating principles in form of an inverted U-shape function between variety and purchase probability (Beneke et al., 2013; Chernev, 2006; Chernev et al., 2012; Grant & Schwartz, 2011; Heitmann et al., 2007). Other research examined PoS in-store demonstrations, product presentations and consumer inspiration, which showed positive effects on attention and evaluation processes of consumers (Bottger et al., 2017; Huddleston et al., 2015; Nordfält & Lange, 2013; Phillips et al., 2015; Townsend & Kahn, 2014). Considering the sensory complexity of the PoS, previous research investigated also the store environments and the PoS atmosphere, exploring how multisensory aspects like music, scent and touch influence shopping behaviour in combined fashion. The results indicate that congruent and matching modalities seem to be most favourable by consumers (Mattila & Wirtz, 2001; Michel et al., 2017; Quartier et al., 2014; Spence & Gallace, 2011; Spence et al., 2014). Although it has been shown that investments in PoS atmospherics and product arrangements can pay off, most merchandising activities are still associated with high costs (Spence et al., 2014). Moreover, many operating stimuli at the PoS that have been shown to greatly influence shoppers are only analysed in isolation without considering the complexity of the entire PoS and its various influencing factors. Consequently, the efficient and effective prediction of the success of PoS marketing campaigns on market level is of great interest for research and practice, given that it might provide a holistic picture of the marketing activities at the PoS that may reduce misinvestments. It is, thus, not surprising that retailers and producers, who launch and promote a myriad of new product variations every year, try to implement marketing campaigns that have been effectively tested before.

The selection of merchandising elements is frequently grounded on insights that are received from exploring the consumers’ perceptions of the – advertised – product or service-associated attributes. In order to measure the consumers’ perceptions of these attributes, self-report measurements are often used, asking consumers directly about their subjective opinions in regard to a product or service. Although self-report measurements have been indicated to be beneficial in some marketing studies, social psychology suggests that self-reports, when used in isolation, are unreliable to accurately predict the consumers’ preferences (Nisbett & Wilson, 1977; De Cremer et al., 2008; Petit & Bon, 2010; Baldo, Paulraj, Curran, & Dronkers, 2015). This is mostly because the consumers’ expressed intentions do not always translate into actual (purchase) behaviour or even sales (Ajzen, 1991; Frank & Brock, 2018; Padel & Foster, 2005). Against this background, other measurements might be more expedient to solve the indicated matter (Ariely & Berns, 2010; Karmarkar & Yoon, 2016; Plassmann et al., 2015).

The application of neuropsychological methods, using neural brain activity data to forecast products and marketing campaigns success, has been indicated to offer a promising approach to gain further knowledge about the consumers’ perception processes (Ariely & Berns, 2010; Berns & Moore, 2012; Daugherty et al., 2016; Falk et al., 2012; Falk et al., 2016; Karmarkar & Yoon, 2016; Kühn et al., 2016; Motoki et al., 2020; Plassmann et al., 2015; Tong et al., 2020; Venkatraman et al., 2015). Plassmann and colleagues (2007) explored, for example, how neuropsychological methods could be used to investigate brand equity as a determining factor that influences the perception and, consequently, the behaviour of consumers. Subsequently, multiple studies demonstrated the predictive power of neuropsychological data, displaying the capability of forecasting music and movie success or advertising elasticities of television ads (Baldo, Parikh, Piu, & Müller, 2015; Boksem & Smidts, 2015; Cha, Suh, Kwon, Yang, & Lee, 2019; Tong et al., 2020; Venkatraman et al., 2015). Although the predictive power of neuropsychological methods has been demonstrated to outperform ‘traditional’ marketing methods (Venkatraman et al., 2015), neuropsychological methods and the generated neuropsychological insights are only partially adapted in practice. One reason for this might be that previous research often emphasised reward associations in order to predict sales with the utilisation of neuropsychological methods (Ariely & Berns, 2010; Plassmann et al., 2015). Thereby the predictions rely on medially and subcortical located brain regions of the reward evaluation system, such as the nucleus accumbens (NAcc), the ventral striatum (vStr), the OFC and the vmPFC. These brain areas can only be measured with stationary neuroimaging methods, such as fMRI, whose application is quite costly and time-consuming. However, although just recently a study conducted by Cha et al. (2019) indicated that the application of fNIRS allows to correlate mPFC neural activity to popularity of music on YouTube, another – in previous research often neglected – neural signature might as well be decisive to predicting PoS sales, namely the deactivation of the dlPFC. The dlPFC is known to play a major role in decision-making by integrating cognitive evaluations whilst modulating affective reward responses (Hare et al., 2009). Frequently, increased dlPFC activity is associated with cognitive (self-)control in decision-making and other cognitive processes such as working memory, abstract problem solving and exertion of control in order to favour long-term goals (Carlén, 2017; Hare et al., 2009; Miller & Cohen, 2001). For example, in food-related value-based decision-making increased neural activity in brain areas of the dlPFC have been identified for participants that execute a greater self-control on their food choice (Hare et al., 2009). Simultaneously, a reduced neural activity of the dlPFC has been associated for brand-related decisions that require less strategy-based reasoning (Deppe et al., 2005; Koenigs & Tranel, 2008; Krampe, Gier, et al., 2018; Schaefer & Rotte, 2007). First shown in the study by Deppe et al. (2005), decision sets that include the participants favoured brand, emotionalise the choice, which allows a quicker, straightforward and less complex decision-making process in favour of the preferred product, a replicated and robust effect called cortical relief effect.

In conclusion, preferred choice options seem to be easier to process, which makes it easier to choose for the favoured product during a decision-making process that seem to be less cognitively controlled and assumed to elicit a reduced activity in brain regions of the dlPFC (Deppe et al., 2005; Koenigs & Tranel, 2008; Krampe, Gier, et al., 2018; Schaefer & Rotte, 2007). Less self-controlled decisions might, therefore, result in more impulsive decision-making, choosing the option that is preferentially presented in a choice situation (Boettiger et al., 2007; Hare et al., 2009; Kable & Glimcher, 2007). Consequently, merchandising elements that are about to expose a reduced neural activity in brain regions ascribed to the dlPFC might be less cognitively engaging, resulting in more impulsive decisions, which might rescale in increased sales at the PoS. Hence, while earlier neuropsychological studies that aimed to predict consumer behaviour on population level with neuropsychological methods focussed mainly on medial and subcortical located brain regions of the reward evaluation system; only a few studies considered the dlPFC in their prediction models. Consequently, this research work is one of the first to evaluate whether the reduced neural dlPFC activity, as a neural signature, can predict PoS sales, building on insights of the cortical relief effect.

Having this in mind, the current research work aims to explore the predictive power of the cortical brain regions of the dlPFC to forecast the success of PoS merchandising elements. By doing so, the given research work overcomes the limitations of stationary neuroimaging methods by utilising mobile fNIRS as a portable applicable neuropsychological method for the research field of shopper neuroscience, demonstrating its potential application in ecological valid setting, such as the PoS. (Çakir et al., 2018; Kopton & Kenning, 2014; Krampe et al., 2018b). Against this background, the given research work aims to explore whether mobile fNIRS – as a mobile applicable neuroimaging method – has the potential to predict the success of PoS merchandising elements by rendering significant neural cortical relief signatures of the dlPFC.

4.5.3 Predicting success of PoS merchandising elements – the ‘duplo’ case

A special case in the analyses of PoS merchandising elements is the product ‘duplo’ by Ferrero (Ferrero Deutschland GmbH, n.d.). ‘Duplo’ constitutes a special case for research, since its effects on shoppers’ processing and behaviour were not only explored in prior studies with neuropsychological and traditional marketing methods (Kühn et al., 2016; Strelow et al., 2020; Strelow & Scheier, 2018), allowing comparisons between different data types, but also provide unique, real-market stimuli materials for research, that are, in contrast to research stimuli specifically designed for a study, highly ecologically valid. The product ‘duplo’ was introduced to the German market in 1964 and is currently the market leader of chocolate bars in Germany, with a turnover of € 200 million (VuMA, 2019b). There, more than 50 % of the turnover is achieved by secondary (out of shelf) displays, which are displayed with PoS merchandising elements (Briesemeister & Selmer, 2020). Over the past 40 years, many PoS merchandising elements have been used to promote the chocolate bar. Six merchandising elements were explored by prior research, representing a typical choice set for marketing campaigns, including past and recent PoS and TV campaigns as well as similar but unknown merchandising elements (Abbildung 4.12).

An fMRI study conducted by Kühn et al. (2016) investigated the different PoS ‘duplo’ merchandising elements on neural level. In particular, two fMRI-derived sales prediction values were extracted based on the neural BOLD (blood oxygen level dependent) signals measured (1) during the perception of the merchandising elements contrasted to the implicit baseline and (2) for the signal change from the baseline contrast of (the advertised) package ‘duplo’ product seen before and after the merchandising element. The fMRI-derived sales prediction values summarised the signal of multiple neural regions, whereby the prediction was mainly driven by the neural activity of the reward system (NAcc and medial OFC) and the deactivation of the dlPFC (Brodmann area 9 and 46). Furthermore, explicit subjective ratings of the ‘duplo’ merchandising elements were evaluated. In order to measure the actual sales – defined as the revenue generated by the different merchandising elements – the merchandising elements were tested at the PoS in a field experiment in parallel to the fMRI study (for detailed information, please see Kühn et al., 2016; Abbildung 4.13D). Results demonstrated that the fMRI-derived sales prediction value based on the merchandising element presentation were the best predictor for the sales numbers (Abbildung 4.13A). While the first two and last two ranking positions were equivalent between fMRI-derived sales prediction value of merchandising elements and actual sales, only one match at the third position was found for the subjective rankings (Abbildung 4.13C) and no match for the fMRI-derived sales prediction value of the product contrast (Abbildung 4.13B). Inspecting the integrated neural brain areas ad-hoc in detail, Kühn et al. (2016) identified the medial OFC as most predictive for actual market sales.

In order to explore the shoppers’ associations with the different PoS merchandising elements and to understand the shopper response to the merchandising elements, following the fMRI study, the merchandising elements were examined in a second study conducted by Strelow and Scheier (2018), utilising an implicit reward association test (IAT). During the IAT, each PoS merchandising element as well as the brand itself were assessed on different reward values that were spontaneously associated with the brand and the merchandising element. From the results of the IAT for the merchandising elements, Strelow and Scheier were able to discriminate the lower three merchandising from the top three merchandising elements, although the ranking order was not congruent with the actual sales numbers identified by Kühn et al. (2016; Abbildung 4.13E). Subsequently, the fit between the merchandising elements and the brand’s reward associations was analysed, indicating that the first and last two ranks of the actual PoS sale performance can be determined by the data (Abbildung 4.13F). The fit of the brand associations with the merchandising element associations can be interpreted either as an enhancement or at least as a confirmation of the brand reward associations representing the degree of congruence between the expected associations elicited by the brand and the associations evoked by the brands merchandising elements.

In conclusion, data from both (neuro)psychological methods, the fMRI data and the IAT data, seem to outperform self-report shoppers’ ratings of the merchandising elements. A high brand-fit score as indicated by Strelow and Scheier (2018) between the merchandising element and the brand seems to be predictive for the success of a merchandising element, since the shoppers’ expected and experienced brand associations are congruent with the merchandising element, potentially resulting in a cortical relief effect, reducing the experienced cognitive dissonance. In the study conducted by Kühn et al. (2016) the fMRI-derived sales prediction value based on the merchandising element presentation were most predictive for actual sales data. Although, the brain regions of reward evaluation system, especially medial OFC, were again highlighted as the driving force for the prediction, a decreased neural activity in the dlPFC was integrated in the formula to predict sales, an aspect that represents reduced cognitive effort and greater cortical relief (MacPherson et al., 2002; Carter & van Veen, 2007; Cho et al., 2010; Izuma et al., 2010; Bartra et al., 2013). Building on previous research, which demonstrated that mobile fNIRS is particularly capable of measuring neural cortical activity, especially lateral areas of the PFC (Krampe, Gier, et al., 2018; Kühn et al., 2016; Liu, Kim, & Hong, 2018), the investigation of the neural signatures of the dlPFC’s deactivation might be a fruitful avenue to predict the success of merchandising elements. While doing so, this research work opens up the potential application of mobile fNIRS in a realistic shopping environment, namely the PoS, to predict success on market level. Hence, the given research work aims to explore, whether the dlPFC can act as a predictive neural signature for actual market sales by utilising and validating mobile fNIRS as a mobile neuropsychological method for the research field of shopper neuroscience, leading to the following hypothesis:

The neural signatures of the dlPFC during the perception of merchandising elements measured with mobile fNIRS are able to predict the sales associated with the PoS merchandising elements.

