1 Introduction

In view of its explanation and measurement, satisfaction in the home has been studied time and again. As explanatory factors, different physical comfort domains have received the most attention (temperature, light, indoor climate, noise), with studies often considering as controls also attributes and values of the residents as well as characteristics of the neighbourhood (see Aragonés et al., 2017, for review). Comfort measures are particularly useful predictors because they can easily be transformed into rules and standards for the building industry, guided by the belief that certain physical comfort levels elicit wellbeing at home. Thus, if examined empirically, the comfort-satisfaction relationship is modelled as a psychophysical relationship, S = f (C), with S indicating a measure of residential satisfaction and C a comfort measure of a particular physical modality.

However, following this comfort-driven approach, we should note that satisfaction is only one of several possible reactions to the stimulation of subjective delight. In particular, as psychological happiness research has convincingly established (Kahneman et al., 1999), we need to distinguish experienced pleasure from remembered pleasure, or, in more everyday terms: felt wellbeing from well-thought-out judgements of satisfaction (Kahneman, 1999). Whereas felt wellbeing is the immediate impression of pleasure, potentially captured in Benthamite instant utility functions, satisfaction is based on the more cognitive component of our wellbeing experience. Transferring this distinction to the domestic scene, feeling attached to one’s home is among the most striking features of felt wellbeing (Cooper Marcus, 2006; Dovey, 1985; Easthope, 2004). Therefore, extending the simple model of comfort-driven residential satisfaction that dominates this field of research, we propose the idea of attachment-driven satisfaction. The experience of comfort in the home prompts residential satisfaction; it does so, however, through the mediation process of attachment to the home. In its core then, wellbeing at home is the trilateral relation of physical comfort, home attachment, and satisfaction, with the size of the mediating effect to be determined (Fig. 1).

Fig. 1
figure 1

Causal diagram of residential satisfaction

Why is this addition to the comfort-satisfaction scheme important? There are two arguments, giving reasons for the here presented research—one theoretical, one practical.

First, integrating home attachment into the equation is adding psychological substance to the architectural design process that is restricted to the consideration of physiological reactions otherwise. The concept of human comfort is used to study physical and physiological sensations in response to the environmental stimuli in our surroundings, but it does not involve motivation and the prospect of purposeful actions. Hence, the comfort-dominated approach is highly reductionistic when it comes to explaining residential satisfaction and far from being able to establishing a behavioural theory of satisfaction.

From a practical point of view, second, comfort is used by engineers to create building standards for optimizing the indoor environment for occupants. But when engineering comfort, we must consider that comfort, however defined, is of little value if we do not feel at home in the places, we experience the comfort in. Merging the standard-setting intensions of engineers with the sense of place, therefore, is what is needed for applying comfort rules in actual architectural settings.

In view of these theoretical and practical concerns, this paper addresses the causal relationship of comfort sensations, home attachment and residential satisfaction in an empirical mediation analysis (Pearl, 2014). Based on a large cross-country population survey in Europe, we examine which mediation model explains the data of this comparative survey best. The aim is to establish an empirical baseline for the analysis of the tripartite relationship of comfort, attachment, and satisfaction in residential satisfaction research.

2 Residential satisfaction as a concept

Residential satisfaction as a concept is of amorphous quality allowing for a multitude of definitions. Reviews of the research into this field include Easthope (2004) and Aragonés et al. (2017). The most comprehensive bibliometric analysis of residential satisfaction research is Biswas et al. (2021), citing 877 publications since 1961. They identify the most influential residential satisfaction contributions in terms of their citation frequencies. Emami and Sadeghlou (2019) structure their review according to the core determinants of residential satisfaction and dissatisfaction from an interdisciplinary perspective, and Lewicka (2011) covers the residential satisfaction literature within her extensive report on place attachment research on from the 70s.

Conceptually, the common feature in reviewing residential satisfaction research is the requirement to distinguish at least three dimensions that need specification in the formation of the concept: domains, relations, and processes. (1) According to Canter and Rees’s (1982) early distinction, the domain of residential satisfaction can address either the house, the neighbourhood, or the neighbours. The house is understood as the home. This is where the individual resides and has her or his primary environment. The home is more than having a roof and shelter, it is also a reflection of the individual’s socio-cultural place in society closely tight to the prevailing concept of the family. While there have been a number of studies that address satisfaction with home characteristics empirically (Shin, 2014, 2016; Shove, 2003), most of the research is focused on the neighbourhood domain (Aragonés et al., 2017, p. 313). The problem this research is confronted with, however, is the difficulty of defining what a neighbourhood is, how to determine its spatial boundary and meaningful demarcation (Marans & Rodgers, 1975). There is also the question of whether the emphasis should be on the physical aspects when drawing up neighbourhood boundaries, or on social and economic characteristics. To the extent that the latter is given preference over “neighbourhood as a place,” the focus is shifted to the third residential satisfaction domain, the neighbours. Here the residential environment is viewed in its social character. The interaction between residents is seen as either determined by neighbours’ physical proximity to each other or by their shared feelings of community. Both aspects can be studied from either a social-psychological and group-theoretical perspective (beginning with Festinger et al., 1950) or from that of urban sociology (Hunter, 1974; Kasarda & Janowitz, 1974).

