Gravity models and analogue store approaches are inadequate in predicting purchases in neighbourhood stores. This requires a new theoretical and empirically tested approach.
We use the Theory of Planned Behaviour (TPB) to determine which factors predict the choice for a new neighbourhood store. We develop a suitable model using a structural equation model with survey data from two cases in which all households in the catchment areas were surveyed both before and after the store opened.
We find the TPB to be appropriate for predicting store choice. Beliefs about one-stop shopping, social pressure from family members and car availability are most important in explaining the intention to shop in the planned store. These factors also explain the actual shopping in this store after opening.
Our model predicts store choice before a store opens. Using a two-wave survey, we avoid ex-post rationalisation and show that, at least in our cases, quality, price and assortment do not predict store choice.
1. Introduction
The success of a new store depends mainly on its location and whether consumers in the projected catchment area are willing to shop there. It is thus crucial for grocery retailers, on the one hand, and spatial planners, on the other, to forecast the amount of groceries local customers would buy in a new store. Retail research and practice lacks theoretically based models that can be used to predict consumers' grocery store choices before stores are established. In this article, we contribute to the theoretical basis for forecasting the use of neighbourhood stores, defined as small grocery stores, in a new location.
In practice, two forecasting methods are commonly used to analyse store location choices (Wood and Browne, 2007): spatial interaction models, which are based on reductionist models, and the analogue stores approach, which is an atheoretical, data-driven approach. The spatial interaction models—mostly variations of Huff’s (1963) gravitation model—estimate flows of purchasing power on a macro level. These models are deployed to understand consumer’s choice between city centres or shopping malls (Khawaldah et al., 2012). They can also project shopping activity at grocery stores with a sales area of more than 3,000 m2 (Wood and Browne, 2007), as these store formats can act as footfall generators and thus fulfil the model’s underlying assumptions. The spatial interaction models use “objective” indicators for (car) accessibility and attractiveness, despite the fact that the subjective perception of store characteristics is crucial for consumer behaviour (Maruyama and Wu, 2014). Previous studies have shown that driving time has become less important, especially for smaller stores since more customers (spontaneously) go for top-up shopping, combine their grocery shopping with other activities, and regularly buy at several stores (Wilkinson, 2023; Wood and Browne, 2007). Given these particularities of grocery stores and small stores, modelling store choice based on spatial interaction models results in less accurate forecasts.
The analogue stores approach employs data from stores similar to the one planned in order to estimate the potential market share in the catchment area. This approach uses information on demographics or distance from consumers’ home to the store (Khawaldah et al., 2012; Wood and Browne, 2007). This approach is based either on customer survey data collected for all stores in the portfolio of the retail chain, or on average turnover (per square metre) of the respective store format. Ideally, catchment area demographic indicators and micro and macro location characteristics are used to compare the planned store with existing stores and/or to estimate the new store’s revenues using regression models. However, empirical findings suggest that store choice is not dependent on age, gender, income, or consumption behaviour (Goodman and Remaud, 2015). Moreover, data from other stores often cannot be transferred to new stores because specific store qualities (e.g. the competition in the local market or neighbouring footfall generators) interfere with average effects, which is particularly relevant for small stores (Fildes et al., 2022; Wood and Browne, 2007).
Given that both approaches are geared towards making predictions for larger stores and/or draw on experiences from existing stores, they are only partially suitable to predict who will buy how much at a planned neighbourhood store. Thus, to date, this market segment lacks a theory-driven model that can forecast consumer choice for a new neighbourhood store before it opens. In this case, location decisions and business plans are dominated by intuition in practice (Hernández and Bennison, 2000; Pioch and Byrom, 2004). A theory-driven model would not only improve our scientific understanding of store choice (Runyan and Droge, 2008), but could also be used in practice to help retailers improve their sales forecasts for planned stores and identify approaches to increase sales. In addition, the public sector could enhance planning processes with more realistic needs assessments.
An issue faced by existing empirical studies on store choice is ex post rationalisation. Studies in the field tend to employ ex post facto designs with customer surveys— typically at the point of sale—trying to link store choice with perceived store attributes and satisfaction with the shopping experience (Goić et al., 2021; Goodman and Remaud, 2015; Yokoyama et al., 2022). If respondents only justify their current behaviour, then cause and effect cannot be separated properly. For this reason, Bonfrer et al. (2022) conclude in their review paper that longitudinal designs and structural models are needed to address this methodological issue. The appropriateness of the Theory of Planned Behaviour (TPB) to solve such problems has already been empirically proven multiple times (Ajzen, 1991). The TPB is a well-established theory for predicting actual behaviour and identifying factors that interventions can build upon to change behaviour.
To better understand the factors that predict the choice of using a new neighbourhood store, in Section 2 we build a theoretical model based on the TPB, which integrates potential impact factors on store choice identified in the retail literature. We empirically test our model with two-wave surveys in two case study areas in Germany: the first wave collects data on the potential impact factors before a neighbourhood store is established, while the second wave focuses on the actual behaviour after the opening (Section 3). By using a structural equation model, we can determine which factors can predict store choice and to what extent, and which factors proposed in the literature do not contribute to the prediction (Section 4). We discuss the findings in Section 5 and conclude with the implications for research and practice in Section 6.
2. Applying the Theory of Planned Behaviour to store choice
The Theory of Planned Behaviour (TPB) is an integrative model for predicting (and changing) human social behaviour (Ajzen, 1991; Fishbein and Ajzen, 2010). Although human behaviour (such as store choice) is complex, the TPB claims that only a few determinants explain behaviour. Accordingly, behavioural beliefs form the attitude towards the behaviour, normative beliefs shape the subjective norm, and control beliefs result in the Perceived Behavioural Control (PBC). Attitude, subjective norm, and PBC, in turn, explain the intention to perform the behaviour. Intention and PBC determine the actual behaviour. All other factors, such as demographics, habits, or identities are considered to be background factors and may influence the beliefs, but are not required for the model. Moreover, a meta-analysis and a recent review demonstrate that socio-demographics have only a marginal impact on retail patronage (Pan and Zinkhan, 2006) or store format choice (Bonfrer et al., 2022) respectively.
TPB has been widely used in retail research to explain online grocery shopping behaviour (Zhao et al., 2023; Zhou and Liu, 2022), store format choice (Anand and Sinha, 2009), or local and ethical shopping behaviour (Mullis and Kim, 2011). Since TPB has not yet been applied to store choice, we specify the theoretical model for explaining store choice (Figure 1) in this section by using existing literature related to store choice.