4.5.4 Materials and methods

4.5.4.1 Participants

In line with previous research (Krampe, Gier, & Kenning, 2018; Krampe et al., 2018b; Kühn et al., 2016; Rampl, Eberhardt, Schütte, & Kenning, 2012; Strelow & Scheier, 2018) only healthy, female participants (N = 45), who indicated that they were mainly responsible for the grocery shopping in their household, were recruited to participate in this study. Female participants were recruited because women are more frequently responsible for the household’s grocery shopping (BVE, 2020; VuMA, 2019a). Due to bad signal quality, 12 participants had to be excluded from the data analysis, resulting in a final sample size of n = 34 (M_age = 41.06, SD_age = 8.41; Age_min = 23, Age_max = 54). All participants were right-handed and had no history of major psychological or neurological disorders.

4.5.4.2 Experimental task procedure

After participants were welcomed, they were informed verbally and in written form about the aim of the study, the task and the utilised mobile fNIRS device. Once participants fully understood the task, a written informed consent was signed in accordance with the Declaration of Helsinki. Thereafter, participants were seated in front of a computer screen and the mobile fNIRS headband was attached on the participants forehead. In order to increase consistency between the participants measured brain regions, the mobile fNIRS headband was locally standardised on the vertical axis using the craniometric point of the nasion as an orientation point and the middle of the two preauricular points for positioning on the horizontal axis, covering the PFC. Before starting the experimental task, data quality was checked and, if necessary, signal quality was improved by shifting the hair away from the detectors, making direct skin contact. In addition, the fNIRS headband was covered with a light-protecting cap to control for external light sources. Once the preparation was finished, participants were instructed to look at the computer screen while the task was performed.

The task was designed analogous to the paradigm developed by Kühn et al. (2016; Abbildung 4.14), applying an event-related experimental design. During the task, a merchandising element was displayed for 3 s, followed by a randomised jitter of 4–6 s. Before and after the merchandising element, the advertised product was shown for 2 s, again followed by a randomised jitter of 4–6 s. In total, every merchandising element was shown six times, whereby the order of the merchandising elements was totally randomised. The task was performed twice, resulting in a total number of 72 trials, with 12 trials for every of the six merchandising elements. After completing the task, the mobile fNIRS device was removed and participants were asked to complete a final questionnaire, assessing demographics as well as their explicit subjective ranking of the merchandising elements. At the end of the study and a verbal disclosure, participants received a monetary incentive for their participation and were free to leave.

4.5.4.3 fNIRS data collection

The continuous-wave fNIRSport-System (NIRx Medical Technologies, Berlin, Germany) was used for data collection (Boas et al., 2014; Scholkmann et al., 2014). In general, fNIRS measures cerebral haemodynamic responses through near-infrared light sources (Ferrari & Quaresima, 2012). The mobile fNIRS system recorded optical signals on two-wavelengths (760 and 850 nm) at a sampling rate of 7.81 Hz. As imaging depth increases with emitter-detector distance, but signal quality is suggested to be best at a separation of 3 cm, the optodes and diodes are set to the distance of 3 cm (Ferrari & Quaresima, 2012; Gagnon et al., 2012; Gratton et al., 2006; McCormick et al., 1992; Naseer & Hong, 2015). The system consists of 22 channels, compromising eight light sources and seven detectors (Abbildung 4.15). In order to identify the equivalent brain areas of Brodmann area 9 (Abbildung 4.15C1) and 46 (Abbildung 4.15C2), the dlPFC definition had to be transferred to the mobile fNIRS optode montage setup (Abbildung 4.15A). Channels classified as relevant to cover Brodmann area 9 are Ch2, Ch5, Ch7, Ch8, Ch9, Ch10, Ch12, Ch13 and Ch14, and for Brodmann area 46 are Ch16 and Ch21 (Abbildung 4.15B). The NIRS-Star software package (version 14.2) was used for checking signal quality and data collection.

The valid application of mobile fNIRS in the field of consumer and shopper neuroscience has been demonstrated in several studies (Çakir et al., 2018; Kopton & Kenning, 2014; Krampe, Gier, & Kenning, 2018; Krampe et al., 2018b). Most of the consumer neuroscience research using fNIRS focussed on the identification of neural correlates associated with merchandising in virtual in-store settings (Krampe et al., 2018b; Liu et al., 2018) or used fNIRS measurements to predict individual food-choice behaviour (Çakir et al., 2018). A recent fNIRS study conducted by Cha et al. (2019) correlated neural activation patterns of the mPFC to online popularity of pop music on YouTube, presenting an extension of earlier studies that predicted music popularity in the field of consumer neuroscience applying fMRI (Berns & Moore, 2012). Overall, prior fNIRS research suggested that especially cortical regions are measurable, whilst brain regions located medially within in the brain or subcortically are not assessable with mobile fNIRS (Krampe, Gier, et al., 2018). Furthermore, most of previous fNIRS studies focused on the medial brain regions, with only one study correlating neural activity pattern to behaviour on population level. As a result, the predictive value of lateral brain areas has not yet been addressed and mobile fNIRS as an innovative neuropsychological method in the field of consumer and shopper neuroscience, requiring further profound and robust validation.

4.5.4.4 fNIRS data analysis

In order to analyse the collected data, data was pre-processed using the NIRx Software Package (NIRx Medical Technologies, Berlin, Germany). In order to increase signal quality, channels exhibiting discontinuous shifts during the measurement were removed. Furthermore, fNIRS data time series were smoothed, applying a band-pass filter (high and low frequency filter; Naseer & Hong, 2015; Pinti et al., 2019) with the frequently applied low cut-off frequency of 0.01 Hz and high cut-off frequency of 0.2 Hz (Franceschini, Fantini, Thompson, Culver, & Boas, 2003; Hu, Hong, & Ge, 2012; Krampe, Gier, & Kenning, 2018; Nissen et al., 2019; Spichtig, Scholkmann, Chin, Lehmann, & Wolf, 2012) in order to control for physiological noises and artefacts such as heartbeat and Mayer waves (Naseer & Hong, 2015; Pinti et al., 2019; Scholkmann et al., 2014). The modified Beer-Lambert law was used to convert raw light absorption rates into haemoglobin concentrations (Kocsis et al., 2006; Kopton & Kenning, 2014; Scholkmann et al., 2014). Haemodynamic states were computed in accordance with commonly used pathlength factors (for 750 nm set to 7.25 and for 850 nm set to 6.38; Essenpreis et al., 1993; Kohl et al., 1998; Zhao et al., 2002). For the further analysis only oxy-Hb signals were interpreted, as they seem to better correlate with cerebral blood flow (Hoshi et al., 2001). Information on the oxy-Hb concentrations is available in the supplementary material (Abbildung 4.20).

A GLM was set up for every participant and convolved with the haemodynamic response function, including six regressors with one for each merchandising element and an additional 12 regressors for the product stimuli (six before and six after each merchandising element). The GLM was first calculated on a single subject individual level (within-subjects level), and subsequently, a second-level group contrasts analysis was carried out to calculate neural activations across subjects (between-subjects level). In order to extract standardised activation values, a t-contrast was executed for each merchandising element against the implicit baseline, using the t-values in the further analysis. Given that significant activation differences are not of interest, the contrast analysis was used as a procedure to standardise the neural activations, which made a multiple comparison correction redundant. To test the hypothesis, fNIRS-derived sales prediction values were calculated from the standardised activation values of the t-contrasts for every merchandising element, respectively (Formel 3). The resulting fNIRS-derived sales prediction values can be interpreted according to their degree of reduced dlPFC neural activity.

Formel 3. Formula for fNIRS-derived sales prediction value.

The t-values of channel Ch2, Ch5, Ch7, Ch8, Ch9, Ch10, Ch12, Ch13 and Ch14 were allocated to represent Brodmann area 9, while for Brodmann area 46 the channel Ch16 and Ch21 were defined. This calculation was performed for each merchandising element, resulting in six fNIRS-derived sales prediction values per Brodmann area (9 and 49).

$$\begin{aligned} &Let\ Ch_{x}\ be\ defined\ as\ the\ signal\ value\ of\ fNIRS\ channel\ x\\ &on\ the\ contrast\ of\ a\ merchandising\ element\\ &against\ the\ implicit\ baseline: \end{aligned}$$

$$fNIRS\ derived\ sales\ prediction\ value = \mathop \sum \limits_{{x_{i} = x_{1} }}^{{x_{n} }} Ch_{{x_{i} }}$$

$$\begin{aligned} &for\ Brodmann\ area\ 9\ D_{x} = \left\{ {2;5;7;8;9;10;12;13;14} \right\}\ and\\ &for\ Brodmann\ area\ 46\ D_{x} = \left\{ {16;21} \right\}.\end{aligned}$$

Hence, the fNIRS-derived sales prediction values for Brodmann area 9 and 46 were used to rank the order of the merchandising elements from lowest to greatest values, whereby a greater neural deactivation (more negative value) corresponds to a higher rank. Thus, the ranking is a result of the least neural activity, displaying less cognitive interfered processing (cortical relief effect) that is hypothesised to translate to sales at the PoS. Consequently, the resulting rank order based on the reduced dlPFC signal values should coincides with the rank order of the actual sales data. In order to evaluate the predictive success of the fNIRS-derived sales prediction values rankings with the original sales data, the results were compared qualitatively and based on Spearman rho correlation coefficients for the ordinal rank orders as well as on Pearson correlation for the quantifiable sales prediction values and actual sales data at a significance threshold of p < 0.05.

Conclusively, based on the neural data analysis two different types of dlPFC fNIRS-derived sales prediction values were extracted and rank ordered, according to their degree of the reduced dlPFC activity. First, the fNIRS-derived sales prediction values of Brodmann area 9; and second of Brodmann area 46, calculated from the contrasts of each merchandising element against the implicit baseline, have been evaluated. The participants’ explicit subjective rating of the merchandising elements was also evaluated, whereby the total number of 1st rank positions for each merchandising element was taken as an indicator. Finally, and in order to estimate the predictive power of the different data types, the actual sales associated with the merchandising elements – defined as the revenue generated by the different merchandising elements – were adopted from Kühn et al. (2016), who explored the revenues generated by the merchandising elements on a quarter display at the PoS in a German supermarket (for detailed information on data, data collection and analysis, please see Kühn et al., 2016).

4.5.5 Results

Supporting the hypothesis, the results suggest that the neural sales prediction values of brain regions of the dlPFC calculated from the merchandising contrasts (Abbildung 4.16), are able to predict the actual sales associated with PoS merchandising elements. The best predictor is the fNIRS-derived sales prediction values of Brodmann area 46. This finding was confirmed by the correlation analyses that revealed a positive significant Spearman rho correlation on the rank order data (r_s = 0.943, n = 6, p = 0.005) and a positive significant Pearson correlation on the sales prediction values and actual sales (r_p = 0.868, n = 6, p = 0.025; Abbildung 4.17). For the qualitatively comparisons with the actual sales data ranking (Abbildung 4.18i), this rank order has all rank positions matched with the exception of the last 4th and 5th positions, which are reversed (Abbildung 4.18A).

