1 Introduction

Negative partisanship (or Negative Party Identification - NPIFootnote 1) has been on the rise in the last decades. Researchers have observed this increase not only in the US but also in many other democracies. This somewhat new scholarly attention is evident in several studies that demonstrated the importance of negative partisanship in general (Abramowitz and Webster 2016; Webster 2018; Iyengar et al. 2019; Ridge 2020), and, more specifically, on political attitudes, democratic dissatisfaction, and behavior in the US and worldwide (e.g., Medeiros and Noël 2014; Abramowitz and Webster 2016; Mayer 2017; Bankert 2021). One major contribution of this new engagement with the topic consisted in establishing that most people hold negative as well as positive identifications. Yet, it is still possible that a person might experience only one of them (either the positive or the negative partisanship). In other words, they can occur separately (e.g., Ridge 2020).

The fact that the concept of negative partisanship has become a highly salient topic helped improving the discussion about it. However, we argue that its current measurements need further improvement as they mostly focus on the level of attitudes. Yet negative partisanship, as well as its positive counterpart, is an identity, and, as already noticed by Bankert (2021), should therefore be measured as such. As a matter of fact, the idea that Positive Party Identification (PPI) is part of one’s social identity is rather uncontested, and it has indeed always been measured as such, with items referring to the own self-concept and definition (see for an overview Johnston 2006). Nevertheless, when it comes to NPI, most measures are operationalized at the level of partisan attitudes instead of engaging with the concepts of social identity and group memberships (see for a similar discussion Caruana et al. 2015; Bankert 2021; Lelkes 2021). An additional source of conceptual and empirical confusion comes from the fact that some of the current operationalizations employed for NPI partly overlap with those that are commonly used to operationalize the related but yet distinct phenomenon of affective partisan polarization (AP), i.e., “view[ing] opposing partisans negatively and copartisans positively” (Iyengar and Westwood 2015, p. 691) or “hostility between rival political partisans” (Huddy and Yair 2021, p. 1). As discussed by Russo et al. (2023) PPI is considered to be one of the causes of AP, as intergroup attitudes are deeply rooted in people’s social identities (see for instance Abramowitz and Webster 2016; Webster 2018; Iyengar et al. 2019; Ridge 2020). If we accept that NPI is specular to PPI, then we need to consider it as a driver of AP, as we do with PPI. However, as several of the current measurements of AP and NPI often overlap, this is de facto empirically impossible.

The idea that NPI is an identity is not foreign in the field of political behavior. In fact, some recent NPI measurement proposals embrace its identity-based nature, but they are either too lengthy (e.g., Bankert 2021, composed of eight items, thus not fit for surveys in multiparty systems),Footnote 2 or use PPI as a precondition (e.g., Lelkes 2021). Others are instead not explicitly focused on identity (e.g., Garry 2007). In this paper, we develop and validate a measurement instrument for negative partisanship in multiparty systems that satisfy both needs: it measures NPI appropriately as an identity, and it does so by employing two items. In order to develop this new measure, we follow current best practices for instrument development (e.g., Boateng et al. 2018). Specifically, we first identified the domain via six expert interviews, second we collected previous items and generated new items, which we pre-tested via twenty cognitive interviews, third we fielded a survey with national samples in Italy and Germany, and finally construct validate our measure against a nomological network of hypotheses derived from the literature. We use Germany and Italy as countries for the validation because despite some differences (e.g. the stability of the party system), these countries share similar levels of AP (see Wagner 2021), and a pronounced left-right divide that structures the political space (Pelizzo 2003). Another important feature that these two countries share is that they both have experienced right-wing regimes in the recent pastFootnote 3. This, along with the fact that both countries have far-right parties, lead us to think that they were a good test case. In fact, we know from recent research that far-right parties are the ones giving and receiving the highest levels of dislike (Harteveld et al. 2022).

2 Conceptualizing Negative Party Identification

2.1 Negative Party Identity in multiparty systems

A flourishing literature has extensively studied how voters relate to political parties (Campbell et al. 1960), since first proposed PPI as a long-term affective bond with a party. Currently, two strands are identifiable in the literature: the perception of PPI either as an instrumental or expressive partisanship (for a review see Huddy and Bankert 2017). Instrumental partisanship is rooted in the rational choice paradigm, and it implies that voters support a certain party because it is closer to their policy preferences and party performances. Expressive partisanship is instead entrenched in the social identity theory, as it sees partisanship as yet another deeply-rooted social identity (Greene 2002; Iyengar et al. 2012), which has an affective nature and causes motivated reasoning against arguments challenging the favorite party. Both these traditions primarily address PPI, but in recent years some work has focused on NPI as well (Garry 2007; Caruana et al. 2015; Mayer 2017; Bankert 2021).