The diagram maps factors influencing new store shopping. On the left, three gray boxes list belief categories: “Beliefs about store’s attributes: quality, price level, products on sale, service, meeting place, one-stop shopping, trip chains, proximity, assortment, and local products,” “Beliefs about social referents: family, friends, and neighbours or other locals,” and “Beliefs about control factors: car availability, health, financial constraints, time sovereignty, and shopping together”. Each box points right to a blue hexagon: “Attitude,” “Subjective norm,” and “Perceived behavioural control”. All three hexagons point to a green oval labeled “Intention to shop in the new store,” and arrows from this oval—and directly from “Perceived behavioral control”—lead to another green oval, “Actual shopping behaviour”.Theoretical model that applies TPB to store choice
The diagram maps factors influencing new store shopping. On the left, three gray boxes list belief categories: “Beliefs about store’s attributes: quality, price level, products on sale, service, meeting place, one-stop shopping, trip chains, proximity, assortment, and local products,” “Beliefs about social referents: family, friends, and neighbours or other locals,” and “Beliefs about control factors: car availability, health, financial constraints, time sovereignty, and shopping together”. Each box points right to a blue hexagon: “Attitude,” “Subjective norm,” and “Perceived behavioural control”. All three hexagons point to a green oval labeled “Intention to shop in the new store,” and arrows from this oval—and directly from “Perceived behavioral control”—lead to another green oval, “Actual shopping behaviour”.Theoretical model that applies TPB to store choice
2.1 Attitude towards shopping in a store
People’s attitude towards behaviour is determined by their beliefs about the attributes of the behaviour. According to the expectancy-value model (Fishbein and Ajzen, 1975), which is one of the foundations of the TPB, a belief about an attribute is the product of the subjective probability that a behaviour has a certain attribute and the evaluation of that attribute. The attitude, then, is the aggregation of the beliefs over all attributes. Relevant attributes can be drawn from the literature on store choice and customer satisfaction, as will be shown in the next paragraph.
In so far as store choice is concerned, scholars have traditionally highlighted price, service quality, range of goods, and location as important dimensions (Agergård et al., 1970). Low prices and the availability of products on sale represent the first dimension (Goodman and Remaud, 2015; Miller and Kean, 1997; Nair, 2018). The second dimension relates to service quality and consists of product quality and, more recently, the availability of a range of local products (Broadbridge and Calderwood, 2002), as well as staff friendliness (Faria et al., 2022; Goodman and Remaud, 2015). The third dimension comprises the assortment and the opportunity for one-stop shopping (Broadbridge and Calderwood, 2002; Goodman and Remaud, 2015; Sethuraman et al., 2022). The location of a store represents the fourth dimension and includes proximity to the home and the ease of trip chains (Bonfrer et al., 2022; Huff, 1963; Marshall, 2019). Beyond these dimensions, the social function of stores as meeting places or “third places” (Oldenburg, 1999) reflects the opportunity to encounter friends and acquaintances – a highly relevant aspect of rural life (Broadbridge and Calderwood, 2002; Miller and Kean, 1997; Oeser et al., 2019; Satyam et al., 2022). Because the beliefs about such attributes of a store are only weakly correlated with store choice (cf. Liu-Thompkins et al., 2022 for retailer loyalty), the TPB adds subjective norm and PBC to attitude.
2.2 Subjective norm
In the TPB, the subjective norm is determined by the total set of normative beliefs about the behaviour. These beliefs result from the expectation of important referents multiplied with the motivation to comply with the respective referent (Ajzen, 2012). This perceived social pressure to perform a certain behaviour (Ajzen, 1991) was also observed in store choice (Marshall et al., 2018, p. 141) and depends on a person’s level of involvement in the neighbourhood or village community (Miller and Kean, 1997; Addis and Grünhagen, 2014). In addition to neighbours, it can be assumed that family members and friends are also important referents (Zhou and Liu, 2022). If the expectations of these groups are important to a person, and this person perceives the expectation from such groups to buy in a newly built store, this combination (subjective norm) leads to the intention to choose the store.
2.3 Perceived Behavioural Control (PBC)
The PBC refers to people’s perception on the extent to which they have both the required resources and the control over performing a behaviour (Fishbein and Ajzen, 2010). Analogous to the definition of attitude, PBC summarises control beliefs that are the degrees to which each control factor enables or prevents the performance of the behaviour, weighted by the subjective probability that the control factor exists (Ajzen, 2012).
Factors enabling or preventing the choice of a new store include car availability, health condition, time constraints, financial situation of the household, and shopping together with other persons (Bonfrer et al., 2022; Marshall et al., 2018; Mullis and Kim, 2011). However, whether these control factors enable or prevent the choice of the local store is often not pre-determined. Poor health, for instance, may prevent customers from driving or taking the bus to grocery stores elsewhere and thus determine the use of the local neighbourhood store. Moreover, poor health may prevent people from carrying groceries home, leading them to rely on family members or neighbours to drive them to a store other than the local neighbourhood store. Therefore, it is important to consider whether the control factors are perceived to be present and whether they enable, prevent, or have no influence on the store choice of a certain person.
2.4 Intention and behaviour
In the TPB, attitude, subjective norm, and PBC explain the behavioural intention. Intentions stated in surveys are often biased by social desirability, which is also the case here (see subjective norm). Meta-studies show that self-prediction, measured by the subjective likelihood of performing a behaviour, correlates more strongly with actual behaviour than asking people about their desires or intentions (Armitage and Conner, 2001; Sheppard et al., 1988). In addition, the satisfaction with the status-quo of stores can serve as an indirect indicator for the intention to buy in a planned store, given that shopping satisfaction is a very good predictor of customer loyalty (Nam et al., 2011; Loureiro et al., 2014).
When applying the TPB on store choice, it is important to operationalize the respective behaviour. In the literature, researchers often asked whether or not a store was used (Pan and Zinkhan, 2006). However, evidence shows that customers use several stores. Smaller stores, such as neighbourhood stores, are often used only for top-up shopping with minor expenditures (Marshall et al., 2018; Yokoyama et al., 2022). Therefore, measuring the actual behaviour must account for the amount of groceries bought in the new store.
3. Data and methods
To empirically test our theoretical model, we selected two new neighbourhood store openings in rural Germany: one in the Bavarian village of Lützelburg and one in the Lower-Saxon village of Müden (see Supplementary_material_appendix_1.pdf). The selected store in Lützelburg is a community store established and funded by local citizens. The store’s sales area is about 120 m2 and the catchment area encompasses the 1,700 inhabitants of Lützelburg. The store in Müden is owned by a local entrepreneur in cooperation with a large retail group and is financially supported by local citizens. The store’s sales area is about 300 m2 and supplies 2,400 inhabitants in its catchment area, which comprises Müden and the adjacent village of Poitzen (200 inhabitants). Both stores were opened in 2019 and were subsidized by the European Agricultural Fund for Rural Development (EAFRD). Due to closures in previous years, these are the only grocery stores in their respective catchment areas. The rural contexts clearly delimitate the catchment areas and allow all potential customers to be surveyed. Comparing the two cases also enables us to draw conclusions that go beyond the local and conceptual specifics.
We employed a two-wave survey design for both case studies and collected data before and after the opening of the stores using postal household surveys. The person of the household who is usually responsible for the grocery shopping was asked to fill out the questionnaire. We used the tailored design method to develop and conduct the survey (Dillman et al., 2014). We operationalized the constructs of the TPB presented in the previous section by applying example questions from Fishbein and Ajzen (2010). Several pre-tests were conducted with experts for content, survey method as well as statistics, followed by cognitive interviews with potential respondents. The questionnaires were then tested at another store opening with more than 500 completed questionnaires. The final questions and items used are shown in Table 1.