Similarly, the neural results reveal that the first rank position based on the calculated Brodmann area 9 fNIRS-derived sales prediction value of the merchandising contrast (Abbildung 4.18B) corresponds to the rank positions of the actual sales data. However, the associated correlations on rank order and sales prediction value with the actual sales data failed to reach significance threshold of p < 0.05 (r_s = 0.771, n = 6, p = 0.072; r_p = 0.648, n = 6, p = 0.164). For the explicit subjective ranking no matched rank positions could be identified qualitatively (Abbildung 4.18C), confirmed by small, non-significant correlations with the actual sales data (r_s = −0.29, n = 6, p = 0.577; r_p = 0.309, n = 6, p = 0.551). The t-values on each channel and scatterplots on the non-significant predictors are available in the supplementary material (Tabelle 4.1; Abbildung 4.19). Thus, fNIRS-derived sales prediction values aggregating the channels constituting Brodmann area 46 could resample the actual sales data best.

4.5.6 Discussion

The current research work aims to explore the predictive power of brain regions ascribed to the dlPFC to forecast the success of PoS merchandising elements, thereby validating mobile fNIRS – as a portable applicable neuropsychological method – and opening up its potential application in realistic shopping environments, such as at the PoS. As one of the first studies, this research work evaluates the neural signatures of the dlPFC deactivation in isolation to predict market sales success with mobile fNIRS, building on the cortical relief effect. More precisely, the integration of mobile fNIRS in the field of shopper neuroscience has been used to investigate six PoS merchandising elements, which have been examined with marketing methods in earlier studies, while overcoming the limitations associated with stationary neuroimaging methods (Kühn et al., 2016; Strelow & Scheier, 2018). The research findings support the hypothesis that the deactivation of the dlPFC is predictive for the shopper behaviour aka sales at the PoS, highlighting an additional crucial neural signature measurable with mobile fNIRS. The results show that fNIRS-derived sales prediction values of Brodmann area 9 and 46 are capable of predicting the actual sales of PoS merchandising elements, whereby Brodmann area 46 (consisting of channels 16 and 21) seem to be the most predictive brain area of the dlPFC.

In the context of prior studies on the ‘duplo’ case, the current research findings suggest that merchandising elements promoting a brand are processed in two neural signatures of the (prefrontal) cortex, leading to different cognitive processes. Whereas in the past the neural activity of the reward evaluation system has been used to predict marketing, advertising and sales effects at the PoS, the role of cortical relief effects and reduced cognitive controlled processes have been neglected. Although occasionally studies integrated the dlPFC besides other brain regions in their prediction models, cortical relief processes have – to the best of the authors’ knowledge – not yet been used to predict and explain purchase behaviour at the PoS.

Supposing that 70 % of the purchases at the PoS are spontaneous and given that an act of purchase takes approximately about 60 seconds (Hertle & Graf, 2009; Valizade-Funder & Heil, 2010), it is suggested that an habituative, less self-controlled process takes place in most of the purchases (Rook & Fisher, 1995). Consequently, any kind of irritation that disrupts the state of cortical relief by incongruency or aspects that require more cognitive effort could potentially interrupt the act of impulsive purchase, resulting in a termination or, at least, a delay in the cognitive or affective purchase process of shoppers. This effect seems to be particularly relevant when shoppers experienced a conflict between their perceived brand image and the triggered reward associations elicited by the PoS merchandising element – a neuropsychological process, which seem to result in an increased neural cortical dlPFC activity (Deppe et al., 2005; Kato et al., 2009; Koenigs & Tranel, 2008; Krampe, Gier, et al., 2018; Plassmann, Ambler, Braeutigam, & Kenning, 2007) and which could be measured with mobile fNIRS. Likewise, the congruency of the brand image and the associated PoS merchandising element might result in a neuropsychological (cortical) relief effect for congruent brand-merchandising PoS elements or vice versa result in an increased neural activity effect in the dlPFC, when the product and merchandising element are perceived as incongruent. Both effects can, consequently, be measured in brain regions of the dlPFC, indicating its specificity to predict sales at the PoS. Consequently, next to the reward association system, brain regions of the dlPFC might also function as a process variable to predict sales in a PoS setting. The utilisation of mobile fNIRS with its technical capabilities to measure cortical brain regions might, therefore, provide an innovative and fruitful method for future research.

4.5.6.1 Implications

The research findings provide several implications for marketing theory and practice. First, from a theoretical perspective, the research findings suggest that the shopper behaviour at the PoS is not only driven by reward associations offered by brands, but is also influenced by the perceived (in-)congruency and the level of conflicts or cortical relief experienced between the shoppers’ brand image and the experienced PoS merchandising element. While earlier neuropsychological studies investigated mainly medial and subcortical located brain regions of the reward evaluation system to forecast population success; only a few studies considered the dlPFC to predict shoppers’ behaviour. Consequently, this research work is one of the first that evaluates the predictive power of brain regions ascribed to the dlPFC neural deactivation, providing an innovative approach to interpret consumer responses to merchandising elements at the PoS. Second, as a methodological contribution, the validation of a mobile and in its application fast-growing methodology of mobile fNIRS demonstrates its potential to predict success in real-world settings such as the PoS. Due to its mobile application, it provides a great variety of application options for research and practice to measure shoppers’ neural responses directly in complex settings such as the PoS, increasing the ecological validity of research results. From a practical point of view, the research results offer an innovative perspective on how to design, evaluate or forecast the success of PoS merchandising elements in combination with the to-be-advertised products – including all kind of merchandising elements such as lighting, furnishing, display screens, price tags and information displays. Cortical relief disrupting conflicts can arise on all levels of the customer journey, beginning with the perception of a stimulus and ending in cognitive overload effects elicited by, for example, the overwhelming assortment in the shelves. To carefully match the shoppers’ brand image with PoS merchandising elements in order to reduce conflicts and cognitive dissonance might, consequently, be of high value for producers and retailers. The integration of the idea to investigate the (in-)congruency and potential conflicts as well as its repercussions enables the analysis of the shoppers’ PoS journey by evaluating different merchandising elements, with its aim to reduce or at best avoid conflicts in the perception of the product specific attributes (e.g. the brand image) and the PoS merchandising elements to be used. A comprehensive investigation of all cues that appear at the PoS during a customer journey, to explore all potential reactions of the shoppers’ brain during a shopping trip, to identify cues that potentially reduce the overall net-incongruence at the PoS, might be beneficial. The neuropsychological neuroimaging method of fNIRS may, therefore, be of particular interest as it enables the investigation of the hypothesised effect directly at the PoS because of its mobile, ecological valid usability. Following from this, the research results might be used to explore different PoS merchandising elements to quantify the cognitive engagement represented by the neural activity of the dlPFC evoked by a shopping trip, measured with the use of mobile fNIRS. The ultimate goal would be a measurement of all rewarding and conflicting cues during an average shopping trip, possibly enhanced by the identification of additional motivating cues, to generate a deeper understanding of the shoppers’ behaviour at the PoS.

4.5.6.2 Limitation and future research suggestions

One aim of the research work is to indicate the usefulness of mobile fNIRS to predict shopper behaviour at the PoS. The current study provides a first step to actually measure shoppers’ neural activity, when confronted with PoS merchandising elements and products at the PoS, using mobile fNIRS. Nevertheless, this research work investigates the neural signatures on basis of a laboratory setting with an experimental paradigm performed in front of a computer screen. The next logical step for future studies should be to explore whether the research findings received under laboratory settings remain also valid in a naturalistic environment measurement at the PoS, utilising mobile fNIRS in realistic PoS settings. Furthermore, mobile fNIRS is a relative innovative neuroimaging method, at least for the research field of shopper neuroscience, indicating the need to consider the continuous development of its technical capabilities. Future research might, thus, use other more advanced mobile fNIRS devices to improve data quality and reduce the application costs. Finally, whilst interpreting the neural activity and the neural reactions associated with PoS merchandising elements, it is implicitly assumed that the cortical relief effect is measured. However, it might be that the merchandising elements have been seen in a TV or PoS campaign before, leading to the measurement of a familiarity effect. This effect might be evoked because the familiar merchandising element might require less cognitive effort to be processed, resulting in a reduced neural activity of the dlPFC. In order to cope with this potential limitation, future studies might replicate the given study with only novel PoS merchandising elements that vary in the degree of their brand fit.

4.5.7 Conclusion

Whereas previous research work mainly focused on the reward association system and its associated subcortical brain regions to predict sales, utilising stationary neuroscientific methods (e.g. Berns & Moore, 2012; Venkatraman et al., 2015; Tong et al., 2020), the research findings of the current study not only suggest that the shoppers’ reward associations seem to be predictive for sales at the PoS, but indicate the importance of the conflicts perceived by the shopper and the congruency between the perceived brand image and the displayed PoS merchandising elements. In other words, the research results signify that the brand ‘duplo’ activates expectation of rewards, which either fits with the associations triggered by the merchandising PoS element or do not fit with the brand’s image perceived by shoppers, leading to either conflicting or supporting, cortical relief effects, displayed by an increase neural activity or a decreased neural activity of the dlPFC, respectively. These neuropsychological processes can, therefore, be quantified with the measurement of the neural activity of the dlPFC, using mobile fNIRS. Consequently, the quantified neural activity of the dlPFC, indicating the congruence between the brand’s image and the triggered reward associations of the PoS merchandising element, might, next to the reward association system, be decisive for the prediction of sales at the PoS, acting as an additional process variable, measurable with mobile fNIRS.

4.5.8 Supplement material

Tabelle 4.1 T-values of relevant channels for merchandising element contrasted to the implicit baseline. Signal values differentiated for each channel representing the Brodmann area 9 and 46 and the six merchandising elements

Full size table

4.6 Beitrag 6: Predicting sales of new consumer packaged products with fMRI, behavioral, survey, and market data

Dieser Beitrag ist als Working Paper hier veröffentlicht: Varga, M., Tusche, A., Albuquerque, P., Gier, N. R., Weber, B., Plassmann, H. (2021). Predicting sales of new consumer packaged products with fMRI, behavioral, survey and market data. Marketing Science Institute Working Paper Series 2021, 21-139 https://www.msi.org/wp-content/uploads/2021/11/MSI_Report_21-139.pdf

4.6.1 Abstract

This paper studied the relevance of different types of data for a retailer’s ability to predict sales of new products before their launch. Our approach combined four information sources: (1) in-house observable market data such as price and promotion level, (2) customer attitudes based on a representative survey, (3) incentivized purchasing decisions, and (4) functional magnetic resonance imaging (fMRI) data from a relatively small sample of individuals collected in a laboratory. We used a large German retailer’s weekly sales data to define an estimation data set containing 34 packaged foods and drinks. This estimation data set was used to estimate the parameters of our model. We then used the parameter estimates to predict sales of 17 different products before they were launched. Results indicate that using fMRI data to forecast sales of new products significantly increased forecasting accuracy: It led to a 28.6 % better forecast than a naïve model that considered historic data only, while the model combining all data led to an improvement of 38.6 %. Using our approach, managers can quantify the benefits of collecting different types of data beyond observable market data – including neuroscientific data – to predict the market success of new products.

4.6.2 Introduction

One of the key decisions of marketing managers is whether to launch a new product (Beard & Easingwood, 1996; Biyalogorsky, Boulding, & Staelin, 2006) to maintain or grow market share or to conquer new markets (Hultink, Hart, Robben, & Griffin, 2000). The launch of a new product depends on significant amounts of resources allocated to that initiative before and during the launch (Bhaskaran & Krishnan, 2009). It involves a broad range of firm decisions, such as promotions and pricing on the marketing side, and capacity planning, production, and inventory scheduling on the supply chain side (Cooper, 1979; Petersen, Handfield, & Ragatz, 2005). For example, in 2006 the French company Danone spent over € 10 million introducing the new yogurt brand Essensis, which later failed and was removed from the market (Bruno & Plassmann, 2014). Researchers have estimated that about 40 % of new products fail at launch, even after extensive evaluation, and only one innovation out of ten achieves commercial success (Cooper, 2011; Cooper, Edgett, & Kleinschmidt, 2004). Thus, correctly predicting the success of new products is crucial and of great interest to firms (Cooper, 1979; Cooper & Kleinschmidt, 1987, 1995; Rothwell et al., 1974; Ryans, 1988).