(2) The second dimension in the concept formation of residential satisfaction is the relational aspect. We are satisfied or dissatisfied in relation to something. Following Amérigo and Aragonés’ (1997) classification, the objective of residential satisfaction can be physical or social, depending on whether the stimulus belongs to the physical or the social environment, and it is, second, either subjective or objective in the sense of being of factual reality versus the result of an evaluation. Amérigo and Aragonés’ (1997) and Aragonés et al. (2017) organize the possible predictors of residential satisfaction according to the resulting four-fold table of these types of relations.

(3) Finally, there is the process dimension. Its relevance received attention after the notion of place attachment was included in residential satisfaction reasoning. Focused primarily on the neighbourhood domain (Hidalgo & Hernández, 2001), attachment became an important variable in residential research, standing side by side with the concept of satisfaction (Bonaiuto et al., 1999). Because of the obvious proximity of both ideas, distinguishing them calls for considering different mental processes in generating satisfaction and attachment, respectively. It is generally held now that satisfaction is first of all a cognitive reaction to the residential environment, while attachment qualifies as a feeling and is as such linked to the emotional element of residential wellbeing (Aragonés et al., 2017, pp. 323–325). Because of these process differences, attachment to a place is of a different nature then residential satisfaction.

Based on the above distinctions of residential satisfaction of domains, relations, and processes, we proceed by delineating the two explanatory concepts of this study, comfort and home attachment, vis-à-vis satisfaction. While the residential satisfaction domain we study, clearly is the home, comfort indicates a particular relational aspect towards satisfaction, whereas home attachment is distinguished from residential satisfaction by a difference in process.

3 Home attachment and comfort measurements

3.1 Home attachment

The distinction between attachment and satisfaction, we focus on, refers to Daniel Kahneman’s theory of the mind and his idea of the self as being double layered: There is the self, existing in system 1, and the self, existing in system 2. The main difference is that in system 1 thinking is fast, in system 2 it is slow. As Kahneman writes: “System 1 operates automatically and quickly, with little or no effort and no sense of voluntary control. System 2 allocates attention to the effortful mental activities that demand it, including complex computations” (Kahneman, 2013, pp. 20–21). Satisfaction is a system 2 notion then because it results from reflections and anticipations. Attachment, in contrast, is immediate indulgence, that is, a system 1 entity.

In terms of causality then, if we maintain this distinction, moving it to the domestic scene, we propose that home attachment should be thought of as contributing to residential satisfaction. We base our sense of satisfaction on our instantaneous enjoyment of feeling at home (Coolen & Meesters, 2012; Cooper Marcus, 2006; Easthope, 2004; Porteous, 1976; Soleimani & Gharehbaglou, 2021), relating it next to some comparison standard of previous experiences and benchmark frames as well as future prospects and concerns. But our attachment to the home, first of all, is what we ponder on when we derive residential satisfaction judgements. In our analysis therefore, we predict a causal relationship between home attachment and residential satisfaction or, in Kahneman’s phrasing, a system 1 phenomenon being causally relevant for a system 2 phenomenon (Kahneman, 1999, pp. 4–6).

3.2 Sense of place—directly perceived and socially constructed

The above argument is based on recent developments in environmental psychology and sense of place research. Environmental psychology, in as much as it was established during the second half of the twentieth century originating from James Gibson’s pioneering work in the field of perception (Gibson, 1979/2015) rejected the representational logic of cognitivism as well as the physicalist idea of the stimulus–response chain of behaviourism. Instead, perception is conceived as an active process in which the surroundings of an organism are related to possible actions, giving objects of perception ecological meaning. The substance of this organism-environment combination is, in Gibson’s term, affordance, reminiscent of the Aufforderungscharakter in the Gestaltist tradition (Lewin, 1936). As objects of perception, affordances are non-representational, instantaneous and without any reflective extras (McConnell & Fiore, 2017; Michaels, 2003).

The idea of perceptual directness and immediateness is picked up by place attachment researchers who are critical of the attention long-term processes of place bonding have received. In this research, it is generally accepted that place attachment develops over time, changes slowly and increases with the length of residence in a given place (Giuliani, 2003; Lewicka, 2010, 2011). Thus, place attachment research “privileges the slow” (Raymond et al., 2017, p. 2), the steady and temporal progression through cognitive and social construction of the meaning of place. In contrast, the theory of affordances “engages the fast” (Raymond et al., 2017, p. 4), the immediate and direct perception of place resulting from a perception–action process which couples the individual instantaneously with the actions she or he can undertake in a particular setting (Chemero, 2003; Raja, 2019).

Sense of place therefore comes in two forms, as direct perception that is immediate (the “fast”) and as a higher order socially constructed reasoning process consuming time (the “slow”). While dual-process theories in general have not been exempted from criticism (Evans & Stanovich, 2013), in place research there is widespread agreement that there is both: immediate place bonding as well as a reflective evaluation process which develops out of the direct sensation of being locally attached (Marsh et al., 2009; Lewicka, 2011, pp. 225f). In terms of causality then, the “fast” will normally function as a precondition for the “slow.”

In as much this distinction is mirrored into the concepts of (directly perceived) place attachment and (socially constructed) residential satisfaction, the relationship between both has mostly been studied without referencing affordance and dual-process theory (exceptions are Coolen & Meesters, 2012; Heft, 2003, 2010; Marsh et al., 2009; Raymond & Gottwald, 2020; Raymond et al., 2017). Rather, attachment and satisfaction are seen as attitudinal constructs (Jorgensen & Stedman, 2001; Scannell & Gifford, 2010) distinguished by appropriate measurement schemes and scales. As attitudes however, little can be said about causal priorities. Does place attachment promote residential satisfaction or is the latter a prerequisite for feeling locally attached?