Operationalisation of constructs and descriptive statistics in both study areas
| Construct | Item/Question | Mean | Median | SD | N |
|---|---|---|---|---|---|
| Evaluation of attributes | Many reasons can play a role in choosing a store. How important are the following attributes to you when choosing a store for purchasing everyday goods? Please answer the question using a scale with values from 0 to 6. A value of 0 means that these attributes are “not at all important” to you, a value of 6 means that they are “very important” | ||||
| Importance of quality | High quality of goods | 5.0 | 5 | 1.1 | 744 |
| Importance of price level | Low prices | 4.0 | 4 | 1.4 | 746 |
| Importance of products on sale | Products on sale | 3.5 | 4 | 1.7 | 743 |
| Importance of service | Friendly staff | 5.0 | 5 | 1.3 | 749 |
| Importance of meeting place | Good opportunity to meet friends and acquaintances | 2.2 | 2 | 1.8 | 741 |
| Importance of one-stop shopping | Possibility to shop everything in one place | 4.3 | 5 | 1.6 | 750 |
| Importance of trip chains | Possibility to combine shopping with other activities (e.g. commuting, doctor’s visit, leisure activities) | 3.1 | 3 | 2.0 | 749 |
| Importance of proximity | Proximity to the place of residence | 4.8 | 5 | 1.4 | 750 |
| Importance of assortment | Wide range of goods | 4.2 | 4 | 1.4 | 747 |
| Importance of local products | Local products | 4.6 | 5 | 1.4 | 732 |
| Probability of attributes | How likely do you think it is that the neighbourhood store [name of the village] will have the following attributes? Again, please use the scale with the values from 0 to 6. The value 0 means that you think it is “not at all likely” that the new store will have these features, the value 6 means that you think it is “very likely”. You can use the values in between to grade your estimate | ||||
| Probability of quality | High quality of goods | 4.7 | 5 | 1.1 | 754 |
| Probability of low prices | Low prices | 2.8 | 3 | 1.4 | 755 |
| Probability of products on sale | Special offers | 3.0 | 3 | 1.5 | 754 |
| Probability of good service | Friendly staff | 4.9 | 5 | 1.2 | 751 |
| Probability of meeting place | Good opportunity to meet friends and acquaintances | 3.8 | 4 | 1.7 | 747 |
| Probability of one-stop shopping | Possibility to shop everything in one place | 3.2 | 3 | 1.9 | 751 |
| Probability to combine paths | Possibility to combine shopping with other activities (e. g., commuting, doctor’s visit, leisure activities) | 2.1 | 2 | 2.0 | 751 |
| Probability of proximity | Proximity to the place of residence | 5.4 | 6 | 1.2 | 755 |
| Probability of assortment | Wide range of goods | 3.0 | 3 | 1.5 | 752 |
| Probability of local products | Local products | 4.8 | 5 | 1.3 | 754 |
| Perceived social pressure | When choosing a store, people are sometimes guided by the expectations that other people have of them. How important is it to you that you meet the expectations of the following people? (response scale from 0 “not at all important” to 6 “very important”) | ||||
| Importance of family’s expectations | My family | 3.3 | 4 | 2.3 | 754 |
| Importance of friends' expectations | My friends | 2.0 | 2 | 1.9 | 747 |
| Importance of other locals' expectations | The other locals | 1.6 | 1 | 1.9 | 749 |
| Expectation of referents | And how likely do you think it is that the following people will expect you to buy your everyday goods mainly at the new store? Again, please use the scale with the values from 0 to 6. 0 means that you think it is “not at all likely” that these people expect you to shop in the new store, 6 means that you think it is “very likely” that they expect you to shop in the new store | ||||
| Probability of family’s expectations | My family | 2.4 | 2 | 2.1 | 752 |
| Probability of friends' expectations | My friends | 2.1 | 2 | 1.9 | 749 |
| Probability of other locals' expectations | The other locals | 3.0 | 3 | 2.1 | 750 |
| Probability of control factors | The following statements describe situations that may influence your choice of a store. For each statement, please indicate how often you find yourself in that situation. Again, please use the frequency scale with values from 0 to 6. A value of 0 means that you are “never” in this situation, the value 6 means that you are “always” in this situation | ||||
| Probability of car availability | I have a car available for my shopping | 5.5 | 6 | 1.4 | 755 |
| Probability of health | My health restricts me from moving around or carrying shopping bags | 1.0 | 0 | 1.7 | 751 |
| Probability of financial constraints | I have very little money for shopping | 1.9 | 2 | 1.9 | 750 |
| Probability of time sovereignty | I can determine the time for shopping myself | 5.1 | 6 | 1.4 | 754 |
| Probability of shopping together | I shop together with other people | 1.2 | 0 | 1.7 | 751 |
| Facilitating or inhibiting effect of control factors | And to what extent would these situations make you more likely or unlikely to shop at the new store? (response scale: 1 = more likely, 0 = no impact, -1 = more unlikely) | ||||
| Importance of car availability | I have a car available for my shopping | 0.1 | 0 | 0.5 | 746 |
| Importance of health | My health restricts me from moving around or carrying shopping bags | 0.4 | 0 | 0.6 | 737 |
| Importance of financial constraints | I have very little money for shopping | −0.2 | 0 | 0.6 | 730 |
| Importance of time sovereignty | I can determine the time for shopping myself | 0.2 | 0 | 0.6 | 741 |
| Importance of shopping together | I shop together with other people | 0.1 | 0 | 0.5 | 732 |
| Probability of shopping in the new store | How likely is it that you will buy your daily goods mainly in the new neighbourhood store in [village]? Again, please use the scale with the values from 0 to 6. The value 0 means that you consider it “not at all likely” to shop mainly in the new store, the value 6 means, that you think it is “very likely” | 3.5 | 4 | 1.7 | 753 |
| Satisfaction with shopping facilities | How satisfied are you at present, all in all, with your shopping opportunities for daily goods? Please use the scale with the values from 0 to 6. 0 means that you are “not at all satisfied” with your shopping opportunities for daily goods, the value 6 means that you are “completely satisfied” | 3.8 | 4 | 1.6 | 753 |
| Actual shopping behaviour (after the opening of the new store) | How many of your daily goods do you usually buy at the following retailers? Please use the scale with the values from 0 to 6. The value 0 means that you buy “nothing” at a retailer, the value 6 means that you buy “everything” there | ||||
| Purchase in the neighbourhood store | Neighbourhood store [name] | 2.9 | 3 | 1.4 | 605 |
Source(s): Authors’ own work
The survey of all households in the catchment areas started three months before the opening of the store in Lützelburg and two months before the opening in Müden. Because both stores were financed with citizens’ capital, we assumed that the local population would be aware of the topic of the survey and that the willingness to answer the questionnaire would be correspondingly high. The second wave was conducted eight (Lützelburg) and twelve (Müden) months after the opening, allowing households to develop new shopping routines in accordance with the new offering. We surveyed all 655 households in Lützelburg and received 319 filled out questionnaires in the first and 263 in the second wave (response rate = 48.7% resp. 40.2%). In Müden, we distributed 1,029 questionnaires, of which 431 were returned in the first and 361 in the second wave (response rate = 41.9% resp. 35.1%). Using a code generated by the respondents themselves, we were able to link 304 questionnaires from the first and second wave (127 in Lützelburg and 177 in Müden). We thus connected beliefs about attitudes, subjective norms, and PBC as well as shopping intention from the first wave with the actual behaviour (self-reported) in the second wave at a microlevel.
To analyse the collected data, we used a structural equation model (SEM), which is the common method for testing the TPB (Fishbein and Ajzen, 2010). SEM tests whether a theoretical model fits the actual empirical data (Byrne, 2010). In addition, SEM allows the use of latent variables and the estimation of direct and indirect effects considering mediation processes. Further, we applied a multigroup analysis to integrate the data sets of both case studies. This allowed us to test the reliability of our measurement constructs and estimated effects (Byrne, 2010; Urban and Mayerl, 2013).