Whilst the literature engaging with PPI has always thrived (Johnston 2006), negative party identification (NPI) has so far received considerably less attention. Even though NPI was part of the original definition of party identification, it was mentioned only in passing (Campbell et al. 1960). It was in fact only noted that identification can be “negative”, and it is the “repelling quality” of the party that is central for NPI, without anything else regarding the concept or possible ways of measuring it. Consequently, the few existing studies do not work on the base of a shared notion of NPI and engage in different measurements (Maggiotto and Piereson 1977; Richardson 1991; Rose and Mishler 1998; Garry 2007; Mayer 2017; Spoon and Kanthak 2019; Ridge 2020; Bankert 2021).

As to why NPI has suffered from this lack of interest compared to PPI, the mostly likely answer is that it was for long just considered to be simply the other side of PPI. However, when trying to conceptualize NPI as a self-standing concept, the first question that arises is whether a negative identity can exist without its flip positive side. If earlier research indeed questioned that NPI could exist independently of PPI (Maggiotto and Piereson 1977), nowadays there is broad agreement that individuals may hold both types of identification independently from one another (Rose and Mishler 1998; Medeiros and Noël 2014; Bankert 2021). According to recent works based on social identity theory (e.g. Leonardelli and Toh 2015), individuals vary when it comes to the categories/groups they emphasize for obtaining positive distinctiveness (Turner et al. 1987). Most people hold intergroup categorizations and clearly defined demarcations of the ingroup (“us”) and outgroup (“them”), in our case positive and negative party identification. However, psychological research has shown that individuals may just use a clearly defined outgroup for differentiation of the social self but do not have a fixed conception of the ingroup and thus only hold a negational identity (Zhong et al. 2008).

By rooting our research in the social identity approach, we can identify the outgroup as a central element of importance when it comes to NPI (e.g., Ridge 2020). Following Zhong et al. (2008)’s work on “negational identities”, in this paper we consider NPI as a case of such a negational identity. Individuals either have a fully formed partisanship relying on intergroup categorizations, i.e., having positive partisanship or NPI, or only hold such a negational identity that is defined by outgroups only. In both cases, NPI should be studied as a concept independent from PPI, and that brings the same advantages for individuals as PPI: it provides structure for dealing with the political environment, and it offers affective benefits based on not belonging to a certain social category or group.

The fact that in some studies negative partisanship and affective polarization are used almost interchangeably (e.g. Webster 2018), even though they are distinct concepts, probably stems from the fact that they are both of affective nature and connected to outgroups with the capacity to overwrite cognitive information by motivated reasoning. Regarding their difference, a clear demarcation line can be drawn with the help of the socio-psychological constructs of identity and attitude: NPI would be closely related to the former, and AP to the latter (similar suggestions are made more implicitly by Bankert 2021, Areal 2022, and explicitly by Mayer and Russo 2022). AP is actually already largely operationalized as an attitude. In fact its more often employed measure, the like-dislike scale, does indeed capture an attitude – the (dis)liking of a political object –, but also other operationalizations are attitudinal measures (for a discussion on the different operationalizations see Röllicke 2023). Psychological literature has established that although identity and attitudes are of course intertwined, they are still conceptually distinct (Oppenheim 1982; Hallajow 2018). We also know from social identity theory that social identities cause intergroup attitudes due to various processes of evaluations and group-based comparisons (Tajfel et al. 1971; Turner et al. 1987). This entails that social identities such as PPI and NPI drive AP and not the other way round. We hence argue that it is fundamental for researchers to be able to distinguish, both at the theoretical and at the empirical level, between a (negative) identity and a (negative) attitude.

In a recent contribution Lelkes (2021) has thoroughly engaged in how NPI unfolds in the USA, arguing that partisan disdain does not necessarily imply a negational identity, and that the latter, also shown by Lee et al. (2022), is anyways not a large phenomenon among the American electorate. However, both studies focus on the USA, which features a two-party system with very different implications than a multiparty one. We argue that the peculiarities of multiparty systems might affect NPI, hence, it is particularly important to measure it separately from PPI in these settings.