Both views are exemplified in the sense of place literature. For instance, in the context of predicting pro-environmental behaviour (Daryanto & Song, 2021), Ramkissoon et al. (2013) consider place attachment as antecedent of place satisfaction in structural equation modelling (see Ramkissoon et al., 2012, for a review of similar studies), whereas others, e. g. Ramkissoon and Mavondo (2015) or Casakin and Reizer (2017), choose to reverse the relationship placing satisfaction before attachment. Vada et al. (2019) use satisfaction with tourism experience as predictor for place attachment, whilst Rollero and De Piccoli (2010) as well as Scannell and Gifford (2017) view place attachment as a cause for satisfaction experiences.

The obvious arbitrariness of such causal sequencing is inevitable as long as there is no theoretical frame for connecting both concepts, or, in Guiliani’s words, as long as “the distinction between satisfaction and attachment rests more on empirical results than on a theoretical basis” (Giuliani, 2003, p. 149; also Fried, 2000; Lewicka, 2011, p. 218). Without a theory of perception, the order of the relationship is bound to be ad hoc. In contrast, the dual-process scheme of fast and slow perceptions in place research, superseding empirical generalisations, allows for causal inference.

3.3 Comfort measurements

Also, sensations of comfort can contribute to residential satisfaction. The research in this area, however, does usually not address satisfaction with the home environment directly but bodily wellbeing in response to particular physical stimuli, i. e. the indoor environmental quality (IEQ) mostly assessed in terms of air temperature, air quality and speed, illuminance, and sound level.

Comfort measures are generated either objectively by defining comfort zones of specific energy levels of the respective physical modalities or subjectively through surveying occupants of buildings and record how pleasing they find certain thermal, lighting, or noise conditions. The prototypical project of assessing comfort by objective means was Fanger’s climate chamber research (Fanger, 1970) resulting in his famous PMV-PPD model of thermal comfort. The model is based on the difference of predicted mean vote (PMV) of experimental subjects and the predicted percentage of the dissatisfied (PPD). With some modification, the model has been cast into standard norms such as EN ISO 7730 (ISO, 1994), and is now widely used by engineers.

Architects are more inclined to follow the subjective measurement approach relying on field-based research that focuses on the occupants and their adaptive expectations, as it is implemented, for instance, in the Adaptive Comfort Standard (ACS) originating from the works of de Dear and Brager (1998). The ANSI/ASHRAE standard in its latest edition (ASHRAE, 2017) attempts to merge both types of measurement in its comfort standard. This is also true for the European standard EN15251, which provides guidance on IEQ measurement but is mainly used in energy simulations in accordance with Directive 2002/91 on the energy performance of buildings of the European Parliament and Council (EPBD, 2003; Olesen, 2012). Great effort has also been put into developing weighting schemes for combining effects of multiple environmental factors—e. g. indoor air quality, thermal environment, acoustics, lighting—to derive a holistic index of comfort in buildings (see Heinzerling et al., 2013; Altomonte et al., 2020; Larsen et al., 2020; Leccesse et al., 2021, for reviews).

3.4 Physical and functional comfort

Empirical survey-based studies of comfort in buildings have come up with an additional distinction. There are comfort dimensions proper and comfort dimensions related to the functionality of the house. The latter refer to operational parameters like energy consumption, renovation status, technical facilities, home size and, for instance, a dwelling’s resistance against mould. Functional considerations do not involve immediate feelings of physical comfort vis-à-vis temperature, proper lighting, or indoor climate. In fact, exploratory studies leading to the present research have established two salient factors of comfort measurement: physical comfort and the assessment of functionality (Wegener et al., 2014; Wegener & Fedkenheuer, 2016), a distinction already familiar in the measurement of the environmental quality of work life (Vischer, 1989, 2007) as well as in the literature on housing preferences and choice (Jansen, 2014; Jansen et al., 2011; Sirgi et al., 2005). Thus, we need to examine how these different elements of comfort relate to the attachment to home and, eventually, residential satisfaction.

4 Mediation analysis models

4.1 Causal analysis

Studying satisfaction predictors is different from studying causation. If we have established a close correlative relationship between satisfaction predictors and satisfaction, this does not tell us how both are connected or what mechanism is causally relevant for the outcome (Francescato et al., 2018). An advanced field of research is the causal analysis of estimating and testing mediation processes with latent and observed variables that, in contrast to conventional mediation analysis, takes random and non-random measurement errors into account (Muthén & Asparouhov, 2015; Pearl & Mackenzie, 2018; Vanderweele, 2015). As such, it has transformed the Hempel-Oppenheim covering-law model of explanation in defining the concept of causality (Hempel & Oppenheim, 1948) and has been shown to be appropriate for experimental as well as nonexperimental data (Pearl, 2010; Pearl & Mackenzie, 2018).

The core of causal analysis, based on directed cyclical graphs, is that it is assumed that a variable C is causing another variable S, the former being the causal variable, and variable S that it causes is the outcome. If we define C = comfort and S = satisfaction path c in Fig. 1 is the total effect of C on S. However, the effect of C on S may be mediated by a variable A (for attachment), while the variable C may still affect S. In this situation path c becomes the direct effect, and variable A is the mediator or process variable.