The assumptions for standard SEM with maximum likelihood estimation are not met as the variables are not multivariate normally distributed (Lei and Wu, 2012). Therefore, we employed a Bayesian SEM implemented via a Markov Chain Monte Carlo (MCMC) algorithm (Metropolis) to our data, as provided by the AMOS software (Arbuckle, 2016). Such estimations provide further advantages when dealing with relatively small sample sizes and missing data (Lee, 2007; Song and Lee, 2012).
The standard TPB uses a summative index to aggregate the beliefs (subjective evaluation*probability of an aspect of the behaviour – see Supplementary_material_appendix_2.pdf) in order to indirectly measure attitude, subjective norm, and PBC (Fishbein and Ajzen, 2010). In contrast, we built composites (Grace and Bollen, 2008) with the beliefs as suggested by Hennessy et al. (2012). This enabled us to include only those beliefs that have a significant positive effect on the intention to buy in the new store in at least one case study. Furthermore, we were able to estimate the effect of each belief on both the intention and on the actual behaviour by calculating the indirect effects. Using the multigroup analysis, we could also test whether the estimates differ between the cases. Therefore, we iteratively compared the fit (determined by DIC) of models that estimate each coefficient for each group (case) separately with those models that estimate the coefficient common to both groups (Song and Lee, 2012, p. 143). This has led us to choose the model with the best fit.
4. Results
In what follows, we present the results of our empirical analysis in two steps. First, we describe the two samples and test the measurement models in preparation for building the structural model. This step led to the elimination of beliefs irrelevant for explaining behavioural intention. Second, we present the results of the structural model and show the effects of the different factors on intention and behaviour.
4.1 Descriptive statistic and preparation for modelling
Starting with demographics, the average age of the respondents is 58 years in both study areas, with more than a third being retired (approx. 36% in both areas). 60% of the respondents in Lützelburg and 65% in Müden are women, indicating that they are more often responsible for purchasing daily goods in the household and were therefore asked to answer the questionnaire. On average, 2.6 persons live in a household in Lützelburg and 2.2 persons in Müden. Concerning the financial situation, the median income of the responding households in Lützelburg lies in the class between 3,000 to 3,500 Euro; in Müden the median income is situated in a lower class between 2,500 to 3,000 Euro.
Table 1 shows the descriptive statistics of the items relevant for the model. Concerning the evaluation of different attributes of the grocery stores, the data shows that quality, service, and proximity are most important. The expected attributes concerning the new store reflect the general perception of small stores (Jürgens, 2020; Yokoyama et al., 2022): convenient location, but a high price level. With regard to subjective norm, respondents mostly anticipated that other locals expect them to shop in the new store, while this group is least important as a reference for their buying behaviour compared to family and friends. The queried control factors are either present or absent, with little gradation in between. Moreover, most respondents answered that the factors neither facilitate nor inhibit them to shop in the new store. Respondents most strongly expect that personal health restrictions would lead them to shop locally. Even though the responding consumers tend to be more satisfied with their shopping opportunities, most respondents answered that using the new store mainly for their daily shopping is rather probable. However, the survey carried out after the store’s opening reveals that, on average, less than half of daily goods are bought in the new store.
For the structural equation model, we used the collected data to calculate the variables measuring beliefs in relation to attitude, subjective norm, and PBC (as described in Section 2). The descriptive statistics for all variables used in both cases are presented in Supplementary_material_appendix_1.pdf. In preparation for the structural modelling, we tested the validity of our latent variables. Firstly, we tested the measurement model for intention and actual behaviour ( Supplementary_material_appendix_3.pdf). The results indicate a good model fit (posterior predictive p = 0.49), sufficient standardized factor loadings (>0.4), and the possibility to restrict loadings, error variances, and factor variances among the two groups (Urban and Mayerl, 2013, pp. 224–225). Secondly, we evaluated the indicators of the composites by calculating the model without the composites. We iteratively eliminated those indicators with no positive and probable (90%-CI ∉ 0) effect on the intention (external validity according to Bollen, 2011, pp. 361–362). This applied to the attitudinal beliefs about quality, price level, products on sale, trip chains, and assortment. Concerning the subjective norm, the variable “friends” was also eliminated for this reason, but for the PBC all variables were included in the model.
4.2 Modelling the shopping intention and choice of the new store
The structural equation model allows us to test our theoretical model based on TPB and literature related to store choice (see Section 2). In addition, we can compare both case study areas using a two-group model, which shows the effects that apply to only one or both cases. The result is depicted in Figure 2. Posterior predictive p amounts to 0.50 and indicates a good fit of the theoretical model to the data. The model shows that attitude, subjective norm, and PBC explain the intention to shop in the new store to a large extent in both cases (R2 = ca. 0.7 and 0.6). The intention, in turn, also explains the actual shopping behaviour in the new store very good; in Müden even better than in Lützelburg. However, the PBC does not contribute to the explanation of the actual behaviour, which was to expected according to the TPB.
The diagram visualizes drivers of shopping intentions and actual purchasing behavior in a new store context. On the top left, five vertically stacked gray boxes labeled “Service,” “Meeting place,” “One-stop shopping,” “Proximity,” and “Local products” connect to a hexagon labeled “Attitude” at the center by arrows with standardized coefficient values: “0.792 over negative 0.199 and (0.138 over negative 0.107),” “0.198 and (0.081 over 0.107),” “1 and (0.468 over 0.633),” “0.629 and (0.269 over 0.332),” and “0.529 and (0.216 over 0.293),” respectively. On the middle left, two vertically stacked gray boxes labeled “Family” and “Other locals” connect to a hexagon labeled “Subjective norm” at the center by arrows with standardized coefficient values: “1 and (0.833 over 0.633)” and “0.542 over 1.044 and (0.244 over 0.486),” respectively. On the bottom left, five vertically stacked gray boxes labeled “Car availability,” “Health,” “Financial constraints,” “Time sovereignty,” and “Shopping together” connect to a hexagon labeled “Perceived behavioural control” at the center by arrows with standardized coefficient values: “1 and (0.491 over 0.458),” “0.035 over 1.463 and (0.008 over 0.320),” “1.334 and (0.325 over 0.342),” “0.911 and (0.389 over 