The main layer to take into account is that multiparty systems offer the opportunity to include party families as structuring element and observe how those affect NPI. Previous works indicate that the division into ideological camps may have a central role (Richardson 1991; Mayer 2017), but research in this area is still scarce. A recent study by Meléndez and Kaltwasser (2021) stresses the importance to consider party system make-up with regard to ideological divides especially with regard to one party family. As populist radical-right parties (PRR) have become a stable feature in Western Europe, their unique position can offer key insights to advance our knowledge of NPI in European multiparty systems. PRR are a heterogeneous party family characterized by the combination of populism, nativism as well as authoritarianism with considerable anti-elitist but also anti-pluralist elements – according to them, they are the only legitimate option to represent the pure will of the people. Consequently, their sympathizers show lower levels of support for liberal democracy as well as lower levels of democratic satisfaction (Müller 2017; Harteveld et al. 2022). PRR are also reported as the party family that has the highest share of negative partisans in Western Europe (Harteveld et al. 2022; Meléndez and Kaltwasser 2021). Those who identify negatively with PRR show higher levels of democratic support, as NPI in this case could not only relate to how one perceives the political environment, but rather a statement towards liberal democracy and the legitimacy of political pluralism. Thus, one might argue that NPI towards PRR and towards other parties fulfil different functions. In other words, if NPI towards mainstream parties could be based on policy preferences, NPI towards PRR could also be motivated by a broader support to democracy in one’s country.

2.2 Nomological network

As a result of our new proposal of NPI’s theoretical concept, we rely on the process of construct validation (Cronbach and Meehl 1955) and derive a nomological network including hypotheses that link this construct to other constructs which we use to test how our new measurement instrument performs against others related constructs. In addition to proposing a measure of NPI, we aim to develop a short scale capable of addressing situations involving more than two parties. We then begin by developing with a longer measurement instrument, which can include several aspects of NPI, which we then shorten based on theoretical considerations, factor analysis, and validation hypotheses.

Following the procedures already used by others for validating PPI measures, we identify categories that are at the core of NPI (Budge et al. 1976). Hence, a valid measure of NPI is one that demonstrates the empirical relationships that we now discuss. First, it is unquestionably true that, despite the fact that the old NPI measures targeted primarily attitudes (instead of identity), these attitudes are intertwined with elements related to negativity and political parties. Several authors agree that the current measures are good enough proxies to be employed in empirical analyses (Medeiros and Noël 2014; Caruana et al. 2015), thus we expect that the new and old NPI measure correlate positively on a moderate level (Hypothesis 1).

A second step consists in testing the relationship between NPI and other variables. We employ three major concepts that are often used in the literature in relation to partisanship. First, we look at satisfaction with democracy. PPI is related to the internalization of citizenship norms which, in turn, normally increases democratic satisfaction (Spoon and Kanthak 2019). On the contrary, NPI implies considering some parties not a viable option, whilst they are still part of the system. Literature has shown that this makes voters question the general way in which democracy works (Spoon and Kanthak 2019; Ridge 2020; Bankert 2021). However, a distinction needs to be made among those holding an NPI towards a PRR, and those who hold an NPI towards another party family, as they entail different considerations. Thus, we expect that NPI should in general relate negatively to satisfaction with democracy (Hypothesis 2a), however, for those having a NPI towards a PRR we expect the relation to be less negative (Hypothesis 2b).

Third, as previously demonstrated, PPI and NPI have their own substantial, positive relation with individual turnout (Medeiros and Noël 2014; Mayer 2017): PPI increases turnout because people have a party for which they root and whose election can be experienced as an expressive act (Huddy et al. 2015). NPI also increases turnout because one has stakes in an election, in the sense that they have a party they want to prevent from succeeding. Thus, NPI should increase the probability to turnout, alongside PPI (Hypothesis 3).

Fourth, following from the argument made earlier, we argue that NPI and its related attitudinal level (that is, negative partisan affect and/or affective polarization) are related yet distinct concepts. We expect their relation to be positive and moderate to high, as social identities relate to increased intergroup bias (Abramowitz and Webster 2018), with positive identities driving ingroup favoritism and negational identities driving outgroup derogation. As negative partisan affect has been often used to operationalize NPI (Mayer 2017; Ridge 2020), we were rarely able in the past to empirically study this relation – with our new measure, this should now be possible. We thus expect that NPI and AP should be highly correlated (Hypothesis 4).