This mediational model is a causal model as it is assumed that the mediator causes the outcome. Complete (full) mediation is given when variable C no longer affects S after A has been introduced, cutting path c to zero. Partial mediation is the case in which the path from C to S is reduced in absolute size but is still different from zero. The amount of mediation in the model is called the indirect effect, and the total effect is defined as the direct effect plus the indirect effect. Often in research, there is inconclusive evidence whether there is full or partial mediation. It is important therefore to test the two alternative nested models, in particular if it turns out that direct and indirect effects have opposing signs causing the total effect to be diminished or disappear altogether (Muthén & Asparouhov, 2015).

4.2 Mediation models

For the different classes of predictors of residential satisfaction, we use mediation analysis for testing the appropriateness of competing models. First, we have the conventional relationship where physical comfort is assumed to determine satisfaction, S = f (C), with S indicating a direct measure of residential satisfaction and C indicating comfort measurements with regard to different physical modalities (Model Ia). According to our causal diagram of Fig. 1, the total effect of C on S in this case is c. Alternatively, this model can also take the form of S = f (F), with F symbolizing measurements of functional comfort instead of physical comfort (Model Ib).

The next model (Model IIa) takes physical comfort as the dominant driver of satisfaction but considers home attachment (A) as the comfort mediator, formally: S = f (C, (A ← C)). This is the classical partial mediation scheme with the direct effect of C on S. Again, reflecting on the role of different comfort realms, we can replace physical comfort with functional comfort such that S = f (F, (A ← F)). This is Model IIb.

Finally, we can leave out the direct effect of either physical or functional comfort on satisfaction in order to test for complete mediation (c = 0). We then have Model IIIa: S = f (A ← C), and Model IIIb: S = f (A ← F).

Thus, three different models are defined for the mediation analysis of physical and functional comfort, respectively: no mediation (Models I a–b), partial mediation (Models II a–b), and complete mediation (Models III a–b). These are summarised in Table 1.

Table 1 Satisfaction models for physical and functional comfort

It should be noted that the dependent variables C, F, and A can be conceived as being latent constructs; as such the intended mediation analyses of our study deviate from the classical meditation scheme applied to observables only (Hayes, 2018).Footnote 1 Accordingly, we will apply the alternative model strategy (Jöreskog, 1993; Kline, 2016, p. 11) within structural equations methodology (SEM) to the models, seeking appropriate measurements of the involved observables (Pearl, 2012). And we will ask whether the trilateral relationship of satisfaction, comfort and attachment captured in the models holds under all circumstances, in particular whether it is valid for different groups of respondents.

As will be described in the next section, we utilised recent survey data of a comparative housing quality study in 14 European countries to test the models. Because different climatic regions are covered, we differentiate between three climate zones in which we study residential satisfaction in our sample: the European Continental zone, the Atlantic zone, and the Mediterranean zone. Respondents living in these zones form three different groups. In addition, we define groups of respondents by country groups and test for model-specific group differences with regard to these different population groups.

5 Data

The database for the study is the Healthy Homes Barometer (HHB), an annual survey of attitudes and behaviour in Europe regarding housing quality, residential satisfaction, comfort, health perception, energy consumption, and environmental impact. It was first fielded in 2014 (HHB, 2015) and reported to the European Commission in 2015. In the present analysis, the Healthy Homes Barometer 2016 is used (Beranova et al., 2017; HHB, 2016). Participating countries in this cross-sectional wave were: Austria, Belgium, Czech Republic, Denmark, France, Germany, Hungary, Italy, Norway, Poland, Spain, Switzerland, the Netherlands, and the UK. The survey was administered as an online-panel survey in the involved countries in October 2015. The mean completion time of the questionnaire across all national samples was 17 min.

Based on online panel sampling procedures (Callegaro et al., 2014), national representative samples of 1,000 respondents, aged 18 and higher, were drawn in each country. The number of respondents from each country was set to ensure statistical representation. The samples were drawn from national online panels that secure representative distributions of key demographic variables (age, gender, education, regional representation).Footnote 2 The validity of results was checked with distributions from national resemblance census data and yielded no significant aberrations (Wilke, 2014). The 14 countries surveyed represent more than 430 million Europeans, accounting for more than 70% of the total European population. Furthermore, the selected countries embody a variety of sizes, geographic locations, and climate conditions.

When concluding on the European level, responses can be weighted according to a specific country’s share of the population of the 14 European countries surveyed as a whole. For analyses based on association measures (regression and factor analysis), however, unweighted estimates are usually preferred (Winship & Radbill, 1994). This approach is also followed in this study.

6 Measurement

The variables relevant for the satisfaction models were operationalised as questionnaire items in the Healthy Homes Barometer with category rating scaling methods. Constructing these items, we relied on exploratory case studies of comfort and satisfaction in residential housing. The studies were part of the Velux Model Home 2020 project (Velux, 2015). Covering a period of three years, the studies were carried out in model homes especially designed for this purpose. During the experiment, the houses were closely monitored both in terms of physical performance and of the psycho-social functioning of the residents. Several methods were used for exploring residential life: group discussions, structured face-to-face interviews with family members, self-reports using diary methods, and online questionnaires. These different measurements led to a very detailed recording of the family life in the houses. Eventually, using factor analytic methods and external validation studies, this exploratory material was transformed into questionnaire items suitable for population surveys. The methodological procedures that were employed are described in Wegener et al. (2014), Fedkenheuer and Wegener (2015), and Wegener and Fedkenheuer (2016, 2017). The outcomes were cross-checked with monitoring results of the complete Model Home 2020 series in Denmark, Austria, France, and the UK. These cross-country comparisons are reported in Fedkenheuer and Wegener (2014).