Given this importance, there is a continuous search for new methods and information sources that can improve the accuracy of forecasts of commercial success (Kahn & Chase, 2018). Compelling work in consumer neuroscience has shown that neuroscientific data, measured with e.g. fMRI and EEG, can predict market-level outcomes such as music sales (Berns & Moore, 2012), movie box office sales (Boksem & Smidts, 2015), and advertising elasticities (Venkatraman et al., 2015). These findings suggest that neuroscientific data from a few participants might outperform traditional marketing research measures such as attitudes and preferences (Knutson & Genevsky, 2018).

The main objective of this paper is to investigate the contribution of fMRI data in combination with other data types that marketers typically use to predict sales of new products. With the collaboration of a large German food retailer, we obtained weekly sales and market data (including prices and promotional activities) for 56 different food and beverage items. We also used surveys to collect information about consumer attitudes toward these products from a representative set of customers of the retailer. Finally, we conducted a brain imaging study to collect fMRI data and non-hypothetical, incentive-compatible purchase decisions regarding these products. Here, a small number of customers not representative of the retailer’s customer base were exposed to images of the product and price information and indicated their incentivized willingness to buy. This setup allowed us to measure the impact of each of the four data types we collected (market, survey, incentivized purchase behavior, and fMRI data), on its own or in combination with the other data, on the prediction of sales of new products above and beyond our baseline model, which used average weekly historical sales data.

After data cleaning, our sample of products was divided into an estimation data set (34 products) and a prediction data set (17 products). We assessed the change in forecast accuracy of our models in terms of the mean average percentage error (MAPE). Using regression models^{Footnote 1}, our results show that using fMRI data led to an improvement of 28.6 % in prediction accuracy compared to a naïve model that considered only the average sales of old products to forecast new product performance. When considered in isolation as the only information source, fMRI did better than models that considered the other available data types (i.e., market data, surveys, and incentive-compatible purchase decisions) on their own. When all data were combined, the improvement in prediction accuracy reached 38.6 % compared to the naïve model. Model predictions and additional information about the costs of collecting each data type provide insights into the value of each source of information for the firm. Taken together, our results can assist managers in justifying the acquisition of the different data types to improve forecasts. This is especially important for fMRI data, with which managers are likely to have less experience and thus less knowledge of costs and return on investment.

4.6.3 Literature review

Our paper is related to two streams of past work: (1) the contribution of brain imaging data to predict consumer choices and (2) the prediction of the performance of new products through the use of different types of data. In what follows, we summarize previous work in these two streams.

4.6.3.1 Market-level predictions using brain imaging data

One promise of the nascent field of consumer neuroscience has been to improve predictions about what consumers like and thus decide to buy (Plassmann et al., 2015). Being able to more accurately predict whether consumers will buy a product has important marketing applications for new product development (Ariely & Berns, 2010). Tabelle 4.2 presents an overview.

In a seminal paper, (Knutson et al., 2007) developed an fMRI purchasing task in which participants evaluated the desirability of consumer products, considered whether they were worth the price, and decided to buy or not (see Abbildung 4.24). Brain responses obtained in this task improved the prediction of the sample’s purchase decisions above and beyond self-reported liking of these products, albeit only marginally. The authors identified three brain regions that were predictive of purchasing decisions: (1) the vStr, (2) the vmPFC, and (3) the anterior insula (aI). Evidence on the ability of these brain regions to predict consumer preferences and choices has been replicated and extended across studies and various product categories (Genevsky & Knutson, 2015; Tong et al., 2020; Tusche, Bode, & Haynes, 2010). Our selection of brain regions from which to extract fMRI data was based on this evidence, as detailed in the methods section.

These initial empirical findings showcase the consistency of brain regions involved in purchasing decisions on the level of single individuals. More recent papers (summarized in Tabelle 4.2) have demonstrated the ability of neuroscientific data to predict out-of-sample behavior at the market level – a new method commonly referred to as neuroforecasting (Knutson & Genevsky, 2018). Berns and Moore (2012) provided early evidence in favor of neural predictions of market-level outcomes. They found that brain imaging data from a few music listeners (N = 27) could predict whether a song would become a national hit three years later, as indicated by commercial sales data from Nielsen SoundScan. Data from brain activity in the vStr – obtained using fMRI while subjects listened to music – were successfully used to predict the future sales of those songs, while self-reported liking ratings taken at the fMRI experiment showed no significant correlation with future sales. This study was the first to suggest that brain data from a relatively small sample of individuals could predict commercial sales at the market level better than self-reported liking ratings.

Tabelle 4.2 Summary of neuroforecasting literature

Full size table

A related pioneering fMRI paper asked smokers who intended to quit (N = 30) about their liking and perceived effectiveness of three different anti-smoking campaigns after their brains were scanned while watching them repeatedly (Falk et al., 2012). Neural activity in the vmPFC predicted the overall success of the three campaigns, measured in call volume of the advertised quit hotline. Behavioral rankings from the same participants made less accurate predictions.

The efficacy of brain data for forecasting market-level outcomes extends beyond fMRI data. For instance, several papers demonstrated that brain activity measured using EEG predicted market-level outcomes such as U.S. box office sales (Barnett & Cerf, 2017; Boksem & Smidts, 2015) and TV audience size (Dmochowski et al., 2014), above and beyond self-reported liking and related preference measures. These studies used a greater variety of methodological approaches and metrics to capture people’s brain activity in response to the marketing stimuli, such as different oscillation bands, different components of time-locked EEG signals, and how much participants’ brains had the same reaction (using correlations between participants’ EEG signals). Thus, less consistency exists regarding the type of EEG signal best suited for which type of neuroforecasting exercise (for a review see Hakim & Levy, 2019).

The idea that “brain beats behavior” in predicting market-level success has since been conceptually replicated and generalized across product categories – examples include forecasting the success of microloan appeals (Genevsky & Knutson, 2015), advertising elasticities (Venkatraman et al., 2015), movie sales (Boksem & Smidts, 2015), chocolate sales (Kühn et al., 2016), news article popularity (Scholz et al., 2017), crowdfunding appeal success (Genevsky et al., 2017), and YouTube viewing frequency and duration (Tong et al., 2020) – and also across different brain imaging techniques (for a review see Knutson & Genevsky, 2018).

All these studies compare different data types from a few individuals in a laboratory environment with their brain imaging data (except Venkatraman et al., 2015). To advance the neuroforecasting literature and demonstrate the value of consumer neuroscience for marketing managers and neuromarketing companies, the comparisons need to include richer data sets that companies typically have access to or acquire to predict sales and success. Against this background, this paper investigates whether the combination of different data types can predict sales of newly introduced food and beverage products. These data sources are (1) market data such as price and promotion level that are accessible for retailers and manufacturers, (2) representative surveys asking customers about their attitudes and intentions, (3) incentivized purchasing decisions, and (4) functional magnetic resonance imaging (fMRI) data from a relatively small sample of individuals collected in a laboratory. Abbildung 4.21 gives an overview of the general methodological approach underlying this paper.

4.6.3.2 Combining different data sets to predict the performance of new products

In marketing, work on new product performance began by using data from initial sales of a launched product to predict whether that product was going to be successful in the long run, mostly drawing from repeat-purchase patterns and loyalty rates (Fourt & Woodlock, 1960). Early papers on new product performance prediction reported that sales of fast-moving consumer goods were easier to predict than those of other product categories, due to the repetition of purchase decisions. In parallel, Bass (1969) established that the consumer’s initial purchase decision is a function of the number of previous buyers of the product, and since his seminal work, papers using diffusion models to study the success of new products have become commonplace in the marketing literature (e.g., Chandrasekaran & Tellis, 2017; Fan, Che, & Chen, 2017). Given the focus of our research question, we next discuss a subset of the subsequent literature on prediction of performance of new products, concentrating our attention on papers that examined how different types of data can be used or combined to improve the accuracy of predictions of new product sales.

Given that more data – in terms of both quantity and variety – have become increasingly available and at a faster pace, researchers have made efforts to answer the question of how to combine alternative data types and sources in a managerially relevant way. Kahn (2002) suggested that surveys, expert opinions, and average sales of comparable products are the most widespread techniques for predicting demand of new products, highlighting that these methods are popular due to their interpretability. As Armstrong, Green and Graefe (2015) argued, practitioners should be overly conservative when they do not understand the forecasting procedures. Our aim is to provide a parsimonious method of combining different data, with the intent of investigating which data set or data sets can best improve the prediction of sales of new products.

The objective of combining data is to make use of the advantages of each data type while reducing the disadvantages. Phaneuf, Taylor and Braden (2013) provided a review of how data on revealed preferences and stated preferences have been combined in marketing, transportation, and environmental economics literature with this purpose in mind. While the main advantage (disadvantage) of revealed preferences data is that it is based on real choices (it is historic in nature), the main advantage (disadvantage) of stated preferences data is that it is flexible in scenario creation (it is hypothetical in nature). Morikawa, Ben-Akiva and McFadden (2002) also highlighted this and the fact that the two types of data have complementary characteristics and proposed a methodology to use multiple types of data to estimate discrete choice models. The combination of the different data sets allows for a better prediction of scenarios, such as new product introduction (Phaneuf et al., 2013), that go beyond the scope of the revealed preferences data, in our case previous sales and price data, and consider possible trends or behavioral perspectives from survey participants.

Several papers have tackled similar research questions. In their seminal paper, Rossi, McCulloch and Allenby (1996) combined data on past choices, causal variables (such as price, display, and feature), and demographics to better predict individual price and promotional elasticities, which is essential information for targeting marketing activities. The authors showed that previous choices are very informative about consumer preferences. Urban, Weinberg and Hauser (1996) described how pre-market forecasting can be done for automobiles, using methods with a multimedia virtual-buying environment (an experiment with about 600 participants) to simulate a user experience, combined with tasks where consumers could seek more information about the product, surveys about their purchase intent, and the use of diffusion models and conjoint analysis. The authors quantified the value of each type of data by comparing implementation costs with benefits regarding the final launch decision of the product. We use a similar approach: the collection and implementation of several studies that allow us to obtain data, which is then used to predict the success of new products.

Feit, Beltramo and Feinberg (2010) combined different data sets to better predict market shares of products with different levels of attributes. The authors argued that estimates of the importance of product attributes that rely solely on hypothetical choice experiments (for example, conjoint analysis) frequently show inconsistencies that can and should be corrected through the combination of these data with individual-level purchase data. The authors applied a general framework using Bayesian models and individual-level data to the evaluation of attributes in the U.S. minivan market, predicting holdout purchases better than an approach that excluded individual characteristics and motivations.

The data used in some papers goes beyond the traditional revealed and stated preferences data. For example, Mueller, Osidacz, Francis and Lockshin (2010), in a two-stage approach, applied an online discrete choice experiment combined with product consumption tasks to understand the interplay between sensory (e.g., taste) and product (e.g., packaging) characteristics to predict liking and repurchase intention of Australian red wines. The study was designed in such a way as to integrate the entire purchase process, from the initial choice through the consumption process and the repurchase decision, with the intent of predicting repurchase decisions. The authors found that data on both types of characteristics are important in explaining repurchase decisions, although the findings in terms of the combination of the data seem to have limited suitability to find the drivers of purchase decisions, in part because wine might be too complex a product for consumers to base their repurchase intention on taste (Mueller et al., 2010). Schneider and Gupta (2016) used both numeric and textual data from consumer reviews to predict the sales of existing and new products, using a parsimonious linear regression approach, in a similar way as our proposed approach.