3 Constructing the NPS scale

3.1 Previous measurement instruments for Negative Party Identification

Items measuring attitudes, such as “liking” or “disliking” a party, cannot be used as indicators of NPI, because an attitude, although linked to it, is not an identity (see Mayer and Russo 2022). If we accept, as the literature tends to agree, that partisanship is not a mere preference, but based on identity (for a review see Bartle and Belluci 2009, p.5), then we need measurement instruments that are conceived to capture a social identity. In line with the procedure illustrated by Boateng et al. (2018), we propose a new measurement instrument following these steps: (1) the identification of domain and items generation via a review of the literature and expert interviews; (2) we assess content validity via cognitive interviews; (3) we develop a scale and we administer it to a large sample; (4) we shorten the scale to assure parsimony via a IRT model; 5) we test dimensionality via both an exploratory and confirmatory factor analysis; (6) we test reliability via Cronbach’s alpha; and finally (7) we test construct validity.

Starting with point 1, Table 1 offers a non-exhaustive overview of the main attempts so far to operationalize NPI. We selected these operationalizations because these are the ones that more implicitly or explicitly engage with NPI as an identity.

Table 1 Three examples of NPI operationalization from previous research
Full size table

Garry (2007) heavily relies on the terms/concepts of supporter and opponent. In categorizing supporters/sympathizers as opponents, the out-group categorization process is captured as a stable trait (“them”) rather than an attitude. However, one can argue that opponent is not necessarily defying a negative identity. Furthermore, opponent is not even necessarily accompanied by the attitude that would be intertwined with a negative identity, that is, strong disdain, failing to capture repulsion. In her US-based study, Bankert (2021) uses the original social identity scale Identification with a Psychological Group (IDPG) in its adaption by Huddy et al. (2015) which originates from the early 1990s (Mael and Tetrick 1992). She changes the wording to reflect negative partisanship. While Bankert’s eight-item scale performs well, it has two key disadvantages. Firstly, it is difficult to apply in large-N surveys in multiparty systems because respondents would have to answer eight questions for each party. This would lengthen survey time and increase mental load and fatigue in the data collection process. She tests a shorter version composed of only three items (marked in Table 1 with an asterisk) but, it remains to be verified whether the distinction in “them” vs. “us” truly captures NPI in the eyes of respondent, especially in a multiparty system. This is not a minor setback considering that this is the only item that focusses on self-categorization, and, as we will see later on, it has a poor characterization of voters’ relationship with a party. Second, and more importantly, this instrument relies mainly on an older scale from organization studies that is rarely used nowadays in social psychology, not accounting for different dimensions nor focusing on the unique nature of partisanship as a social identity.Footnote 4 Lelkes (2021) recently proposed a single item based on Zhong et al. (2008) to measure NPI. However, as he uses PPI as a filter question, and asks dichotomously about the own or the other party, this is not a feasible solution in multiparty contexts and does not allow for capturing negative partisans without PPI.

We argue that the existing approaches advanced tremendously the field, but that further efforts need to be made to develop a measure fully equipped to capture NPI in multiparty systems. New items should target the underlying identity dimension and stem from the social identity theory literature to ensure a close connection to previous research in social psychology. They should take into account previous research on negational identities (Zhong et al. 2008; Leonardelli and Toh 2015) as well as newer studies on the measurement of social identity that do not rely on older scales, such as the IDPG, anymore, but focus more on complex, multi-dimensional measurements (e.g., Leach et al. 2008). Thus, we propose to develop and validate a new measure of NPI which engages directly with one’s social identity, is suitable to be used in multiparty systems, and can be used in a short version (with one or two items only).

3.2 Using expert interviews as a starting ground

These elements identified in the previous section guided our development of a new measure. After our own extensive review of the literature, in October 2021, to gather further insights into the field, we conducted six expert interviews. Experts were selected based on previous published research in the area, but we also considered variation in terms of country of origin/research (US, Israel, the Netherlands, Switzerland) and gender (five men, one woman). We conducted semi-structured interviews (the full guidelines are available in the Supplementary Information, part A) on the definition, dimensions, and nature of NPI, the proposed origins as well as its relation to affective polarization. We also asked the experts for a general assessment of existing measures,Footnote 5 based on how well they capture NPI according to Caruana’s (2015) definition.

Our experts agreed with the proposition of NPI as an expressive and multi-dimensional form of partisanship that is also part of a person’s social identities. They all agreed that NPI is more relevant for the study of multiparty systems (compared to two-party systems), as it allows to better understand underlying currents of dislikes and rejections. Four out of six recognized to have a NPI themselves. They also acknowledged that it may be closely linked to affective polarization, but still conceptually distinct from it. Our experts all agreed that current measures did not entirely fit Caruana’s (2015) definition, as either the repulsion failed to be captured, as well as the identity component (commenting Garry 2007 and Mayer 2017), or shortening for a multiparty system application was needed (commenting Bankert, 2021).