0.408),” and “1.193 over negative 0.453 and (0.190 over negative 0.089),” respectively. All three hexagons point to a green oval labeled “Intention” by arrows labeled “0.030 and (0.470 over 0.356),” “0.019 and (0.175 over 0.259),” and “0.096 and (0.410 over 0.397),” respectively. Arrows (“0.497 over 0.658 and (0.645 over 0.658)” and “negative 0.011 and (negative 0.062 over negative 0.052)”) from this oval—and directly from “Perceived behavioral control”—lead to another green oval, “Actual shopping behaviour”. Additional arrows branch to indicators: from “Intention” to “Probability of shopping in the new store” and “Satisfaction with shopping facilities”; from “Actual behaviour” to “Purchasing quantity” and “Purchasing share,” all with respective coefficients. The ovals “Intention” and “Actual behaviour” are also labeled “R-squared equals 0.708 over 0.587” and “R-squared equals 0.372 over 0.525,” respectively. Below, the notes read, “Values on the left for Lützelburg, right for Müden; One unstandardised effect: equality constraint across groups; Standardised coefficients in parentheses; cap equals unlikely effect (90 percent credibility interval includes 0); and Model fit: convergence statistic less than 1.005; posterior predictive p equals 0.50”.Estimated model to explain actual shopping in the new store with factors collected before its opening
The diagram visualizes drivers of shopping intentions and actual purchasing behavior in a new store context. On the top left, five vertically stacked gray boxes labeled “Service,” “Meeting place,” “One-stop shopping,” “Proximity,” and “Local products” connect to a hexagon labeled “Attitude” at the center by arrows with standardized coefficient values: “0.792 over negative 0.199 and (0.138 over negative 0.107),” “0.198 and (0.081 over 0.107),” “1 and (0.468 over 0.633),” “0.629 and (0.269 over 0.332),” and “0.529 and (0.216 over 0.293),” respectively. On the middle left, two vertically stacked gray boxes labeled “Family” and “Other locals” connect to a hexagon labeled “Subjective norm” at the center by arrows with standardized coefficient values: “1 and (0.833 over 0.633)” and “0.542 over 1.044 and (0.244 over 0.486),” respectively. On the bottom left, five vertically stacked gray boxes labeled “Car availability,” “Health,” “Financial constraints,” “Time sovereignty,” and “Shopping together” connect to a hexagon labeled “Perceived behavioural control” at the center by arrows with standardized coefficient values: “1 and (0.491 over 0.458),” “0.035 over 1.463 and (0.008 over 0.320),” “1.334 and (0.325 over 0.342),” “0.911 and (0.389 over 0.408),” and “1.193 over negative 0.453 and (0.190 over negative 0.089),” respectively. All three hexagons point to a green oval labeled “Intention” by arrows labeled “0.030 and (0.470 over 0.356),” “0.019 and (0.175 over 0.259),” and “0.096 and (0.410 over 0.397),” respectively. Arrows (“0.497 over 0.658 and (0.645 over 0.658)” and “negative 0.011 and (negative 0.062 over negative 0.052)”) from this oval—and directly from “Perceived behavioral control”—lead to another green oval, “Actual shopping behaviour”. Additional arrows branch to indicators: from “Intention” to “Probability of shopping in the new store” and “Satisfaction with shopping facilities”; from “Actual behaviour” to “Purchasing quantity” and “Purchasing share,” all with respective coefficients. The ovals “Intention” and “Actual behaviour” are also labeled “R-squared equals 0.708 over 0.587” and “R-squared equals 0.372 over 0.525,” respectively. Below, the notes read, “Values on the left for Lützelburg, right for Müden; One unstandardised effect: equality constraint across groups; Standardised coefficients in parentheses; cap equals unlikely effect (90 percent credibility interval includes 0); and Model fit: convergence statistic less than 1.005; posterior predictive p equals 0.50”.Estimated model to explain actual shopping in the new store with factors collected before its opening
In terms of attitude towards shopping in the new store, we can conclude that one-stop shopping is the most important belief, followed by proximity to one’s home and availability of local products. Service plays a relatively large role only in Lützelburg, while this effect is not credible in Müden. The preparatory analysis showed that the belief about the meeting point function had a probable effect on the attitude in Müden. The group analysis, however, supports a cross-case consideration, which makes the effect less probable, although positive. For the subjective norm, family plays the central role, with the other locals in Müden exerting a comparable (unstandardized) effect. This reference group does not play a likely role in Lützelburg. In both cases, the PBC is significantly influenced by financial constraints, followed by car availability, and time sovereignty. Health constraints are a very important and probable influence only in Müden and shopping with others only in Lützelburg.
The attitude toward shopping in the new store and the PBC have an approximately similar influence on the intention to shop in the new store (due to the different scales in the compositions, the standardised effects must be taken into account). The influence of the subjective norm is smaller, but still likely. Furthermore, these effects are the same in both case study areas. In contrast, the effects of intention on actual purchases in the new store differ significantly between the two cases. However, the effects are strongly positive in both cases, as expected.
The use of the SEM makes it possible to examine indirect and total effects in addition to direct effects. Table 2 shows the direct effects of the input variables on the composites as well as the indirect effects on intention and actual behaviour. It shows that the credible direct effects also lead to credible indirect effects (the 90% credibility interval is above 0). Since the indirect effects result from multiplying the direct effects along the path, beliefs about one-stop shopping, expectations by family, and car availability have the largest indirect effect on intention and actual behaviour. As Figure 2 shows, the direct effect of PBC on actual behaviour is slightly negative, but unlikely. However, the total effect is still 0.037 (standardized: 0.203) and credible (90% CI: 0.015, 0.060).
Direct effects of beliefs on attitude, subjective norm, and Perceived Behavioural Control (PBC) and indirect effects on intention and actual behaviour, incl. credibility intervals
| Variables | Case study area | Direct effect on attitude | Indirect effect on intention | Indirect effect on actual behaviour | |||
|---|---|---|---|---|---|---|---|
| Effect | 90%-credibility interval | Effect | 90%-credibility interval | Effect | 90%-credibility interval | ||
| Service | Lützelburg | 0.792 (0.318) | 0.349, 1.297 (0.160, 0.471) | 0.023 (0.146) | 0.011, 0.036 (0.070, 0.223) | 0.011 (0.094) | 0.005, 0.019 (0.043, 0.153) |
| Müden | −0.199 (−0.107) | −0.633, 0.239 (−0.336, 0.113) | −0.006 (−0.036) | −0.018, 0.007 (−0.111, 0.040) | −0.004 (−0.027) | −0.013, 0.004 (−0.087, 0.030) | |
| Meeting place | Lützelburg | 0.198 (0.081) | −0.113, 0.531 (−0.053, 0.203) | 0.006 (0.038) | −0.003, 0.014 (−0.023, 0.095) | 0.003 (0.024) | −0.002, 0.008 (−0.015, 0.065) |
| Müden | 0.198 (0.107) | −0.113, 0.531 (−0.067, 0.270) | 0.006 (0.037) | −0.003, 0.014 (−0.023, 0.095) | 0.004 (0.028) | −0.002, 0.010 (−0.017, 0.072) | |