In the following, we describe the variables for assessing residential satisfaction, comfort, and home attachment that were included as part of the Healthy Homes Barometer, forming the database of the present study.

6.1 Satisfaction

As the dependent variable, residential satisfaction, a variant of the universally applied survey question on which satisfaction research is based, was modified to suit the domestic environment (Cheung & Lucas, 2014; Diener et al., 2012; Jovanović & Lazić, 2020).

All in all, how satisfied are you with your current home? (S),

respondents were provided with “Not at all satisfied/Slightly satisfied/Moderately satisfied/Very satisfied/Extremely satisfied” as response options.

6.2 Home attachment

Based on the exploratory case studies in residential housing mentioned above (Velux, 2015), the measurement of home attachment comprises two items that capture residents’ affection to their home best. The stimulus question was:

How do you perceive your current home? Please indicate to what extent the following statements are true?

Respondents could choose from “Very untrue/Somewhat untrue/Neutral/Somewhat true/ Very true.” Specifically, the two attachment items were:

I feel at home where I live (A1)

I don’t really like to spend much time in my dwelling (A2).

6.3 Comfort

For evaluating the physical comfort in the home, a battery of statements was used that were to be scaled by respondents according to their assumed truthfulness, using the same format as above. The physical comfort items, presented in randomised order, were:

In my home, I can make full use of the daylight (C1)

The temperature in my home can easily be adjusted according to my needs (C2)

The sleeping conditions in my bedroom allow a restful sleep (C3)

My dwelling can easily be aired out (C4).

The functionality aspects were measured with the following four (negative) statements, also in randomised order:

My dwelling is in need of renovation (F1)

I sometime wonder if my home uses too much energy (F2)

I have a problem with mould in my home (F3)

My dwelling is too small (F4).

6.4 Climate zones and country groups

Attempting to generalising results over Europe, it is necessary to control for geographic differences. Can we distinguish particular regions in Europe that are characterised by typical residential evaluation patterns? The climate zones classification of regions is the obvious candidate for running this test. Climate zones do not respect country borders, so classifying respondents according to the climatic zones of their residency calls for matching sub-national topological units to the climate zones in Europe.

Numerous studies have shown that climate provides a statistically significant explanation of cross-country variations in measures of subjective wellbeing (Maddison & Rehdanz, 2011; see Parker, 1995, for an early review listing more than 2000 studies from sociology, psychology and physiology). In most of these studies, either revealed preference techniques for expenditures or the household production function approach to measure the compensating surplus (CS) of climate are used (Maddison, 2003). In the so-called hedonic approach, the marginal changes in climate variables are measured directly via individual satisfaction estimates (Maddison, 2014; Maddison & Bigano, 2003; Maddison & Rehdanz, 2011; Mendelsohn, 2014; Rehdanz & Maddison, 2005, 2009; Van der Vliert et al., 2004; Zapata, 2022). Confronted with climate change, this area of research has gained imperative importance.

Based on the most widely used bio-geographical classification of regions (European Environmental Agency, 2020), we make use of the regional information in the Healthy Homes Barometer and derive a climate zone classification of respondents. The classification involves the international codes III2, III3, and IV1 and makes use of the bio-geographical labelling of regions in Europe: Continental climate, Atlantic climate and Mediterranean climate.

We can also look at country differences instead of climate zones in order to test whether political instead of topological divisions make a difference (Phillips et al., 2022). Grouping the individual 14 countries of the study, a natural classification of countries of the European map would be “North” (Austria, Belgium, Denmark, Germany, Norway, Switzerland, the Netherlands, the UK), “South” (France, Italy, Spain), and “East” (Czech Republic, Hungary, Poland) (Beranova et al., 2017).

6.5 Model specification

The specified models are tested with structural equation modelling (SEM) techniques allowing to test alternative models against each other and to control for measurement error. We also investigate the parameter invariances of the studied models with SEM’s group comparison options (Brown, 2015; Jöreskog, 1971; Kline, 2016).

As exogenous latent constructs, the models highlight physical comfort in Model Ia and functional comfort in Model Ib. In Model IIa and Model IIb, attachment is a mediation process variable for physical as well as for functional comfort on satisfaction (see Fig. 1). In Models IIIa and IIIb, complete mediation via attachment is assumed, setting the direct effect of both comfort measures on satisfaction to zero. For physical comfort, there are four indicators (C1–C4), functional comfort has also four observables (F1–F4), and the measurement model of attachment has two indicators (A1, A2). The final dependent (endogenous) variable is the observed satisfaction measure S.

6.6 Goodness of fit statistics

According to Hu and Bentler (1999), a two-index combination of incremental and absolute fit indices is advisable for reporting SEM results.Footnote 3 In this paper, we go beyond Hu and Bentler’s suggestion and report the two most commonly used incremental indices, CFI and NNFI (also known as TLI for Tucker-Lewis index), along with two absolute indices, RMSEA and SRMR.Footnote 4 Model chi-square values and their degrees of freedom are presented as well, even though there are well known limitations in using them in large samples (Bentler & Bonnett, 1980; Jöreskog, 1993). For additional information, we also report the overall coefficients of determination (R2) for the whole model (Bentler & Raykov, 2000).