By combining existing literature and elements that emerged from the expert interviews, we identify five non-exhaustive and non-distinct domains that carry NPI based on socio-psychological literature (especially Leach et al.’s (2008) five-component hierarchical model). Table 2 shows the items developed in this phase. The three dimensions of sense of self, ingroup emotional significance, and relation with other adherents were modelled to account for the general quality of social identities based on previous Social Psychology literature (Cameron 2004; Mayer 2017). Unlike categorizations such as gender or ethnicity, partisanship is a social identity that is self-chosen, and that relies on an organization who itself acts in the political sphere, namely the party. We thus wanted to account for this by including two more domains that also emerged from the interviews: namely the importance of the general values of a party (opposition to worldviews) and an evaluative dimension, which takes into account longstanding trends. The dimension related to values was directly or indirectly very prominent in most of the expert interviews, whilst the evaluation of the past performance only emerged twice. However, we thought that it might have been useful especially in relation to parties with no clear ideological connotation and/or worldview, such as the Five Star Movement in Italy (which labels itself as post-ideological).

3.3 Scale development employing cognitive interviews

As recommended by the literature (Loevinger 1957; Clark and Watson 1995; Boateng et al. 2018), we developed a comprehensive and broad set of items based on the expert interviews along with the existing literature review (see Table 2). Two or three items were designed for each of the five areas discussed in the paragraph above, for eleven items to be tested in the cognitive interviews.

We conducted cognitive interviews to understand whether the items where consistently understood in both countries, and independently from education levels. This is an important quality stage that gives important input to finalize the questions for the survey (McColl 2006). As suggested by e.g,. Boateng et al. (2018) and Beatty and Willis (2007), respondents were asked first to verbalize their understanding of the questions, and then to answer them.

Table 2 Dimensions and items used for cognitive interviews
Full size table

The questionnaire (available in SI, part B) was administered to twenty respondents, ten in Italy and ten in Germany, in late October and early November 2021, either by personal interview, by telephone or by web interview (e.g., via Zoom or Skype). Each country set of interviews included five respondents with university education and five without university education. From the interviews, it clearly emerged that some of the items were more consistently understood in both countries. It also seemed that the one very classical item “When I talk about this party I rather say they than we” did not perform particularly well even though it is often used (Bankert 2021). We decided to take for every dimension the item that worked better in the interviews, but we also decided to include two for the sense of self-dimensions, because of the widespread use of the us versus them item. In the end, we selected the following six items for the next stage:

  1. 1.

    Because of their worldviews, I could never vote for this party (rejection of worldviews).

  2. 2.

    It is important to me that I am not one of those that vote for this party (sense of self).

  3. 3.

    When I talk about this party I rather say ‘they’ than ‘we’ (sense of self II).

  4. 4.

    I am glad when this party loses an election (emotional significance).

  5. 5.

    I have nothing in common with people voting for this party (rejection of other adherents).

  6. 6.

    I would never vote for this party based on their political performance in the past (longstanding trend).

We measure all items on seven point rating scales in order to insure reliability (Krosnick and Presser 2010) and to treat them as interval variables (Rhemtulla et al. 2012).

4 Data and methods

The data for this study were collected in Germany and Italy in March 2022. The market research company Respondi, with matched quotas for age, gender, and federal state/region, fielded it online using an Online Access Panel. All participants with missing values on the independent variables were excluded (97 cases in Germany, 68 in Italy, final N Germany = 1,911, N Italy = 1,440). Due to the data originating from a non-probability sample, although quotas were applied, generalizations will be drawn with caution. Replication material is available at OSF: https://osf.io/brw4u/?view_only=ed55fa7db956469ba4118b62b2a8ebb1.

4.1 Operationalization of key concepts

Both Germany and Italy have a multiparty system, which includes bigger and smaller parties. To make the questionnaire feasible and limit the drop out, we decided to include only a limited number of parties. We selected the six parliamentary parties for Germany (CSU/CDU, Social Democrats, The Left, The Greens, Free Democratic Party, Alternative for Germany) and seven for Italy (Brothers of Italy, Lega, Go Italy, 5 Stars Movement, Italy Alive, Democratic Party, Free and Equals/Art.1).