| One-stop shopping | Lützelburg | 1.000 (0.468) | - (0.359, 0.596) | 0.030 (0.214) | 0.023, 0.040 (0.160, 0.284) | 0.015 (0.139) | 0.010, 0.022 (0.089, 0.201) |
| Müden | 1.000 (0.633) | - (0.506, 0.774) | 0.030 (0.221) | 0.023, 0.040 (0.166, 0.293) | 0.020 (0.167) | 0.014, 0.028 (0.116, 0.230) | |
| Proximity | Lützelburg | 0.629 (0.269) | 0.298, 1.015 (0.143, 0.400) | 0.018 (0.123) | 0.010, 0.027 (0.065, 0.182) | 0.009 (0.079) | 0.004, 0.014 (0.039, 0.125) |
| Müden | 0.629 (0.332) | 0.298, 1.015 (0.178, 0.492) | 0.018 (0.115) | 0.010, 0.027 (0.061, 0.171) | 0.012 (0.087) | 0.006, 0.019 (0.044, 0.134) | |
| Local products | Lützelburg | 0.529 (0.216) | 0.217, 0.877 (0.096, 0.338) | 0.016 (0.099) | 0.007, 0.024 (0.043, 0.157) | 0.008 (0.064) | 0.003, 0.013 (0.026, 0.108) |
| Müden | 0.529 (0.293) | 0.217, 0.877 (0.133, 0.453) | 0.016 (0.102) | 0.007, 0.024 (0.045, 0.160) | 0.010 (0.077) | 0.004, 0.016 (0.033, 0.124) | |
| Variables | Case study area | Direct effect on subjective norm | Indirect effect on intention | Indirect effect on actual behaviour | |||
|---|---|---|---|---|---|---|---|
| Effect | 90%-credibility interval | Effect | 90%-credibility interval | Effect | 90%-credibility interval | ||
| Family | Lützelburg | 1.000 (0.833) | - (0.538, 1.086) | 0.019 (0.135) | 0.011, 0.028 (0.076, 0.199) | 0.009 (0.087) | 0.005, 0.015 (0.045, 0.138) |
| Müden | 1.000 (0.633) | - (0.385, 0.880) | 0.019 (0.149) | 0.011, 0.028 (0.083, 0.217) | 0.013 (0.112) | 0.007, 0.019 (0.061, 0.171) | |
| Other locals | Lützelburg | 0.542 (0.244) | −0.408, 1.690 (−0.316, 0.645) | 0.009 (0.045) | −0.008, 0.025 (−0.040, 0.125) | 0.004 (0.029) | −0.004, 0.012 (−0.026, 0.081) |
| Müden | 1.044 (0.486) | 0.258, 2.236 (0.193, 0.736) | 0.018 (0.116) | 0.006, 0.029 (0.040, 0.187) | 0.011 (0.087) | 0.004, 0.019 (0.029, 0.144) | |
| Variables | Case study area | Direct effect on PBC | Indirect effect on intention | Indirect effect on actual behaviour | |||
|---|---|---|---|---|---|---|---|
| Effect | 90%-credibility interval | Effect | 90%-credibility interval | Effect | 90%-credibility interval | ||
| Car availability | Lützelburg | 1.000 (0.491) | - (0.361, 0.622) | 0.096 (0.202) | 0.067, 0.127 (0.141, 0.268) | 0.037 (0.100) | 0.015, 0.060 (0.043, 0.161) |
| Müden | 1.000 (0.458) | - (0.334, 0.587) | 0.096 (0.182) | 0.067, 0.127 (0.128, 0.239) | 0.052 (0.113) | 0.028, 0.078 (0.062, 0.168) | |
| Health | Lützelburg | 0.035 (0.008) | −0.714, 0.754 (−0.175, 0.178) | 0.004 (0.004) | −0.065, 0.070 (−0.070, 0.075) | 0.001 (0.002) | −0.026, 0.029 (−0.037, 0.040) |
| Müden | 1.463 (0.320) | 0.654, 2.457 (0.156, 0.479) | 0.137 (0.128) | 0.063, 0.212 (0.059, 0.198) | 0.074 (0.080) | 0.028, 0.128 (0.030, 0.136) | |
| Financial constraints | Lützelburg | 1.334 (0.325) | 0.736, 2.038 (0.211, 0.439) | 0.124 (0.133) | 0.079, 0.169 (0.085, 0.181) | 0.047 (0.066) | 0.019, 0.080 (0.027, 0.110) |
| Müden | 1.334 (0.342) | 0.736, 2.038 (0.221, 0.463) | 0.124 (0.136) | 0.079, 0.169 (0.086, 0.187) | 0.067 (0.084) | 0.034, 0.104 (0.043, 0.131) | |
| Time sovereignty | Lützelburg | 0.911 (0.389) | 0.478, 1.493 (0.252, 0.529) | 0.084 (0.159) | 0.053, 0.115 (0.100, 0.220) | 0.032 (0.079) | 0.012, 0.055 (0.032, 0.135) |
| Müden | 0.911 (0.408) | 0.478, 1.493 (0.265, 0.549) | 0.084 (0.162) | 0.053, 0.115 (0.102, 0.224) | 0.046 (0.101) | 0.022, 0.072 (0.050, 0.159) | |
| Shopping together | Lützelburg | 1.193 (0.190) | 0.054, 2.486 (0.009, 0.363) | 0.111 (0.079) | 0.005, 0.215 (0.004, 0.153) | 0.041 (0.038) | 0.001, 0.091 (0.001, 0.084) |
| Müden | −0.453 (−0.089) | −1.452, 0.475 (−0.278, 0.089) | −0.042 (−0.035) | −0.129, 0.044 (−0.107, 0.036) | −0.024 (−0.022) | −0.078, 0.022 (−0.073, 0.021) | |
Note(s): Standardized values in parentheses
Source(s): Authors’ own work
5. Discussion
At the outset of the paper, we noted that current practices for forecasting the use of planned neighbourhood stores involve a significant degree of intuition. This suggests that there are shortcomings in the theoretical understanding of these practices and that further development of existing models is warranted. With this in mind, our research question was: which factors can be used to predict the choice of a new neighbourhood store before it opens? Using two case studies, we demonstrated that the Theory of Planned Behaviour (TPB) allows the development of a corresponding model. By taking into account attitude towards shopping at the new store, subjective norm, and Perceived Behavioural Control (PBC), our model can explain the intention to shop at the new store and, ultimately, the actual shopping behaviour. The estimated effect sizes and variance explanations are comparable to or even better than previous applications of the TPB in other fields (Ajzen, 2012; Armitage and Conner, 2001). Taken together, our results demonstrate that the TPB is well suited for predicting the choice of a new neighbourhood store.
The attitudinal beliefs largely correspond to factors that have already been widely used in analyses of shopping behaviour. These approaches are based on multi-attribute utility theory (Wallenius et al., 2008; Yokoyama et al., 2022). Our analysis shows that many of the factors that were previously thought to be relevant were not important, at least in the two cases under study. These factors include: quality, price level, products on sale, trip chains, and assortment. It must be noted, though, that the belief about local products, which was identified as important, constitutes a partial aspect of quality. A recent meta-analysis of the effect of assortment on sales shows that the impact diminishes for functional products with high purchase frequency, such as many grocery items, and, due to a concave elasticity, for small stores with low assortment (Sethuraman et al., 2022). Price level and special offers, which were also excluded from our model as irrelevant, are presumably reflected in PBC-related beliefs about financial constraints. Moreover, a recent review article indicates an ambiguous effect of price on store format choice (Bonfrer et al., 2022), and the expectations about the price level in neighbourhood stores are particularly low (Yokoyama et al., 2022). The meta-analysis of Liu-Thompkins et al. (2022) confirms a decreasing effect of price (and assortment) on retailer loyalty over the last 50 years.
Furthermore, it is noticeable that the effect of subjective norm on intention, and therefore on actual behaviour, is smaller than that of attitude and PBC. As meta-analyses show, this is rather normal for models built on TPB (Armitage and Conner, 2001, p. 481). A further reason for this may be that we did not apply the extension of the TPB to the Reasoned Action Approach (RAA) (Fishbein and Ajzen, 2010), as we have only considered the injunctive norm and not the descriptive norm. While the injunctive norm measures perceived expectations by relevant others, the descriptive norm captures the observed behaviour of social referents. Including the descriptive norm in a model (alongside attitude, injunctive norm, and PBC) increases the proportion of variance explained in intention by an average of 5% (Ajzen, 2012, p. 17). However, since shopping by reference persons in a store that has not yet opened cannot be observed, it was not possible to collect the descriptive norm before the store opened. Therefore, only the injunctive norm could be included in our model, which may contribute to the smaller effect of the subjective norm.