Beyond marketing, other fields such as healthcare have also benefited from similar methods. For example, Harris and Keane (1998) studied elderly consumers’ choice among health plans using attitudinal data and revealed preference data (choices), showing that the combination of these data sets provided more reliable estimates of their preferences for and perceptions of the attributes of choice alternatives. Kappe, Venkataraman and Stremersch (2017) combined historic data on prescriptions and firm detailing efforts with data from subject-matter experts obtained through a conjoint experiment to predict how firms would react to unprecedented marketing policy changes in the pharmaceutical industry.

To summarize, motivated by these papers we collected data from several information sources, estimated a parsimonious model that allowed us to predict sales and do a hold-out prediction evaluation, and conducted a cost-benefit analysis of each type of data, providing insights to managers regarding which studies might be relevant.

We finish the discussion of the literature on prediction by highlighting that there are alternative methods for prediction and measures to evaluate the accuracy of predictions. In terms of modeling approaches and their applicability to forecasting sales of new products, Hardie, Fader and Wisniewski (1998) found that simple models provide significantly better forecasts than complex model specifications. Although there have been recent attempts to predict sales of new products with complex approaches (Chong, Han, & Park, 2017; Kulkarni, Nikumbh, Nikam, Anuradha, & Nikam, 2012), W.-I. Lee, Chen, Chen, Chen and Liu (2012) showed that the simple logistic regression model is often a better choice than the more complex neural network approaches for forecasting the sales of fresh foods. Hence, and in line with other papers that use neuroscience data, we use linear regressions as the main method.

For the measures used to evaluate prediction, we followed Hardie, Fader and Wisniewski (1998) and used the MAPE as evaluation criteria, defined as

Formel 4. Mean absolute percentage error.

$$MAPE = \frac{{\mathop \sum \nolimits_{j = 1}^{J} \mathop \sum \nolimits_{t = 1}^{T} \left| {\left( {Y_{jt} - \hat{Y}_{jt} } \right)/Y_{jt} } \right|}}{JT},$$

where J is the number of products, T is the number of time periods (weeks), $Y_{jt}$ is the value of actual sales per retailer of product j in week t, and $\hat{Y}_{jt}$ is the respective estimated value. In Hardie, Fader and Wisniewski (1998), the authors discussed which measure of prediction accuracy is best suited to product sales forecasting tasks and concluded that MAPE is recommended (see also Makridakis, 1993). Divakar, Ratchford and Shankar (2005) also used MAPE as measure of forecast accuracy in their paper on the practical applications of forecasting models. The authors highlighted that a careful balance between modeling sophistication and practical relevance is key to achieving accurate predictions, with MAPE being one of the easiest measures to understand and interpret. In addition, MAPE has been proven to be very appropriate in planning and budgeting situations (Makridakis, 1993). A number of recent applications have used MAPE, including ArunKumar et al. (2021), Jadhav, Reddy and Gaddi (2017), Kaewtapee, Khetchaturat, Nukreaw, Krutthai and Bunchasak (2021), Prayudani, Hizriadi, Lase and Fatmi (2019) and Wickramasinghe, Weliwatta, Ekanayake and Jayasinghe (2021).

4.6.4 Setting and data description

One of Germany’s largest food retailers provided us with data on 56 products (23 beverages and 33 food items). Product selection by the retailer’s marketing managers ensured representation across 18 product categories (e.g., canned tuna, carbonated soft drinks; three (once six) products each) and sufficient variation in launch dates.

For each product, we observed the average weekly number of units sold per retailer and the number of retailers that decided to carry each product. For products launched before January 2014, this data covered close to six years (until September 2019). For products launched after January 2014, we observed weekly sales and number of adopted retailers since their launch date. Abbildung 4.22 shows the sales evolution of three products in four product categories, as an illustration. It highlights the significant variation in the level of sales, even within each category, suggesting that historical sales data for previously introduced products are limited in their predictive usefulness for the sales of newly launched products. High variance in sales performance characterizes most categories in our data set and partly motivates the retailer’s managers to use multiple data sets to predict sales.

We divided the products into two sets to implement our analysis approach and evaluate the predictive utility in new product sets. Products launched before November 2016 (35 products) were part of an estimation set, while products launched after November 2016 were the prediction or test set (21 products; see Abbildung 4.27 in the additional materials and methods section for products and launch dates). The threshold date was chosen for practical and data analytical reasons, as the retailer introduced several products soon after this date. Moreover, it yielded estimation and prediction sets of a size consistent with standards for cross-validation and out-of-sample predictions in the field (Berrar, 2019). In a way, adopting a threshold date mimics a manager’s challenge to forecast the commercial success of not-yet-launched products, using the information on overall sales of products in the food and drinks categories – and additional data sources at her disposal – at this point.

Besides sales information, four different types of data formed our explanatory variables: (1) market data of all products, including prices and promotional activities; (2) attitudes toward the products obtained using a survey from an online sample representative of the general customer population of the supermarket chain (N = 1451); (3) the incentive-compatible purchase decisions of laboratory student participants while their brain responses were measured using fMRI (N = 44); and (4) their neural correlates of purchasing the different products, obtained in the same fMRI study. A detailed protocol description is available on the Open Science Framework (OSF)^{Footnote 2}. We received ethical approval from the institutional review board of a German university’s medical school. The study was conducted in accordance with the Declaration of Helsinki, and all participants gave informed consent for their participation. In what follows, we describe the four data sources in more detail.

4.6.4.1 Market data

Along with the weekly average sales per location, the retailer provided information about the average price of each product for a given week and the weekly frequency with which the product was on promotion across the retailer’s stores in Germany. We also included a dummy variable for food versus beverages (taking drinks as base) in the estimation model, to control for the different market size of the two types of items. We refer to this set of variables (price, promotions, food category dummy variable) as market data.

Tabelle 4.3 presents the summary statistics of the market variables of the 35 products in our estimation data set (launched before November 2016) and the 21 products in the prediction set (launched after that date). The table describes the average price (in euros), average promotional level, and weekly sales (in thousands) per retailer. We display the mean and the standard deviations. Products in the estimation set have a higher mean price, a lower mean promotional level, and higher mean sales. Due to the selection of new products in the price-sensitive market of food retailing for the prediction set, it is likely that regularly a penetration strategy is applied for these products, introducing them at a low price with additional promotions to gain a fraction of the highly competitive (Kotler & Armstrong, 2004; Spann, Fischer, & Tellis, 2015). This might explain the significant differences between estimation and prediction sets, which challenge the prediction models as the prediction set is in some aspects fundamentally different than the estimation set. A typical product costs about € 6, is not promoted, and has a sales volume of about four units.

Tabelle 4.3 Summary statistics of market variables by product set

Full size table

4.6.4.2 Representative survey

We recruited 1,451 customers of the supermarket chain using the Qualtrics online panel to be representative of the chain’s customer base (see Tabelle 4.10 in the additional materials and methods section for a description). This survey was done in June and July of 2018; the sample size was determined to match sample sizes traditionally used by the retailer when conducting similar surveys.

We note that the survey (and the fMRI experiment – see below) was done at a later stage than the threshold date chosen to define the estimation and prediction sets. We recognize that this might be a concern because consumers may have been familiar with the products chosen to predict sales. However, looking at the responses about product purchase, only 5 % of the respondents indicated previous purchase of the products, and there was very low familiarity with these more recent products (significantly lower than with the products in the estimation set). This is a limitation of our data and not of the approach, driven by the time periods covered by the market data^{Footnote 3}.

Each participant evaluated 12 products across four different product categories, yielding a questionnaire length of about 30 minutes suitable for an online survey. On average, we obtained 311 evaluations per product, varying between 293 and 331. Participants answered several questions about their attitudes toward the products and their personality and socio-demographic status, and completed an instructed attention manipulation check adapted from Oppenheimer, Meyvis and Davidenko (2009), which was used to exclude participants who did not pay attention^{Footnote 4}. The order in which the products were shown to participants was randomized.

The survey included questions about each product’s desirability, measured through four questions about product liking, product attractiveness, packaging attractiveness, and intention to buy the product (translated from a scale by C.-H. Cho, Lee and Tharp (2001). For all of these items, participants’ evaluations were based on a 7-point Likert scale ranging from 1 (“fully agree”) to 7 (“don’t agree at all”). Respondents answered these questions without knowing the product’s price. Given the high positive correlations among these measures, we computed an average of these four variables per individual and used this as a composite measure of the desirability of the product (Cronbach’s alpha = 0.955). Product desirability scores were reversed, so that higher numbers represent more positive product attitudes.

The survey also asked participants about their perception of the product’s success by translating and shortening the success scale from S. Zhang and Schmitt (2001). More specifically, respondents indicated whether they believed that many customers would purchase the product, that it was an enrichment to the category, and that it would have lasting popularity among buyers. They used a 7-point Likert scale ranging from 1 (“fully agree”) to 7 (“don’t agree at all”). We averaged these three indicators to reflect the perceived success of the product in the eyes of survey participants (Cronbach’s alpha = 0.868) and reverse coded it. After the product’s recommended retail price was revealed to the respondents, they also indicated their hypothetical purchase intention if the product was sold at that price.

Tabelle 4.4 presents the summary statistics related to the measures included in the model. Comparing the products in the estimation and prediction sets, we see that products from the prediction set were perceived to be somewhat more desirable and more successful.

Tabelle 4.4 Summary statistics of survey variables by product set

Full size table

4.6.4.3 fMRI experiment

Data from 44 participants was included in the analysis (49.1 % female, average age of 27.2 years; see online appendix for more details). This sample size is in line with previous neuroforecasting studies (37 % higher than that of the average sample size of the papers reviewed in Tabelle 4.2) and with current standards in cognitive neuroscience (Yarkoni, 2009). The study was conducted from June to August of 2018.

The participants were asked to make purchase decisions for each of the 56 products at three different price levels, resulting in 168 purchase decisions (56 products × 3 price levels). Using a theory-driven approach, we included the brain activity (the neural correlate of product desirability and value) in the three brain regions previously found to be involved in purchasing decisions (see Abbildung 4.24): (1) the vmPFC, (2) the vStr, and (3) the bilateral aI. The section additional materials and methods describes in detail the fMRI data acquisition, analyses, and detailed definition of brain regions of interest (ROIs)^{Footnote 5}.

As a sanity check, we also tested whether the data from these brain regions are correlated with product desirability during the product consideration phase and with purchase decision and willingness to pay in the product and price consideration phase. Our analysis replicated previous findings that (1) the vStr and aI encoded product desirability while subjects considered products and (2) the vmPFC and aI encoded the subsequent purchase decisions (“strong no” to “strong yes,” referred to as decision value) for these items (see Abbildung 4.25 and Abbildung 4.26, Tabelle 4.11 and Tabelle 4.12, and supplemental sanity check and supplemental whole-brain analyses section for further details).

Tabelle 4.5 presents relevant summary statistics related to these measures. We observed that activity changes in the aI during product consideration were significantly lower for the products in the prediction set as compared to the estimation set. Finally, we observed that the incentivized purchase decisions by participants in the fMRI experiment were not significantly different (t = 0.794, p = 0.431) for products in the estimation set (M = 2.274, SD = 0.256) and the prediction set (M = 2.329, SD = 0.238).

Tabelle 4.5 Summary statistics of mean fMRI parameter estimates by product set

Full size table

4.6.5 Data modeling and statistical analyses

4.6.5.1 General approach

Our prediction approach has two stages. In the first stage, we estimated the parameters of our model with information from the estimation set, composed of 35 products launched before November 2016. In the second stage, we used the coefficients of the estimated model to predict the sales of products in the prediction set, which were launched after November 2016. This approach simulates the managerial challenge of predicting the success of newly launched products in the marketplace based on existing data. Our approach also allowed us to compare the predictive utility of the four available data types. More precisely, it enabled quantifying the added value of each data source in terms of model fit and, more important, out-of-sample prediction of commercial success of new products.