The newly developed items for NPI were not asked about all parties, as this would have resulted in a too lengthy questionnaire (36 questions for Germany and 42 for Italy). However, only asking it for one party would have prevent us to capture the different natures of NPI towards a radical right party and the one towards other kind of parties (if any). If we were only measuring the strongest NPI, we might only be able to capture NPI with one of the radical right parties. Thus, we decided to measure all items for two parties: first asking respondents to rank the parties from the best to the least favorite, and then asking the new items only for the least and second least favorite in Germany, and least and third least favorite in ItalyFootnote 6. This difference was implemented because we considered that voters that would choose Brothers of Italy or Lega as least favorite were also very likely to answer the other one, as second least favorite, and the two parties are ideologically too close as they could both qualify as radical right party. We considered different ways to aggregate the separate items for NPI to a scale. We assume it is a scale with subdimensions that can be aggregated to a single factor of higher order. We do not have theory-based reasons to expect that NPI is made of several subdimensions that have to be necessarily fulfilled. Thus, we treat NPI as a compensatory measure for which we use a mean index over all items.

PPI was measured with the standard questions from the CSES, asking about closeness with a political party, including the follow-up question and questions about which party and strength (0 no party ID, 1 not very strong, 2 moderate, 3 very strong). About 37 (Germany) and 46 (Italy) percent of respondents do not identify with a political party.

We measured AP as in the CSES, i.e., by asking respondents to rank each party on a like-dislike scale (7 points). The AP measure was then calculated based on the recommendation by Wagner (2021) for the weighted spread-of-sums variant (mean Germany: 1.46, SD 0.55; mean Italy 1.37, SD 0.73). With regard to NPI, along with the newly developed five items, we also asked the usual questions i.e., whether there was a party the respondent would never vote for. Feeling thermometers were included for all parties, ranging from − 3 to + 3. We also included questions about political interest, trust, satisfaction with the way democracy works in the country (1 not at all satisfied to 7 totally satisfied), left-right self-placement, the probability to vote at the next federal election (1 not at all likely to 7 very likely), and sociodemographic variables (see SI, part C, Table C1 for the distribution for all variables).

We also recoded all independent variables to the range 0–1 to be able to compare unstandardized coefficients across models in the regressions analysesFootnote 7 for which we use robust standard errors as the dependent variable is highly skewed.

4.2 Descriptive analyses for the new scale

To identify the most disliked and second or third most disliked party our respondents were asked to rank parties from most to least disliked. Table 3 displays the distributions for these questions including information on party families as proposed by the CSES. Whilst in Germany, the AfD clearly appears as most disliked party among nearly 70% of the respondents, in Italy the picture is much more fragmented.

Table 3 Distribution most disliked and second/third most disliked party
Full size table

The six items measuring NPI are on average correlated by r = 0.4–0.8. The items correlate with overall similar patterns in both countries but also for both parties (see SI, part D, Table D1).

Next, we calculate mean scores for every respondent over all available items from the scale. The distributions (see SI, part C, Table C1) are left skewed in both countries, and more left skewed for the first than the second/third disliked party. We also test the reliability of our six-item battery by using Cronbach’s alpha (Cronbach 1951). They range between 0.87 (Germany, most disliked party) and 0.94 (Italy, third disliked party), and thus exceed the 0.80 threshold of acceptability indicated in the literature (Boateng et al. 2018).

5 Parsimony and dimensionality test

There is an ongoing discussion in the social sciences regarding advantages and disadvantages of short scales that are easy to field, versus measurement quality, which is higher when using multi-item measures. However, the general goal of measurement scales is to be as parsimonious as possible. We follow the next step suggested by Boateng et al. (2018), that is, an IRT test. This allows us to test which items have large and positive discrimination values (normally from 3 onwards), that is, which items are highly consistent with the underlying trait we aim measuring (NPI). As Table 4 shows, the Them versus Us- item scores a low discrimination value in both countries. Having no things in common seems also to have a low discriminatory value in Germany, but not in Italy.

Table 4 IRT analysis
Full size table

Following previous approaches for scale development, we perform a Confirmatory Factor Analysis (CFA).Footnote 8 How can we optimize the scale? Table 5 shows a CFA performed both on six and five items (without the “them vs. us”-item) for the most disliked party. In both countries, the lowest factor loading was again the one for the Them versus Us-item for the six-item CFA. When eliminating this item the fit of the model for Germany also slightly improved. We see similar results for the second- and third-most disliked party (which is displayed in the SI, part D, table D4). Based on the results from the different approaches, we thus use the five-item measure for the following analyses.

Table 5 Results for confirmatory factory analyses for the most disliked party, standardized factor loadings
Full size table

6 Validation analyses

6.1 Correlation old and new

Our first expectation when validating the new five-item measure was that the old and new measures of NPI should positively relate to each other. We use three different versions of old operationalizations that have been used in previous studies. Specifically, we use a combined measure of a negative feeling thermometer score combined with the answer to the question if one would never vote for this party (as in e.g., Mayer 2017), as well as the feeling thermometer score and the question if one would never vote for the party separately (as in e.g., Ridge 2020).