Similar to the descriptive norm, affective attributes, which are often considered to be critical to store loyalty (Liu-Thompkins et al., 2022), could not be measured. Respondents are not able to evaluate aspects such as atmosphere, image or design of a store that does not yet exist. Therefore, this potentially important dimension is missing from our model, which nevertheless explains a high proportion of the variance in shopping behaviour.
As expected, PBC has an important effect on the intention to shop at a new store. However, there is no direct effect on the actual behaviour, so the effect is perfectly mediated by intention. The meta-analysis by Armitage and Conner (2001, pp. 483–484) shows that there is a significant direct effect (0.40) on average, but it only increases the variance explained by intention alone by 2% and is thus almost negligible. Here, PBC can hardly serve as a proxy for actual behavioural control. Even though control factors, such as car availability, health status, or financial situation are unlikely to change within such a short period of time, the subjective evaluation of the factors in terms of their enabling or preventing influence on shopping at the new store appears to be potentially shaped by social desirability. The descriptive analysis shows, for example, that many respondents assume that the availability of a car either has no influence or a positive influence on shopping at the local store. In contrast, existing research, particularly in rural contexts, indicate that the availability of a car increases the possibility of out-shopping in larger and thus more attractive stores located further away (Christiaanse and Haartsen, 2017; Jürgens, 2020; Marshall et al., 2018).
The limitation of our study to two cases and the presumably selective response behaviour in favour of those personally impacted affect the representativity of our results. However, we assume that the strategic selection of strongly varying cases and the high response rate ensure a certain degree of representativeness. As always in Bayesian statistics, the results only ever apply to existing data, i.e. the respondents. An advantage of Bayesian structural equation modelling is that the model is empirically identified even when n is small. Although the overall internal validity of the measurement model is good, the intention loads low on the satisfaction with existing shopping facilities. Nonetheless, the consistency between both groups supports the validity of this measurement model as well as the external validity due to the high effect of intention on behaviour, as expected according to the TPB. We adhere to the construct of intention because it helps to correct for measurement errors that inevitably arise from questionnaire surveys.
In total, only four of the 21 estimated effects had to be estimated separately for the two cases, indicating a certain reliability of the model. Nevertheless, additional cases should be examined in order to further substantiate the model and, if necessary, to explain local specificities. For example, it could be hypothesized that the effect of other locals on the subjective norm was likely in Müden and not in Lützelburg because Müden’s status as a health resort depends on the presence of a grocery store. Another avenue for further research is offered by the larger effect of intention on behaviour in Müden. A hypothesis to test could be that the sales area in Müden is larger than in Lützelburg, therefore making it easier to fulfil the customers' expectations in the sense of the expectation disconfirmation model (Yokoyama et al., 2022) or loss aversion (Goić et al., 2021). Further research could, for example, establish whether the belief about price level is not a relevant explanatory factor for attitudes in other cases when real income is much lower. Therefore, we would welcome replication studies in low-income situations as well as for other store formats and urban contexts where competition is more intense (Çakır et al., 2020) and thus quality, price, and assortment may play a significant role.
6. Conclusion
Our study advances the scientific debate on store choice forward in three ways. First, we develop a model to predict the expenditure in a new neighbourhood store before its opening. To our knowledge, this is the first time that the TPB has been applied to store choice. Our study stands out by testing the model using a two-wave survey, which enables us to chronologically separate cause and effect. Second, we demonstrate how attitude, subjective norm, and PBC can be measured indirectly, thus avoiding a tautologic correlation between the direct measure of these constructs on the one hand and intention on the other. By using composites rather than a summative index to measure these constructs, we can specify the relevant beliefs and their impact on intention and actual behaviour. Third, our model contributes to the discussion on the importance of social interactions, which Liu-Thompkins et al. (2022) have recently proposed as a promising research direction. Our contribution integrates service (employee-customer), meeting places and shopping together (customer-to-customer), and social pressure.
Our results have three main practical implications. First, retail companies could use our model and our items to develop more realistic sales forecasts. Therefore, we would recommend testing our model with an “objective” measure of actual behaviour using, e.g. rewards apps over a longer period of time. Planning authorities could also require more serious profitability calculations based on our model. The EAFRD could fund such surveys as preparatory studies for rural neighbourhood stores, which would otherwise be too costly for small retailers to carry out. Second, the results indicate the potential levers that companies have at their disposal to increase purchasing power retention. Accordingly, retailers should select easily accessible locations that are, for example, located on main thoroughfares in village centres and have sufficient parking spaces. They should preferably offer a wide range of products, including local products, rather than a deep range and premium or discount products. They should also pay staff well to keep them motivated and friendly. Conversely, they should not offer gastronomic services as meeting points.
However, a systemic consideration of these attributes in the context of subjective norms and PBC also reveals the limitations of such strategies. Beliefs about PBC, in contrast to those about store attributes, can hardly be influenced directly, so that, thirdly, communication strategies might be used primarily to increase the perceived social pressure. In order to identify suitable action approaches, there is a need for future intervention studies (Fishbein and Ajzen, 2010) that demonstrate the effects of, for example, communication and participation measures in the planning phase before a store opening on beliefs and thus on purchasing behaviour.
Funding: This work was supported by the German Federal Ministry of Food and Agriculture (BMEL) [grant number 2817LE018].