All models are estimated in Mplus 8 (Muthén & Muthén, 2017) applying the robust maximum likelihood estimation method procedure (MLR) in order to compensate for skewed distributions of observables. We report standardised regression coefficients, the specified goodness of fit measures, and indirect and total effects and their significance based on bootstrap estimation and the Sobel test.

6.7 Exact measurement invariance test and alignment optimisation

When comparing results for groups, differences in means and regressions can only be interpreted meaningfully if the underlying measurements are invariant. If measurement invariance is established, we can safely assume that the same constructs are measured in the same way and that question wording and other conditions do not affect the outcomes across different groups (Davidov et al., 2018; Meredith, 1993; Millsap, 2011).

There are several techniques for testing for measurement invariance. The most widely used procedure is multigroup confirmatory factor analysis (MGCFA), which is applied across different groups of participants. Depending on which types of coefficients are constrained to be equal, three invariance levels are usually differentiated: configural, metric and scalar measurement invariance (Davidov et al., 2014). Configural invariance is established when a particular latent construct is measured by the same items in all groups, regardless of loading sizes. For metric measurement invariance, the loadings are assumed to be equal across groups, and for scalar invariance, in addition to the constrained factor loadings, also the intercepts are constrained to equality. Metric invariance is necessary for comparing covariances and regression coefficients over groups or countries; scalar invariance is necessary for comparing latent means (Davidov et al., 2014).

The problem with MGCFA is that it leads almost always to rejection with regard to scalar measurement invariance, even if only partial invariance is tested (Asparouhov & Muthén, 2014), because the test strictly demands that all relevant group parameters are exactly equal.Footnote 5

In this paper, we will make use of the exact measurement invariance test with MGCFA across population groups. However, in as much as it is likely that exact measurement invariance is not supported by the data, we also make use of an alternative. If exact measurement invariance does not hold, forms of criteria relaxation have been suggested for performing meaningful comparisons (Byrne et al., 1989; Muthén & Asparouhov, 2013). The approach receiving the most attention presently is the alignment optimisation method (Asparouhov & Muthén, 2014; Cieciuch et al., 2018). Tests of cross-cultural scale invariance of place attachment measures using alignment are reported by Nartova-Bochaver et al. (2022), whereas Jones et al. (2017) and Boley et al. (2021) use MGCFA only.

As integrated into the Mplus 8 program, the alignment optimisation procedure estimates group means and variances without constraining loadings and intercepts of indicators to be equal across groups. The method provides a list of model parameters that are noninvariant. According to the simulation results of Muthén and Asparouhov (2014), a maximum of about 25% noninvariant parameters, averaged over loadings and intercepts, are tolerable to conclude that the means based on the alignment optimisation are trustworthy.

6.8 Sequencing of model tests

Based on the forgoing methodological considerations, we proceed with the following sequence of structural equation models and measurement invariance tests (Fig. 2):

Step 1: Tests of full versus partial SEM mediation models (I[a, b], II[a, b], III[a, b]) for the cross-country sample (total, N = 14,000)

Step 2: Multigroup SEM of the best fitting partial meditation models for the three climate zones and three country groups

Step 3: Exact measurement invariance tests (MGCFA) of the best fitting partial mediation models for the three climate zones and three country groups

Step 4: Approximate measurement invariance tests by alignment optimisation for the three climate zones and three country groups

Step 5: In conclusion, we test the multiple mediation model for the cross-country sample (total, N = 14,000) determining the influence of physical and functional comfort and attachment on residential satisfaction simultaneously.

Fig. 2
figure 2

Sequence of steps of analysis

7 Results

7.1 Model tests, cross-country sample

In Table 2, goodness of fit values of the structural equation models are reported. The subsequent section takes up the results of the group comparisons of climate zones and country groups by detailing the goodness of fit measures of the group models, testing also for measurement invariance with both the exact invariance measurement and the alignment optimisation method.

Table 2 Model fit indices (cross-country sample)

In Table 2 we first see that the CFI index indicates a satisfactory fit for the no-mediation physical comfort model (Model Ia) for the full sample but displays a relatively low R2 value.Footnote 6 This is not particularly surprising, however, because the predictive strength is limited to only one independent variable. For functional comfort in Model Ib, results are very similar.

The path estimates of the mediation models are presented in Table 3. Noteworthy are the strong effects of physical comfort on satisfaction (0.452) and of (negatively phrased) functional comfort (− 0.503), expressed in the form of standardised coefficients, in Model Ia and Ib, respectively.

Table 3 Standardised path estimates of mediation models (cross-country sample)

Of the models involving physical comfort as predictor, Model IIa is the most credible model, yielding the best overall goodness of fit indices CFI und TLI, well surpassing the conventional criterion values (Barrett, 2007; Bentler & Bonett, 1980; Hu & Bentler, 1999). The mediation process variable of attachment reduces the direct effect substantively to a coefficient of only 0.142 here. The next step leads to eliminating the direct effect in the hierarchical confirmatory analysis altogether, as is demonstrated in Model IIIa. While the CFI index is only minimally affected by this and TLI not at all, the overall R2 value increases notably, indicating a higher proportion of explained variance. From this we argue that home attachment tends to provide the complete mediation of physical comfort on satisfaction.