Table 6 shows the correlations for the two parties that were also used to test our NPI measure. We see that correlations are positive and significant (p < 0.001), but the strength is moderate. We argue that this is because NPI captures much more deep-going feelings than merely sympathies.

Table 6 Correlations between new measure NPS and old measures for NPI
Full size table

6.2 Predicting democratic satisfaction and turnout

Both in Germany and in Italy, we can mostly observe the expected significant negative relationship between NPI and democratic satisfaction, which is displayed in Fig. 1. We find substantial significant effect sizes of about − 0.5 scale points for the most disliked party in the models including the interaction between NPI and having a NPI with a populist radical right party (M1b and M2b), both in Italy and Germany. For the second/third most disliked party, NPI also relates negatively to democratic satisfaction, but the effect size is smaller and it does not reach conventional levels of statistical significance in Italy. Furthermore, having a NPI towards a populist radical right party relates to greater democratic satisfaction in the models without the interaction (M1a and M2a). Those having an NPI with the AfD in Germany or Fratelli d’Italia or Lega in Italy show higher levels of democratic satisfaction. When we look at the moderating effect of NPI with a PRR on the relationship between NPI and democratic satisfaction (M1b and M2b) that indeed that a NPI with a PRR, compared to negative identifiers with other party families, relates significantly much more positive to democratic satisfaction in Germany, supporting Hypothesis 2b. However, we do not find such an effect in Italy.

Fig. 1
figure 1

Linear regression on democratic satisfaction, coefficient plot with unstandardized coefficients. Note: Omitted from display age, gender, education, and political interest. 95% CI displayed. Full model, see SI, Table C2

Full size image

Turning to turnout, Fig. 2 shows that both PPI and NPI for the most disliked party affect turnout likelihood positively by about 0.5 to 1.75 scale points, meeting the expectations outlined in Hypothesis 3. The effect of NPS with the most disliked party on turnout is in both countries not much lower than PPI showing the importance of including at least NPS with the most disliked party in analyses of turnout. Compared to models containing only PPI and not NPI (SI, Table Table C3), we see that the explanatory power increases while the coefficient for PPI stays the same; NPI is thus likely not mediated by PPI, but largely independent.

Fig. 2
figure 2

Linear regression on turnout probability, coefficient plot with unstandardized coefficients. Note: 95% CI displayed. Full model, see SI, Table C3

Full size image

6.3 Relation with affective polarization

As stated in Hypothesis 4, we expect that AP and NPI, although distinct concepts, are related and should therefore correlate positively. In line with our expectation, as Table 7 shows, a low and highly significant correlation can be observed between AP and the new NPI measures in both countries (r = 0.13–0.31, p < 0.001). Next, we correlate affective polarization also with the old NPI measures (I would never vote for this party and feeling thermometers), expecting to find a somewhat high(er) correlation. However, in particular for the feeling thermometer the correlations are very similar. This might be due to the calculation procedure of the spread-of-sums variant, but also to the specific country selection for which AP and NPI do not correlated as highly. However, this low correlation can also be observed for the older NPI operationalizations.

Table 7 Correlation between affective polarization and NPI
Full size table

6.4 Towards a shorter scale

So far, we have developed and validated a new scale for measuring NPI as a social identity consisting of five items. However, in a multiparty system using the full scale might not always be feasible or desirable, especially if the battery needs to be asked for all parties. Even though the measurement quality of multi-item instruments for latent constructs such as party identification is usually higher than for single-item instruments, questionnaires often need to be parsimonious. Hence, we deemed it necessary to shorten this scale. Following Gerber et al. (2010) we propose to use the items that provide the biggest range of the concept based on their discriminatory value (see Tables 4 and 5, and SI, Table D3). Two items fall in this category: the ones referring to repulsion of worldviews and sense of self.Footnote 9