We modeled the sales per retailer of product j during week t as a flexible function of the covariates contained in the four different data types,

Formel 5. Equation to include four data types.

$$Y_{jt} = f\left( {X_{jt} , S_{j} ,Z_{j} ,W_{j} } \right).$$

(4.1)

In equation 4.1, $X_{jt}$ includes market variables (market price, promotional activities, product type); $S_{j}$ stands for the variables included in the representative survey (average perceived product desirability, product success, and respondents’ intent to purchase at the recommended retail price); $Z_{j}$ consists of the fMRI data (product-specific parameter estimates during the product consideration phase in the vStr and aI, and during the price and product consideration phase in the vmPFC and aI); and $W_{j}$ refers to the incentive-compatible purchase decisions during the purchasing task in the fMRI experiment.

To compare the benefit of including different data sources for sales forecasts, we specified models based on different subsets of the data. For example, a model that uses only market data will take the specification

Formel 6. Equation including market data.

$$Y_{jt} = f_{1} \left( {X_{jt} } \right),$$

(4.2)

while a model in which we augment the market data with data from the representative survey will be

Formel 7. Equation including market and representative survey data.

$$Y_{jt} = f_{2} \left( {X_{jt} ,S_{j} } \right),$$

(4.3)

and so forth for other combinations of the several data sets. This approach enables us to investigate which combination of data type optimizes the prediction of sales and to quantify the benefit of adding other types of data.

4.6.5.2 Empirical specification

In our analysis, we use as a dependent variable the average weekly unit sales of the product per retailer that has decided to sell the product on its shelves ($Y_{jt}$). Although we could have instead used the overall sales of a product, we decided against it because the volume of sales depends on both consumers’ demand and retailers’ decision to carry. Given that we do not have information about retailer characteristics or about the decision process retailers go through to adopt a new product, we decided to focus our analysis on explaining and predicting consumer demand, conditional on the retailer offering the product.

Our approach to estimating the parameters is the ordinary least squares (OLS) method, which is simple, widely used, and easy for managers to understand, with a linear form for the $f\left( . \right)$ function in Formel 3. Given that we have a limited number of products, divided into estimation and prediction sets, more complex models that allow for interactions between variables are likely to increase overfitting, as the degrees of freedom go down with more explanatory variables (Hawkins, 2004)^{Footnote 6}.

4.6.6 Data details for estimation

First, to estimate the model, we used weekly data from the 60 weeks before the threshold launch date, from October 2015 to November 2016. To test the model’s predictive ability, we predicted sales per retailer for 60 weeks after the launch date, from November 2016 to January 2018. Given that four products were launched after January 2018, our data set dropped to 52 products, for which we have a total of 2,407 observations.

Second, for each product, we excluded the first eight weeks immediately after product launch, as these weeks are typically marked mostly by stocking up and placement decisions by the retailers, leading to more variation in sales not related to the overall performance of the product. This left us with 2,247 observations and 51 products, as one product was launched in November 2016 (so this filter eliminated it from the estimation set).

Third, we excluded outlier observations, defined as time periods when promotional activity was above the 95th percentile across observations in the estimation set. These are periods when managers likely combined the promotional activity with unobserved-to-the-researcher activities that supported sales. Hence, these extreme cases can influence predictive outcomes, although in our case the results are not substantively different if they are included.

After applying these three criteria, the estimation set consisted of 34 products and 1,600 observations, and the prediction set included 17 products and 505 observations.

4.6.7 Results

We first describe the fit and predictive accuracy of the OLS model, using different sets of data. We then discuss the coefficients of the explanatory variables for the subset of best models.

4.6.7.1 Model fit and predictive accuracy

Following the approach outlined in the previous section, we estimated several models using all possible combinations of the four sets of data and collected as fit measures the adjusted R-square and the in-sample MAPE. To evaluate predictive accuracy, we computed the MAPE for the out-of-sample prediction values, using the estimated OLS coefficients and the data available for products launched after November 2016.

Tabelle 4.6 shows the estimated and predicted errors, using all possible combinations of data types (smaller numbers – i.e., smaller errors – represent better outcomes). We group the models based on the amount of data available, from models that use a single data set to the full model, which uses the four different types of data. Besides the MAPE, we also compute a measure of how much the fit and predictive accuracy changes, shown as a percentage, when compared to a baseline model in which only the constant is included, defined as

Formel 8. Accuracy measure.

$$\left( { - \left( {{\raise0.7ex\hbox{${MAPE_{model} }$} \!\mathord{\left/ {\vphantom {{MAPE_{model} } {MAPE_{baseline} }}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${MAPE_{baseline} }$}} - 1} \right)} \right)$$

The baseline model represents the manager’s best guess on the performance of newly launched products, based solely on the average of sales per retailer of previously launched products, and assuming no access to any additional data. We observe an in-sample MAPE of 0.72 and an out-of-sample MAPE of 0.84 for the baseline model.

As one might expect, more data is better in terms of the estimated in-sample performance, with the in-sample MAPE improving in all cases with additional data. Looking at the models estimated with a single data set, we observe that market and fMRI data sets provide the best increments in in-sample accuracy. For instance, adding market data (fMRI data) to a constant-only model reduced the in-sample MAPE to 0.52 (0.62), an improvement of 28.3 % (14.6 %).

The combination of these two data sets leads to a large reduction in the in-sample MAPE, to a value of 0.50, an improvement of 30.6 %. Interestingly, the combination of fMRI and survey data also leads to a similar improvement of the in-sample MAPE. However, it seems that this combination leads to an overfitting of the model, as in out-of-sample, the respective MAPE is worse than in the baseline model. Finally, in terms of in-sample results, when looking at the cases when the three and four types of data are combined, the values again show in-sample improvements above the 30 % mark, with the full model using all data reaching the value of 34 %.

Looking at the out-of-sample prediction results, and starting with the single-data models, we find that the fMRI data alone performs better than other single-data models – and even better than some of the models that use two or three data sources – with a prediction MAPE of 0.60, an improvement of 28.6 % compared to the baseline model. This finding suggests that the fMRI data from a few participants were powerful predictors of the sales of not-yet-launched products at the market level. This finding is even more notable given that participants in the fMRI study were not representative of the retailer’s customer base (a student convenience sample).

When combining two data types, the fMRI and market data provide the best improvements in terms of out-of-sample MAPE, with a value of 0.55, an improvement of 34.1 %. This result provides evidence that capturing additional information directly related to the product and marketing decisions – in this case the type of product, price, and promotional activities – can complement the data variation captured by the fMRI experiment and account for additional in-store elements that are at least in part under the control of the manager or the retailer, and that are relevant to explain sales of new products.

Tabelle 4.6 Fit and prediction accuracy

Full size table

It is important to note that some models overfit the data when additional data is included. This overfitting seems driven mainly by the survey data variables. For instance, adding the survey data variables to a model combining them with one other data type, including the incentivized purchase decisions and fMRI data variables, leads to poorer predictions and increases the MAPE in most cases. One possible explanation is the lack of an incentive when respondents answer the survey in a laboratory and/or their lack of experience with the category. Similar results in which stated or liking preferences do not match with market outcomes have been found in previous studies (e.g., Kühn et al., 2016; Phaneuf et al., 2013).

Overall, combining all data sets maximized the forecast accuracy, leading to a MAPE of 0.51, an improvement of close to 38.6 % over the baseline model. For this specification, Abbildung 4.23 illustrates the match of predicted and actual sales per retailer. Overall, sales were well captured by this combination of data sources and align better than in the naïve model. The dotted line represents the intercept of that model (i.e., the average sales per retailer of products used in the estimation data set), with an intercept of 4.03 and a standard error of 0.49. The “missed” predictions on the top of the figure are all from one product that did extremely (and, according to retailer managers, unexpectedly) well in the market.

4.6.7.2 Parameter estimates

Tabelle 4.7 presents the coefficients from the OLS models for several combinations of data types, obtained using the estimation data set. These coefficients measure the marginal effect of variables on the dependent variable (sales per retailer in a given week) using solely the data until November 2016 for estimation. The first four columns present the specifications using only one data source for the estimation, with each data set considered separately. In columns 5 and 6, we show the results of the best models (in terms of MAPE prediction) for the combination of two and three data types. The last column shows coefficient estimates using all available data types.

The estimates were consistent overall across models. For the incentivized purchases collected during the fMRI task, we observed a significant positive coefficient when the data was used alone. This finding suggests that incentivized purchase decisions observed in a small (nonrepresentative) sample in a well-controlled laboratory context reveal relevant information about product sales on the market level. The significance goes away when other variables are included in the model.

The market data variables show effects in line with commonly held notions in the field: The positive coefficients for the promotion level suggest that a more heavily promoted product is likely to have more sales. The significant negative coefficients for average price indicate that a more expensive product is likely to have lower sales. These findings suggest that, to some degree, the store manager has control over the success of new products, using different levels of marketing mix variables.

Tabelle 4.7 Parameter estimates of selected combinations of data types within the estimation set

Full size table

The survey data variables did not show a significant relationship with sales in the estimation set in our estimation model, suggesting that consumer attitudes, liking responses, and hypothetical purchase intentions are less powerful to predict the sales of new products.

Finally, we observed that the fMRI data were a significant predictor of sales, except for estimated brain data in the vmPFC. This finding is in line with recent results that while activity in the vmPFC predicts purchases within the same individual (Chib, Rangel, Shimojo, & O’Doherty, 2009; Knutson et al., 2007; Litt, Plassmann, Shiv, & Rangel, 2011; Tusche et al., 2010), it is likely less suited to predict other people’s purchases (Genevsky, Tong, & Knutson, n.d.).

The vStr during product consideration correlated negatively with the product’s sales per retailer. This is notable because the vStr has positively correlated with purchases and product desirability within the same individual in previous literature (Knutson et al., 2007) as well as in our data (see Tabelle 4.12 in the additional materials and methods section). It has also consistently shown a positive coefficient in similar in-sample regressions in the neuroforecasting literature that directly investigate not purchasing of packaged goods, but other related behaviors such as sales of songs in the U.S. charts (Berns & Moore, 2012), advertising elasticities (Venkatraman et al., 2015), and promotional sales after ad exposure (Kühn et al., 2016). However, if we consider vStr activation during the product and price consideration phase instead of the product consideration phase, as done in the previous literature (Genevsky & Knutson, 2015; Genevsky et al., in press, 2017; Knutson et al., 2007), we do find that the vStr during this phase is a significantly positive predictor (see Tabelle 4.13), and including the vStr during this phase in the prediction model does not substantially change our prediction results (see Tabelle 4.14).

The aI activation has a significantly positive relationship with per-retailer sales, during both the product phase and the product and price consideration phase. This brain region has been repeatedly linked with consumer choices. Yet evidence regarding the directionality of the effect is mixed (Knutson et al., 2007; Tusche et al., 2010). The neuroforecasting literature has generally paid less attention to this brain region. Notable exceptions are studies outside the consumer domain that predict microlending rates (Genevsky & Knutson, 2015) and crowdfunding outcomes (Genevsky et al., 2017), which found a negative coefficient for the aI in their regression analyses. Our results indicate that the aI might play a more important role for predicting sales than previous papers have suggested.

4.6.7.3 Robustness checks

We tested whether our findings regarding sales forecasts were robust to alternative model specifications. To this end, we performed four robustness checks. First, we moved the threshold date – which assigned products to the estimation and prediction sets – to four weeks later. Second, we moved the threshold date to four weeks earlier, which led to two more (fewer) products in the prediction (estimation) set. With these two checks, we tested whether our results are robust to the chosen timing for the prediction exercise. Third, we excluded from the estimation set a widely popular and well-known product, which can be considered an outlier in sales per retailer. Fourth, we excluded the two products with the lowest sales per retailer from the estimation set. This approach allowed us to test whether possibly niche items caused a bias in the forecasts and drove our main results.