Functional comfort instead of physical comfort is inserted in Model IIb resulting also in a satisfactory, above threshold CFI index for partial mediation. The TLI index is notably lower, and the nonmediated direct effect changes from − 0.503 to − 0.305. If, however, the model is reduced to complete mediation (Model IIIb), both goodness of fit values fall below acceptable levels and large RMSEA values emerge. The conclusion here is that complete mediation does not fit the data for functional comfort. In contrast to physical comfort, functional comfort is only partially mediated by home attachment, leaving a substantial direct influence of functional comfort on satisfaction. Obviously, both forms of comfort sensations tend to stimulate different cognitions in the evaluation of our contentment in the home, thereby involving attachment to a different degree.

7.2 Climate zones and country groups differences

7.2.1 Exact measurement invariance tests using MGCFA

Next, we report the results of the multiple group structural equation models for the different mediation types of comfort, attachment, and satisfaction. The question that needs to be answered is whether the relationships that have been corroborated for the full sample hold also true for different population groups (Davidov et al., 2018). To this effect, we report climate zones and country group differences.

We look at the model fit of partial mediation, the most complete model, for the individual groups first (Models IIa and IIb). As documented in Table 4, the individual group analyses by and large indicate a satisfactory fit of the model with the group data. A fit slightly below standard is visible only in the functional comfort models in the Continental and Atlantic climate zones and in the Northern and Eastern country groups. Overall, after ruling out these exceptions, configural invariance of the geographic population groups is established.

Table 4 Model fit indices for partial mediation in climate zones and country groups

Following the group comparison options for MGCFA, we test possible hierarchical model restrictions employing the free-baseline strategy of letting all parameters be unconstrained in the beginning and then constraining groups of parameters step-by-step. We report results of the three most relevant levels of invariance: configural invariance, metric invariance and scalar invariance. The goodness of fit indices for the invariance models are then compared, and the observed changes in fit statistics guide us in selecting the valid form of invariance (Cheung & Rensvold, 2002).

From the MGCFA results presented in Table 5, we can see that with regard to the two topological groupings, climate zones and country groups, configural and metric invariance is given for the latent constructs in the physical comfort models. Scalar invariance, however, is only established for the country grouping. In contrast, for the functional comfort models, measurement invariance at all three invariance levels must be rejected. Trying to fit models with either configural, metric, or scalar measurement invariance fails as evidenced by the insufficient goodness of fit measures obtained for these models.

Table 5 Exact measurement invariance tests for climate zones and country groups

7.2.2 Alignment approximation of climate zones and country groupings

For estimating means of the latent constructs, we next apply the approximate measurement invariance test with the alignment method, as described in the method section above. For the three latent constructs, physical and functional comfort, and attachment, we find that there are only very few invariance violations for the climate zones and country groups on the level of metric measurement invariance (equal factor loadings), whereas violation of scalar invariance (equal intercepts) is more prominent (see Table 7 in the Appendix). In all, there are 22% noninvariant parameters; over the two types of groupings, this includes 12% noninvariant loadings and 31% noninvariant intercepts. These results stay well below the cut-off criterion of 25% noninvariant parameters, averaged over loadings and intercepts, that are tolerable in order for means based on the alignment optimisation to be trustworthy.

Thus equipped, we turn our attention to the factor means in the climate zones and country groupings. In Table 6, for highlighting statistical differences, the means are arranged in descending order. We find that home attachment is highest in the Continental climate zone and in the South; physical comfort is lowest in the Atlantic zone as well as in the East and South; and functional comfort scores highest in the Atlantic zone and in the Eastern countries. For satisfaction, the dependent variable, the values are the highest for the Continental climate zone and the Northern countries. If nothing else, these results demonstrate that the three latent constructs capture very different concepts and realities, giving credence to the differences in the mediation of the two comfort types on satisfaction.Footnote 7

Table 6 Factor means comparison in descending order

7.2.3 Multiple mediation model

After analysing the influence of physical and functional comfort on residential satisfaction independently, we are now in a position to conclude our analysis with a summative model of the full sample across all groupings: the multiple mediation model. The model includes both comfort types, with physical comfort (C) fully mediated by attachment (A) in determining satisfaction (S), and with functional comfort (F) partially mediated through attachment but also affecting satisfaction directly. This can be formalised as S = f (F, (A ← F), (A ← C)) and is presented graphically in Fig. 3.

Fig. 3
figure 3

Multiple mediation causal model with standardised estimates (S = satisfaction, A = attachment, C = physical comfort, F = functional comfort)

This figure also includes standardised parameter estimates. Using the same robust estimation procedure as in the previous models, the model fit yields a chi-square value of 1026.876 (df = 37), CFI = 0.957, TLI = 0.937, RMSEA = 0.044, and SRMR = 0.027; R2 = 0.868. To mirror possible applications, residual error correlations between the functional comfort items F1–F3, that involve possible home renovation efforts, were allowed. Table 8 in the Appendix summarises the total direct and indirect effects of the comfort types on satisfaction.Footnote 8

8 Discussion

User-oriented architectural planning and post-occupancy evaluations usually rely on comfort measurements for determining residential satisfaction. An alternative predictor for the level of satisfaction, however, is the assessment of home attachment because the sense of home serves as a prerequisite for physical or functional comfort to shape our level of residential satisfaction. In terms of our model specification then, home attachment is a mediator variable for the comfort varieties that prompt satisfaction. Using recent European cross-country survey data, this proposition is brought to a straightforward test by comparing models of complete, partial, or no mediation of comfort in explaining residential satisfaction.