To test whether the shorter options work as well as the full scale, we rerun all our analyses (see SI, part E). We compare our full scale against two one-item options (our preferred: the ones about values and sense of self), and with a two-item scale combining both. Results show a similar low correlation pattern with the old measurement (see Supplementary Information, Table E1) for all three options. In terms of democratic satisfaction, we see similar patterns across all three options (see SI, Figure E1 & Table E2). Considering the effect size of only sense of self, it is smaller than the other options, but not significantly different from the other coefficients. It is the same when explaining turnout: the two-item operationalization had a much larger effect size in Germany than the other versions (see SI, Figure E2 & Table E3). Last, the correlations with affective polarization are also similar no matter which shorter version one uses (see SI, Table E4). This shows that shorter versions of the scale are viable options for research purposes. On both theoretical and empirical grounds, we recommend using the two-item version because when thinking about definitions and what a negative identity conveys, sense of self and repulsion to a party’s world view provide an encompassing picture of NPI, and the importance for a voter to be not among those who vote for a given party clearly taps into identity. We argue that these are encompassing items that would capture NPI across countries. This said, our analysis clearly shows that they also both work well as single items, with the one for repulsion of a party’s worldviews performing slightly better (see Table D2 in SI). We thus suggest that for a one-item measurement, the preferred one should be the one for repulsion of a party’s worldviews. We would also argue that although the importance of not being among those who vote for a given party is indeed another appropriate item (as we also show empirically), it does not explicitly tackle why one does not want to be among those voters. Whilst the item on worldview explicitly mentions the rejection of values (which are an important part of one’s social identity), hence pointing out the reason for negative identification.

7 Conclusion

Partisanship is one of the most pivotal concepts in political behavior research. While Positive Party Identification has received a great deal of attention, its negative counterpart has been largely neglected. Despite some recent interest, Negative Party Identity remains a largely understudied topic. This lack of consideration translated in an under-conceptualization of the construct, and in a scholarly debate that calls for further thinking and reflection on operationalization. A clear example of these shortcomings is the occasional implicit use of NPI and Affective Polarization as the same phenomenon, whilst (positive) party identity is known to be a cause of AP. In this paper, we have argued that Negative Party Identification (NPI) is an important construct that deserves a clearer conceptualization and, consequently, a more precise operationalization, especially when one wants to use it in a multiparty context. To achieve this objective, we developed a new measurement instrument reflecting this fundamental feature. The validation of this new instrument has followed best practice guidelines: we have first engaged in a critical review of the literature, we further investigated the theoretical foundation of the concept by conducting six expert interviews, which led, in turn, to a long battery of eleven items that have been pre-tested via twenty cognitive interviews. Based on these steps, we were able to select six items that we have tested in a large-N survey in two countries: Germany and Italy, leading to a final five-item battery. We have shown a high internal consistency for the scale, which we construct validated subsequently by analyzing its relation with satisfaction with democracy, turnout and affective polarization. In a last step, we provided additional analyses for several short versions of the scale to ensure its feasibility in multiparty systems.

We validated and tested our measure in Italy and Germany, two countries with a clear left-right divide, and in one point in time. To further strengthen reliability the new measure should be included and tested on a set of randomly selected countries (Boateng et al. 2018) to demonstrate its cross-country applicability. Collecting NPI longitudinally would also shade light on how this construct evolves overtime in relation to political events or changes in party systems.

As a more general point, we would like to remark that NPI, as mentioned above, is still neglected compared to its positive counterpart. Considering its importance and the new attention that it is raising, it would be a great help for advancing out knowledge on this topic that survey consortia (such as the Comparative Study on Electoral System), along with researchers involved in this field, would consider including a measure on NPI in their data collection. This would advance knowledge on substantive areas of research that are still under investigated, such as the relationship between NPI and underlying ideological identification.

We hope that by offering a viable measurement instrument, future research will explore the role of NPI in multiparty systems, but its relations with a number of concepts and construct, such as institutional trust, political attitudes, and voting decisions. Future research needs of course to consider several further aspects. First, it remains to be discussed whether it is desirable to establish a NPI threshold that would make it possible to define when somebody could be categorized as a negative partisan, in the same fashion in which some of the previous operationalizations have been used. Instead, we have treated the NPI measure as a mean index without any weight in a compensatory way. We did not rely on fixed thresholds, as we believe that dichotomizing the measure leads to loss of information. However, there are situations where researchers could find it useful to rely on such a dichotomy or discrete categories, e.g., for constructing a typology of partisans in the tradition of Rose and Mishler (1998) or conducting experiments. In such cases, we suggest taking the mid point of the answer scale as the cut-off point.

A second important point lies in the novel possibility of deciding whether to measure NPI only for one (or a limited number of) party, or for all of them. Because we started from a larger number of items (six), we had to make a selection in terms of parties. As discussed above, we decided to measure NPI only on two parties per country. However, our new scale offers the possibility to take such decision depending on the objectives of a research design. This seems especially useful in a multiparty system.

Finally, the relationship of NPI with underlying ideological identification needs to be further explored to determine whether NPI is ideologically anchored or not, as this would have interesting implications in the study of the emergence of NPI with new parties.