References
Appendix 1
Description of the case study areas
| Lützelburg | Müden | |
|---|---|---|
| Number of inhabitants | 1,730 | 2,420 |
| Number of households | 655 | 1,029 |
| Municipality | Gablingen | Faβberg |
| Distance to the next large city | 19 km to Augsburg | 77 km to Hanover |
| Inhabitants per km2 (municipality) | 178.4 | 61.5 |
| Share of inhabitants 65 years and older (municipality) | 21.0% | 25.0% |
| Purchasing power per inhabitant | 26,776 Euro | 24,765 Euro |
| Distance to the next grocery store | 3 km | 4 km |
Source(s): Compiled by the Authors with data from internet research (number of inhabitants, distance to large city and grocery store), www.inkar.de (population density, elderly population, purchasing power) and own data (number of households)
Appendix 2
Method of calculation and descriptive statistics for the variables used for modelling, differentiated between the two cases Lützelburg and Müden
| Variable | Calculation procedure | Case | Mean | Median | Standard deviation | Minimum | Maximum | Skewness | Kurtosis | N |
|---|---|---|---|---|---|---|---|---|---|---|
| Quality | importance of quality * probability of quality | Lützelburg | 23.9 | 25 | 8.4 | 0 | 36 | −0.436 | −0.182 | 305 |
| Müden | 23.5 | 25 | 8.2 | 0 | 36 | −0.304 | −0.334 | 432 | ||
| Price level | importance of price level * probability of low prices | Lützelburg | 9.6 | 8 | 7.6 | 0 | 36 | 1.353 | 2.251 | 304 |
| Müden | 12.4 | 12 | 8.2 | 0 | 36 | 0.907 | 0.693 | 436 | ||
| Products on sale | importance of products on sale * probability of products on sale | Lützelburg | 10.3 | 9 | 8.7 | 0 | 36 | 1.154 | 1.244 | 308 |
| Müden | 11.9 | 10 | 8.9 | 0 | 36 | 0.912 | 0.358 | 432 | ||
| Service | importance of service * probability of good service | Lützelburg | 23.6 | 25 | 9.4 | 0 | 36 | −0.339 | −0.746 | 302 |
| Müden | 25.5 | 25 | 9.1 | 0 | 36 | −0.648 | −0.219 | 437 | ||
| Meeting place | importance of meeting place * probability of meeting place | Lützelburg | 9.0 | 6 | 10.0 | 0 | 36 | 1.310 | 0.981 | 304 |
| Müden | 10.2 | 8 | 10.0 | 0 | 36 | 1.092 | 0.493 | 431 | ||
| One-stop shopping | importance of one-stop shopping * probability of one-stop shopping | Lützelburg | 13.5 | 12 | 10.7 | 0 | 36 | 0.614 | −0.483 | 306 |
| Müden | 15.7 | 15 | 11.2 | 0 | 36 | 0.359 | −0.971 | 435 | ||
| Trip chains | importance of trip chains * probability to combine paths | Lützelburg | 6.7 | 3 | 9.4 | 0 | 36 | 1.670 | 2.106 | 307 |
| Müden | 9.0 | 5 | 10.3 | 0 | 36 | 1.082 | 0.139 | 437 | ||
| Proximity | importance of proximity * probability of proximity | Lützelburg | 25.1 | 30 | 10.1 | 0 | 36 | −0.812 | −0.140 | 305 |
| Müden | 27.2 | 30 | 9.6 | 0 | 36 | −1.105 | 0.436 | 437 | ||
| Assortment | importance of assortment * probability of assortment | Lützelburg | 12.2 | 10 | 8.8 | 0 | 36 | 0.800 | 0.294 | 305 |
| Müden | 13.4 | 12 | 9.3 | 0 | 36 | 0.807 | 0.087 | 436 | ||
| Local products | importance of local products * probability of local products | Lützelburg | 24.4 | 25 | 9.6 | 0 | 36 | −0.515 | −0.404 | 295 |
| Müden | 22.1 | 24 | 10.0 | 0 | 36 | −0.346 | −0.615 | 430 | ||
| Family | importance of family’s expectations * probability of family’s expectations | Lützelburg | 9.1 | 4 | 10.7 | 0 | 36 | 1.085 | 0.156 | 311 |
| Müden | 11.2 | 8 | 11.9 | 0 | 36 | 0.751 | −0.681 | 439 | ||
| Friends | importance of friends' expectations * probability of fiends' expectations | Lützelburg | 4.9 | 1 | 7.8 | 0 | 36 | 2.160 | 4.891 | 309 |
| Müden | 7.4 | 3 | 9.5 | 0 | 36 | 1.402 | 1.320 | 436 | ||
| Other locals | importance of other locals' expectations * probability of other locals' expectations | Lützelburg | 4.6 | 0 | 7.5 | 0 | 36 | 2.125 | 4.720 | 310 |
| Müden | 7.4 | 2 | 10.1 | 0 | 36 | 1.396 | 1.064 | 436 | ||
| Car availability | importance of car availability * probability of car availability | Lützelburg | 0.1 | 0 | 3.2 | −6 | 6 | −0.025 | 0.474 | 306 |
| Müden | 0.9 | 0 | 2.9 | −6 | 6 | 0.320 | 0.649 | 443 | ||
| Health | importance of health * probability of health | Lützelburg | 0.5 | 0 | 1.6 | −6 | 6 | 1.675 | 6.091 | 307 |
| Müden | 0.4 | 0 | 1.4 | −6 | 6 | 1.665 | 8.201 | 439 | ||
| Financial constraints | importance of financial constraints * probability of financial constraints | Lützelburg | −0.4 | 0 | 1.6 | −6 | 6 | −1.079 | 4.521 | 302 |
| Müden | −0.1 | 0 | 1.7 | −6 | 6 | −0.106 | 4.548 | 433 | ||
| Time sovereignty | importance of time sovereignty * probability of time sovereignty | Lützelburg | 0.8 | 0 | 2.9 | −6 | 6 | 0.182 | 0.609 | 301 |
| Müden | 1.4 | 0 | 3.0 | −6 | 6 | 0.168 | −0.114 | 435 | ||
| Shopping together | importance of shopping together * probability of shopping together | Lützelburg | 0.2 | 0 | 1.1 | −6 | 6 | 2.514 | 15.675 | 309 |
| Müden | 0.3 | 0 | 1.2 | −6 | 6 | 1.741 | 11.780 | 437 | ||
| Probability of shopping in the new store | – | Lützelburg | 3.2 | 3 | 1.7 | 0 | 6 | −0.141 | −0.758 | 310 |
| Müden | 3.8 | 4 | 1.7 | 0 | 6 | −0.413 | −0.604 | 443 | ||
| Satisfaction with shopping facilities | – | Lützelburg | 4.0 | 4 | 1.5 | 0 | 6 | −0.734 | 0.198 | 311 |
| Müden | 3.6 | 4 | 1.6 | 0 | 6 | −0.387 | −0.449 | 442 | ||
| Purchasing quantity | purchase in the neighbourhood store | Lützelburg | 2.3 | 2 | 1.2 | 0 | 6 | 0.283 | 0.047 | 238 |
| Müden | 3.2 | 3 | 1.4 | 0 | 6 | −0.083 | −0.760 | 367 | ||
| Purchasing share | purchase in the neighbourhood store/(purchase in the neighbourhood store + supermarket + discounter + hypermarket + speciality stores + organic market or health food store + kiosk or gas station store + farmer’s market, store or box + mobile retailer + e-commerce) *100 | Lützelburg | 17.2 | 15.4 | 12.8 | 0 | 100 | 2.294 | 10.311 | 195 |
| Müden | 23.5 | 21.4 | 13.8 | 0 | 83.3 | 0.979 | 1.257 | 287 |
Source(s): Authors’ own work
Appendix 3
The path diagram connects two main green ovals: “Intention” at the left and “Actual behaviour” at the right, with a curved blue arrow linking Actual behaviour back to Intention. Downward arrows from “Intention” point to two gray boxes: “Probability of shopping in the new store,” with coefficient 1 (0.916), and “Satisfaction with shopping facilities,” with coefficient negative 0.407 (negative 0.402). “Actual behaviour” points to two outcomes: “Purchasing quantity,” with coefficient 1 (0.943), and “Purchasing share,” with coefficient 9.001 (0.850). Below, the notes read, “Standardised coefficients in parentheses; Equality constraints across groups for factor loadings, error variances, and factor variances; and Model fit: convergence statistic less than 1.002; posterior predictive p equals 0.49”.Results of the measurement model for intention and actual behaviour
The path diagram connects two main green ovals: “Intention” at the left and “Actual behaviour” at the right, with a curved blue arrow linking Actual behaviour back to Intention. Downward arrows from “Intention” point to two gray boxes: “Probability of shopping in the new store,” with coefficient 1 (0.916), and “Satisfaction with shopping facilities,” with coefficient negative 0.407 (negative 0.402). “Actual behaviour” points to two outcomes: “Purchasing quantity,” with coefficient 1 (0.943), and “Purchasing share,” with coefficient 9.001 (0.850). Below, the notes read, “Standardised coefficients in parentheses; Equality constraints across groups for factor loadings, error variances, and factor variances; and Model fit: convergence statistic less than 1.002; posterior predictive p equals 0.49”.Results of the measurement model for intention and actual behaviour