Tabelle 4.8 shows the results across the four robustness checks, across all possible combinations of data sources. Across the four sets of robustness checks, the model combining all data continues to provide the best out-of-sample predictions, matching findings obtained using our main specification. We highlight two considerations: First, the MAPE reduces significantly across specifications when the threshold date is moved ahead by four weeks. This is driven mostly by the fact that there are now fewer products in the prediction set, with less variation in sales per retailer, which leads to a more accurate prediction. Second, we note that the MAPE for the full data specification is no longer the best when the best product is removed. This is justified mostly by the fact that the survey data becomes worse at prediction without that product. In other words, the accuracy of survey respondents is better when that product is considered, most likely because it is a product informative about the popularity of new products. Overall, however, our robustness checks demonstrate that our main findings regarding out-of-sample predictions of sales are robust to a variety of alternative specifications.

Tabelle 4.8 Robustness checks for sales forecast

Full size table

4.6.8 Managerial implications

We evaluated the impact of the different data sets on the profits of a retailer by estimating benefits from using the different data to obtain better predictions, and by obtaining estimates of the costs of acquiring the data. We evaluated the magnitude of the data value using the 17 products that were kept in the prediction set.

To ascertain the costs of acquiring the survey data, we reached out to three suppliers of survey services and were quoted € 10,000, € 12,000, and € 22,000 for surveys of similar sample sizes as the ones used in our study, in terms of both number of products and participants. For the fMRI study, which also includes the incentivized-purchase task, we obtained quotes of € 29,000 and € 35,000, again with similar conditions to our study. We assume that the market data is free of charge, as the retailer must keep records of prices and promotions, and they know the type of products sold.

To evaluate the benefits of each data set, we computed back-of-envelope values based on the difference in predictions, with and without the different data sets. Our objective was to have estimates of the benefits of having more accurate sales predictions per retailer, which could translate into an increase in revenue because of reductions in the cost of stock-outs or the cost of holding excess stock^{Footnote 7}. For simplicity, we describe the approach when the prediction of sales with additional data reduces the overestimation of sales and assume that the overall value of margins of lost sales (when the prediction is underestimated) is similar. In practice, these values can be different, and a category manager should know both of them (i.e., the cost of holding excess inventory and the loss in margins from stock-outs).

To ascertain the cost of holding inventory, we obtained information from financial statements of retailers and industry reports: The cost of goods is 75 % of the retailer price of the products, similar to other retailers in Germany^{Footnote 8}, while the cost of holding inventory is assumed to be 20 % of the value of the inventory^{Footnote 9}. With the information about product retail price, this allowed us to compute the cost of holding inventory, per unit of product, which multiplied by the unit sales per week of each of the products in the prediction set gave us the weekly holding costs. Based on discussions with managers, we assumed that the impact of a different prediction of sales lasts for τ weeks, and after that, the retailer can observe the actual level of sales and correct the inventory levels, no matter what the initial prediction of sales was. We tested τ = 4 and τ = 8.

Given that we obtained the out-of-sample MAPE from our estimations (i.e., the value of the prediction error incurred based on different data sets, measured as a percentage of sales), we could quantify the value of each data set, multiplying the difference in the MAPE of the baseline model and each data MAPE by the weekly sales, valued at the cost of goods. That gave us the difference in the value of excess held between two the predictions. We then multiplied that value by the average weekly holding cost and the number of weeks we assumed to be necessary for the manager to adjust the level of inventory based on actual sales. Hence, the benefit a data set provides to the firm, compared to the naïve model, is given by:

Formel 9. Equation to calculate benefit of data set.

$$\begin{aligned} Value_{Data\ Set} = \ & \left( {MAPE_{0} - MAPE_{Data\ Set} } \right) \\ & \times Average\ Weekly\ Sales \left( {valued\ at\ COGS} \right) \\ & \times Weekly\ Holding\ Cost \times \tau\ weeks\ adjustment. \\ \end{aligned}$$

The results for each single-data model and the full model are presented in Tabelle 4.9, with values aggregated across all 17 products used in the out-of-sample prediction set. The average weekly sales of these products were € 188,691, which leads to a holding cost per week of € 28,304, given our aforementioned assumptions.

Our estimates show that the fMRI is the most valuable data set to collect, given that it is the one that provides the best increase, on its own, in the out-of-sample MAPE. The overall benefits can range from € 27,171 to € 54,343, depending on the speed of adjustment to the sales. This justifies its costs of about € 30,000 in most situations, considering that in our application, the incentive-compatible purchase decisions were also part of the fMRI and especially because we limited the benefits to only 17 products. The benefits of collecting the data scale linearly with the number of products and with the number of weeks needed to adjust the inventory to market conditions. A full data set that combines all data types would lead to benefits of € 36,000 to € 74,000, about 5 % of sales. Overall, these results highlight the advantage of collecting fMRI data sets to improve the prediction of sales of new products.

Tabelle 4.9 Evaluation of each data set

Full size table

Our work also has important implications for neuromarketing vendors. Most of the companies that are offering neuromarketing services are using different EEG-based metrics to tweak ads, making them shorter and thus saving media expenses for their clients (for a list of these companies and their services, see Plassmann & Ling, 2020). Our results indicate that another business opportunity for such companies is to offer neuroforecasting services using approaches such as the one described here to help their clients drive revenue in addition to saving costs.

4.6.9 Conclusion

In this paper, we studied the added value of different data types to the forecasting accuracy of market-level sales of new products. Using data provided by a large German retailer on more recently launched grocery products and on similar products that were previously available on the market, we estimated the contribution of market data (price, promotions, and product type), representative surveys (purchase intention, perceived desirability, and success of the products), fMRI data (in three brain regions involved in purchase decisions: vStr, aI, vmPFC), and incentive-compatible purchase decisions to improving forecast accuracy.

We estimated a regression model and used its estimated coefficients to predict the success of new products. We used the weekly average number of units sold by a retailer as the dependent variable of the regression analysis. Our approach mimics the managerial challenge of obtaining a forecast of not-yet-launched products at a given point in time. We found that using fMRI data to predict the sales of new products significantly increased forecast accuracy. Using only fMRI data, we reduced the prediction error by close to 29 % compared to a naïve model (i.e., a model using the average historical sales of previously launched items as an intercept). Such improvement was not possible with any other data type. In addition, we found that although all data types can improve predictions, some are worth more than others, and given the acquisition cost of data, it is likely that in practice some data is not worth collecting or buying.

We performed various robustness checks, in terms of both data and method of prediction. Through these supplemental tests, we confirmed that brain data of a small number of participants are indeed a robust predictor of sales in the marketplace. In fact, we find that fMRI data are better predictors than traditional customer surveys.

Our findings also contribute to academic research on the predictive utility of fMRI data in a variety of cases and settings (Boksem & Smidts, 2015; Genevsky & Knutson, 2015; Genevsky et al., 2017; Kühn et al., 2016; Scholz et al., 2017; Venkatraman et al., 2015). We extend prior work in at least three important ways: First, we predicted real-world sales of new products. Second, we integrated product attitudes of a large representative customer sample, market variables, and historical sales data on related products, allowing for a comparison with other information sources that marketers would typically use. And third, we have given managers and researchers an indication of the monetary added value of collecting fMRI data. Taken together, our paper has important novel implications for both marketing research and practice.

4.6.10 Additional materials and methods: fMRI experiment

4.6.10.1 Participants

We recruited 53 healthy, right-handed participants using the participant pool of a German university and standard fMRI inclusion criteria. They received a total of € 50 for their participation. Of this amount, € 40 was used as a budget to spend (or not) on the purchasing task. Participants were told that the goal of the experiment was to study the neural correlates of consumer decision-making. We excluded nine participants from the fMRI sample due to excessive head movement beyond 3 mm/degrees during scanning. Thus, a total of 44 participants were included in the analyses (52.3 % female, aged 20–39, M = 27.27 years, SD = 4.86 years; see Tabelle 4.10 for a sample description).

4.6.10.2 Procedure

The experiment consisted of three parts. Part one involved a computerized task that took place outside of the fMRI scanner. Participants completed a valuation task to determine their willingness-to-pay (WTP) for the 56 products using a Becker-deGroot-Marschak (BDM) second price auction mechanism (Becker, DeGroot, & Marschak, 1964). In part two, participants completed an incentive-compatible purchasing task while their brain activation was measured using fMRI. In part three, participants went through the same questionnaire as the participants from the representative survey sample did, except that they evaluated all 56 products in randomized order (outside of the fMRI scanner).

Tabelle 4.10 Descriptive statistics of socio-demographic characteristics

Full size table

The main fMRI purchasing task was adapted from the SHOP task from Knutson et al. (2007) and displayed using the Scenario Designer software (for the timing and procedure of a sample trial, see Abbildung 4.24). In each of the 168 trials, participants were presented with an image of a product (4 s; product consideration phase), followed by a fixation cross (1–5 s), the presentation of the price together with the product (4 s, product & price consideration and decision phase), and the response screen (2 s, response phase). Inter-trial intervals (fixation cross) varied from 2 to 6 s. Participants were instructed to make their purchase decisions during the product & price consideration phase. They indicated their purchase decisions on a 4-point scale by pressing the respective button on an MRI-compatible response box during the response phase. The mapping of the purchase decision and button press was consistent across participants (strong yes = left index finger, weak yes = left thumb, weak no = right thumb, strong no = right index finger). All participants underwent a training phase ensure that they understood the meaning of the response buttons.

All 56 products were presented three times across the three runs. Each run included every product once, shown at one of three prices levels. The price levels varied as follows: In every run, one-third of the products were offered for the actual recommended retail price, one-third for a price that was marked up by 20 % of the participant’s WTP, and one-third for a 20 % discount of the participant’s individual WTP. This was done to ensure enough variation in the purchasing decision variable since in the original study by Knutson et al. (2007) many nonpurchase trials had to be removed even though the original retail value of the products was discounted by 75 %. The mapping of a product to the three price levels was pseudo-randomized across runs. The order of products varied across functional runs. Together, participants made 168 purchase decisions in the fMRI task (56 products × 3 presentations at a different price level each).

At the end of the study, one decision from either the BDM auction or the fMRI purchasing task was implemented by the computer. The total time of the fMRI experiment, including preparation and debriefing time, was two hours.

4.6.10.3 fMRI data acquisition

Gradient echo T2*-weighted echo-planar (EPI) images with BOLD contrast were acquired using a 3-Tesla Magnetom Trio scanner (Siemens, Erlangen, Germany) and an eight-channel head coil. Thirty-seven slices were scanned in ascending inter-leaved order, each 3 mm thick with an interslice gap of 0.3 mm (voxel size: 2 × 2 × 3 mm). The flip angle was 90. Other imaging parameters were 2.5 s repetition time (TR) and 45 ms echo time (TE). We also acquired whole-brain high-resolution T1-weighted structural scans using an MP-RAGE sequence resulting in 160 slices (voxel size: 1 × 1 × 1 mm; TR = 1.3 s, TE = 3.97 ms) to permit anatomical localization of the functional activations at the individual level. Diffusion-weighted imaging data was acquired immediately following the acquisition of T1-weighted structural images for purposes not relevant to this paper.

4.6.10.4 fMRI data processing

Functional images were analyzed using the statistical parametric mapping software SPM12^{Footnote 10} implemented in MATLAB. Before statistical analysis, functional imaging data were subjected to the following preprocessing steps: (1) slice-timing correction was applied; (2) the realign procedure was used to perform motion correction; (3) the participants’ T1 structural volume was co-registered to the mean of the corrected EPI volumes; (4) the group-wise DARTEL registration method included in SPM12 was used to normalize the T1 structural volume to a common group-specific space, with subsequent affine registration to Montreal Neurological Institute (MNI) space; (5) all EPI volumes were normalized to MNI space using the deformation flow fields generated in the previous step, which simultaneously resampled volumes to 2 mm isotropic, (6) and the EPI volumes were smoothed using a Gaussian kernel of 6 mm isotropic, full width at half maximum (FWHM).