We find that physical comfort contributes to satisfaction via home attachment, whereas functional comfort affects satisfaction also directly, in other words: physical comfort is fully mediated by attachment, functional comfort only partially. Thus, referencing Kahneman’s (2013) distinction of fast versus slow thinking and Gibsonian perception theory, physical comfort turns out to be an element of system 1 experience, functional comfort is an element of system 2. This is so because attachment, mediating physical comfort fully, is an impulsive reaction whereas functional comfort stimulates a reflective response and as such affects satisfaction directly. This finding is well in line with recent attempts to conceptualise residential satisfaction within the means-end chain approach employing hierarchical value maps (Coolen, 2011; Coolen & Hoekstra, 2001; Sadeghlou & Emami, 2023). According to this approach, residential satisfaction is mediated by housing preferences that are motivated by the expected consequences of choosing a house as well as the realisation of abstract values—clearly a reflective process of Kahneman’s system 2 variety.

Explaining residential satisfaction, we therefore have a two-process situation characteristic of many cognitive and behavioural phenomena (Chaiken & Trope, 1999; Evans, 2010; Evans & Frankish, 2009; Strack & Deutsch, 2004). Though the causal analysis put forth in this article is clearly limited in scope as it utilises only a selective set of both types of comfort variables, it can well serve as a baseline model for delving more deeply into the duality of the causal dynamics of residential satisfaction.

Besides this theoretical outlook, residential satisfaction is also an important yardstick for the definition of comfort norms in architectural design. The engineering of comfort is guided by numerous standards and recommendations in national and international legislation the building industry is obliged to follow. But if comfort is mediated by home attachment in determining residential satisfaction, as the here reported research suggests, the simple comfort rules of these statutes are no longer appropriate. In order to generate satisfaction, the rigid physical and functional comfort norms need to be contextualised and transformed into adaptive comfort standards following recent developments in the measurement and implementation of thermal comfort (Halawa & van Hoof, 2012; Nicol et al., 2012; Kazanci et al., 2019; Nicol et al., 2020). Adaptive comfort rules take into account that perceptions, acceptability, and preferences vary depending on building characteristics and environmental conditions and therefore call for different comfort norms for different situations.

Situational differences need also be considered in view of home attachment. However, in spite of general agreement in place research that home is the prototypical place and that we are “domicentric” individuals (Porteous, 1976), the overwhelming quantity of research that deals with place attachment concerns attachment to the neighbourhood, not home (Giuliani, 2003; Lewicka, 2011). Moreover, as Hidalgo and Hernandez (2001) have shown, replicated by Lewicka (2010), there is an U-shaped relationship between place levels (home, neighbourhood, city) and strength of place attachment, home and city evoking the strongest attachment, neighbourhood the least. The search for predictors of these different attachment forces however overemphasises individual and social properties, whereas differences between places and the physical features of places receive less attention (Droseltis & Vignoles, 2010). Therefore, place attachment research is still in need of a theory of place that identifies causal factors for the development of emotional bonds in particular types of places (Lewicka, 2011).

Regarding the home environment in particular, researchers may want to return to the beginnings of place research in human geography and its roots in the phenomenological tradition (Merleau-Ponty, 1962; Norberg-Schultz, 1979; Graumann, 2002; Cresswell, 2006; Seamon, 2021). The claim of phenomenologists is that the descriptions of places, although subjective, are not idiosyncratic impressions of individual experiences, but reveal universal properties, giving places meaning and emotional value. In this tradition, the work of architectural theorist Christopher Alexander is pertinent (Alexander, 2003; Alexander et al., 1977). In his seminal book The timeless way of building, Alexander (1979) describes a number of architectural factors that should transform houses into homes, and hence generate attachment. Alexander’s order principles of liveable homes include balancing the public and the private, ownership status, size of buildings and number of floors, windows that give pleasing views, openness of access, and an abundance of other design principles. A theory of place attachment could benefit from taking these building and environmental features into account, testing to what extent they predict home attachment (Lewicka, 2011, p. 223; Iwańczak & Lewicka, 2020).

9 Concluding note

A research programme that is directed towards differences in home attachment and the adaptive norm setting of comfort rules needs to be based on causal reasoning and empirical tests of the kind illustrated in this paper. But there are at least five issues future research might improve on:

  1. (1)

    The number and complexity of contributing factors to the emergence of residential satisfaction should be extended including levels of place attachment based on a taxonomy of types of buildings and residential settings.

  2. (2)

    Also, differences in lifestyles should be considered. A certain level of satisfaction may well be contingent on particular objective features of the environment and still result in dissatisfaction. Therefore, lifestyle predictors may undercut the predictive power of physical attributes on satisfaction (Jansen, 2011). Future studies should provide to control for these variables.

  3. (3)

    As is quite obvious, more comfort measures of different physical modalities and functions should be added.

  4. (4)

    Causal effect differences between population groups and regions need to be accounted for.

  5. (5)

    Finally, it is also an option to design intervention studies to systematically vary comfort dimensions and attachment factors in quasi-experimental field studies.

But even without these extensions, the here reported research demonstrates already that comfort (physical and functional), home attachment, and residential satisfaction are the key elements of a core theory to explain our wellbeing at home.