4.6.10.5 fMRI data analyses

For each participant, a GLM estimated regressors of interest for each of the 168 trials in the fMRI task (56 products × 3 presentations), separately for each phase of the purchasing task (product consideration, product & price consideration and decision phase). The trial-specific regressors of interest of a particular task phase served as input for the theory-driven, region-of-interest (ROI) analyses (see details below). Trials in the product and product & price consideration phase were defined by the onset and offset of the relevant information on-screen (i.e., product presentation and price display, respectively; Abbildung 4.24). Trials in the response phase were defined by the onset of the decision prompt and participants’ execution of a purchase decision (button press) in the trial. Note that we cannot reliably distinguish BOLD responses in the response phase from the previous price phase (due to the lack of variable inter-stimulus intervals between both phases and the “sluggishness” of the BOLD response). The GLMs included as covariates of no interest the six motion parameters estimated from image realignment. Neural activation was modeled by distinct regressors convolved with a canonic hemodynamic response function (hrf). A 128s high-pass cutoff filter was applied to eliminate low-frequency drifts in the data.

4.6.10.6 Selection of ROIs

We extracted data of three a-priori defined regions of interest (ROIs). The ROIs of the bilateral aI and vStr were created using the Desai atlas in AFNI^{Footnote 11}. The vmPFC ROI was based on Neurosynth, a platform for large-scale, automated synthesis of fMRI data. Thus, ROIs were independently defined with regard to our key analyses – the neural prediction of market level success of our 56 products – and with regard to our subject sample, reducing the risk of producing false positive results and of circular analysis (i.e. double dipping; Kriegeskorte, Simmons, Bellgowan, & Baker, 2009). The masks we used for all three ROIs are available on OSF^{Footnote 12}.

ROI-specific activation was calculated by averaging across estimated regressor values of all voxels with the specified mask (separately for each of the three brain regions). For each ROI, we extracted and averaged product-specific data across the three regressors estimated for each product (per task phase), corresponding to the three product presentations in the fMRI purchase task. Data were extracted from two task phases: 1) the product consideration period and 2) the product & price consideration period during the purchasing decision for a particular product. In line with the prior literature, we extracted from the vStr (Knutson et al., 2007) and aI (Tusche et al., 2010) mask during the first period and from the aI (Genevsky & Knutson, 2015; Knutson et al., 2007; Litt et al., 2011) and vmPFC (Genevsky et al., 2017; Hare, O’doherty, Camerer, Schultz, & Rangel, 2008; Knutson et al., 2007; Litt et al., 2011) during the second period yielding four values for each of the 56 products for every participant.

4.6.10.7 Sanity check and supplement whole brain analyses

We performed several post hoc analyses at the whole-brain level to further validate the selection of our regions of interest. We aimed to identify brain areas in which measured BOLD signals are systematically modulated by the participants’ purchase decisions and perceived product desirability.

To this end, for each participant, we estimated additional GLMs (separately for our behavioral variables of interest listed above). Below, we describe the GLMs using the example of participants’ purchase decision value (DV) on each trial. DVs are based on participants’ button presses on each trial and coded so that higher values represent a positive purchase decision (1 = strong no, 2 = weak no, 3 = weak yes, 4 = strong yes). During the product & price consideration phase, participants had access to all the information necessary to make a purchase decision (i.e., product & price information) and were instructed to decide whether or not they would want to purchase the product for real. Thus, we hypothesized that DVs are encoded in the brain during the product & price consideration phase of the task (i.e., before the subsequent response phase; see Abbildung 4.24). To test this idea, for each participant, we estimated a GLM with the following regressors:

(R1):: a boxcar function for the product consideration phase on all trials (duration = 4 s);
(R2):: R1 modulated by the subject’s stated DV on each trial;
(R3):: a boxcar function for the product & price consideration and decision phase on all trials (duration = 4 s);
(R4):: R3 modulated by the participant’s stated DV on each trial;
(R5):: a boxcar function for the response period on all trials (duration = reaction time);
(R6):: R5 modulated by the participant’s stated DV on each trial;
(R7–R9):: A boxcar function specifying missed trials, separately for each choice period (durations of 4 s for product & price consideration phase, respectively; duration of the response period = reaction time on that trial); and
(R10–R15):: regressors of non-interest included six motion regressors as well as a session constant.

To examine whether brain responses obtained in the product & price consideration phase are modulated by the DVs, the regressor of interest (R4) was contrasted against an implicit baseline. Subject-specific contrast images were then used in a one-sample t-test at the group level (as implemented in SPM).

We found that the bilateral vmPFC and the left aI were positively correlated with participants’ purchase decision values during the price phase (p < 0.001, family-wise error, FWE, corrected at p < 0.05 at the cluster level; Abbildung 4.25, Tabelle 4.11). The right aI showed a similar response profile at p < 0.001 (uncorrected, whole-brain) but did not survive statistical correction for multiple comparisons (Abbildung 4.25).

Tabelle 4.11 Neural correlates of participants‘ decision value (strong no to strong yes) during the product & price consideration and decision phase of the fMRI purchase task

Full size table

Next, we examined whether participants’ perceived product desirability modulated brain responses during the product consideration phase. To this end, for each participant, we estimated a GLM that was like the one described above, with one exception: DV values in R2, R4, and R6 were replaced with participants’ stated desirability of the product shown on a trial. Product desirability scores were based on participants’ survey responses completed after the fMRI purchase task. Participants rated liking, attractiveness, hypothetical purchase intention, and package liking (all on a scale from 1 = very much to 7 = not at all). We reversed the directionality of the scales such that higher values represented more positive product attitudes. We then integrated these positively correlated product attitudes into one averaged desirability index. To test whether neural signals during product consideration are modulated by products’ perceived desirability, we contrasted R2 against an implicit baseline (for each participant) and subjected these contrasts to a one-sample t-test at the group level.

We found that the left vStr and the right aI positively covaried with individuals’ perceived product desirability (p < 0.001, FWE corrected at p < 0.05 at the cluster level; Abbildung 4.26, Tabelle 4.12). The same was true for the right vStr and the left aI at a slightly more lenient threshold (p < 0.001, uncorrected). Overall, these supplemental analyses provide strong support for the functional role of our a priori regions of interest (vStr, aI, vmPFC) during the choice-relevant periods in the fMRI purchasing task.

Tabelle 4.12 Neural correlates of the perceived desirability of consumer items during product consideration phase in the fMRI purchase task

Full size table

4.6.10.8 Supporting online appendix

Tabelle 4.13 Parameter estimates of selected combinations of data types within the estimation set when adding vStr during product & price consideration phase

Full size table

Tabelle 4.14 Fit and prediction accuracy when adding vStr during product & price consideration phase

Full size table

Notes

1.
Der Beitrag entspricht der folgenden Publikation: Gier, N., Krampe, C., Reisch, L. & Kenning, P. (2018). Zur Konzeption eines Verbraucherinformationssystems als Ergänzung -oder Alternative? – zum klassischen Informationslabel. Journal of Consumer Protection and Food Safety, 13(2), 183–189. https://doi.org/10.1007/s00003-017-1144-7
2.
Dieser Beitrag wurde bei der „Internationalen Tagung Wirtschaftsinformatik“ 2019 im Rahmen eines Vortrags im Workshop „Digital Consumption“ vorgestellt und ist als „Research in Progress“ veröffentlicht: Gier, N. R., Krampe, C., Reisch, L. und Kenning, P. (2019) „Besser statt mehr! Vom Daten-DIY zur ‚Verbraucherinformatik'. International Conference on Wirtschaftsinformatik, Siegen, Germany, https://www.verbraucherinformatik.de/2019/10/20/beitraege-vom-workshop-digital-consumption-auf-der-wi-2019/
3.
We also implemented random forest models, with substantively similar results. We discuss them in the results section.
4.
osf.io/6du3r/?view_only=f130aa7005af42bfa86ea424f2a03069
5.
We note that we could have included more dummy variables to capture the differences in product categories within food and drinks. However, with only 35 products in the estimation set, we found that that model overfitted the data.
6.
The questionnaire is available on OSF: osf.io/6du3r/?view_only=f130aa7005af42bfa86ea424f2a03069. There are more details in the additional materials and methods section.
7.
It is important to note that these three brain regions are also involved in other mental processes unrelated to purchasing; most brain regions are involved in more than one function (Poldrack, 2011). However, given the meta-analysis on value coding (www.sas.upenn.edu/~mcguirej/meta-analysis.html) and emotional intensity coding (www.neurosynth.org) and the sanity check described in the additional materials and methods section (Abbildung 4.24 and Abbildung 4.25, Tabelle 4.11 and Tabelle 4.12), we are confident that our three regions of interest are indeed involved in the formation of purchase decisions in our experiment.
8.
We also estimated our model with a random forest approach. This approach benefits from allowing for interactions between the different terms, and so it is more appropriate when several data sets are included in the estimation. In the full model, we found that the OLS approach produced better predictions, in terms of lower MAPE, than the random forest, and hence we decided to present here only the more parsimonious approach. In terms of importance of the fMRI data for prediction in the random forest estimation, similar results were obtained.
9.
Another possibility would be to consider not launching some of the products, the ones for which predicted sales were too low, but given that we do not have data on the launch expenses, we did not attempt this computation.
10.
For example, the Rewe Group shows 76.4 % of cost of goods sold for 2020 (see www.rewe-group-geschaeftsbericht.de/fileadmin/media/pdf/RZF_FinancialStatements_20201231.pdf).
11.
See, for example, www.retailowner.com/Inventory/Costs-of-Excess-Inventory.
12.
www.fil.ion.ucl.ac.uk/spm
13.
afni.nimh.nih.gov/AFNIAtlases
14.
osf.io/6du3r/?view_only=f130aa7005af42bfa86ea424f2a03069

Author information

Authors and Affiliations

Heinrich-Heine-Universität, Düsseldorf, Deutschland
Nadine R. Gier

Authors

Nadine R. Gier
View author publications
You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Nadine R. Gier .

Rights and permissions

Open Access Dieses Kapitel wird unter der Creative Commons Namensnennung 4.0 International Lizenz (http://creativecommons.org/licenses/by/4.0/deed.de) veröffentlicht, welche die Nutzung, Vervielfältigung, Bearbeitung, Verbreitung und Wiedergabe in jeglichem Medium und Format erlaubt, sofern Sie den/die ursprünglichen Autor(en) und die Quelle ordnungsgemäß nennen, einen Link zur Creative Commons Lizenz beifügen und angeben, ob Änderungen vorgenommen wurden.

Die in diesem Kapitel enthaltenen Bilder und sonstiges Drittmaterial unterliegen ebenfalls der genannten Creative Commons Lizenz, sofern sich aus der Abbildungslegende nichts anderes ergibt. Sofern das betreffende Material nicht unter der genannten Creative Commons Lizenz steht und die betreffende Handlung nicht nach gesetzlichen Vorschriften erlaubt ist, ist für die oben aufgeführten Weiterverwendungen des Materials die Einwilligung des jeweiligen Rechteinhabers einzuholen.

Reprints and permissions

Copyright information

About this chapter

Cite this chapter

Gier, N.R. (2022). Ausgewählte Beiträge. In: Consumer Decision Neuroscience. Springer Gabler, Wiesbaden. https://doi.org/10.1007/978-3-658-38206-3_4

Download citation

DOI: https://doi.org/10.1007/978-3-658-38206-3_4
Published: 19 August 2022
Publisher Name: Springer Gabler, Wiesbaden
Print ISBN: 978-3-658-38205-6
Online ISBN: 978-3-658-38206-3
eBook Packages: Business and Economics (German Language)

Publish with us

Policies and ethics