Introduction

Commuting and buying groceries are regular trips for many individuals and account for a large share of weekly travel (32% of kilometers traveled in Germany (Follmer and Gruschwitz 2019). These trips create a high proportion of negative transport externalities such as noise, congestion, and greenhouse emissions (Cao and Yang 2017). Moreover, long commute distances are connected to lower individual well-being and health (Chatterjee et al. 2020). Daily travel behavior is based on medium-term location choices such as residential location decisions and long-term lifestyle decisions (Van Acker et al. 2010). While grocery shopping destinations largely depend on the supply of shopping locations in the vicinity of the residence, commute trip destinations are more subject to personal factors (Axisa et al. 2012).

In dense, urban areas, trip distances tend to be shorter for commuting and shopping trips (Zhao et al. 2010; Smith et al. 2023). However, even though the connection between the built environment and travel behavior has been confirmed by a multitude of studies (see Stevens (2017) for a meta-analysis), findings differ regarding the direction and strength of the effects as the effect of the built environment on travel behavior is mediated by residential location choice. Individuals might select their residential location according to their travel needs, resulting in travel-related residential self-selection; they might prioritize other criteria; or they might not even have much choice in tight housing markets (Ahrens and Lyons 2021; Schimohr et al. 2023). The influence of residential self-selection on the residential location introduces a bias in the analysis of the impact of the built environment on travel behavior (Heinen et al. 2018). For example, Cao et al. (2006) found that the propensity towards walking to nearby shopping destinations can be partly explained by residential self-selection. While a vast array of studies has been conducted on residential self-selection, there is still considerable disagreement concerning the existence, direction, and magnitude of the bias noted above (Guan et al. 2020). Therefore, ideally, datasets including information about how the residential environment was chosen allow for analyses that account for residential self-selection.

Furthermore, following the hypothesis of reverse causality, the built environment might also influence travel attitudes, either directly or mediated through the impact of the built environment on travel behavior (van Wee et al. 2019). Therefore, the moment of relocation is of crucial interest to untangle these relationships. The analysis of travel behavior and attitudes preceding and following a residential relocation allows us to understand the impact of relocations and thus, changes in the built environment, on travel distances to daily destinations. Several studies investigate the effect of a residential relocation on changes in travel mode choice (see, for example, De Vos et al. (2018), Tao (2023), Aditjandra et al. (2016), Scheiner and Holz-Rau (2013). In contrast, there are fewer findings regarding changes in commute trip length. The number of studies investigating changes in grocery shopping trip distance is even more limited. In previous studies, relocations were found to be associated with changes in trip distances depending on the direction of a move: commute trip distances tend to increase after relocating to less urban locations (Buchanan and Barnett 2006; Næss 2005; Cervero and Day 2008; De Vos et al. 2018). Correspondingly, after a move from suburban or rural locations to urban neighborhoods, commute and shopping distances tend to decrease (Buchanan and Barnett 2006). Movers staying within a similar distance to the city center or the same neighborhood type maintain travel behavior (Næss 2005) or decrease their average commute distance (De Vos et al. 2018). Further, some studies observed attitudes to change in response to changes in the built environment or trip distances after relocation (De Vos et al. 2021; Ramezani et al. 2021).

Due to the complexity of the interrelations involved, panel datasets of movers are ideal to analyze the relationship between changes in the built environment and travel behavior after relocation (van Wee and Cao 2022). Cross-sectional or quasi-longitudinal studies may suffer from recall errors that may lead to lower accuracy and biases. The bias increases with the time span involved and the inclusion of non-tangible items such as attitudes (Dex 1995). To date, however, only a few studies use panel data to analyze changes in travel behavior while measuring attitudes before and after relocation, due to the effort involved in collecting such a dataset (Wang and Lin 2019). Even fewer studies include changes in trip distances in the analysis (Tao et al. 2023). To our knowledge, no study with a focus on trip distances has yet been conducted that uses longitudinal data of movers.

This study aims to determine the role of trip distances in residential location decisions. We investigate whether movers wish to decrease trip distances and whether satisfaction with commute and grocery shopping trips before relocation or travel attitudes affect residential location choices. Thus, we evaluate trip distances to the workplace and grocery stores before and changes after a residential relocation based on a panel study of movers in Germany who were surveyed twice, before and after relocation. The reasons for investigating the two selected activities are twofold. First, both trip purposes represent some of the most regularly undertaken trips (Follmer and Gruschwitz 2019). Second, there are substantial differences in how these trips interact with the built environment. Commute trips depend strongly on individual factors, may cover regional distances, and job locations often persist after relocations. In contrast, shopping trips are highly dependent on the local built environment and shopping destinations usually change after residential relocation (Buchanan and Barnett 2006). We model the relationship between trip length, spatial structure, and travel attitudes before and after relocation, taking into account sociodemographic characteristics and housing preferences in residential location search through structural equation modeling (SEM). This gives insight into whether respondents decrease trip distances in their relocation decisions, which conditions support or hinder a decrease, and which other relationships exist between the variables under consideration.

Literature and conceptual model

While a number of studies investigate factors that are associated with commute or grocery shopping trip distances, most of them are based on cross-sectional analyses. Exceptions using panel data include Clark et al. (2003), Beige and Axhausen (2017), and Prillwitz et al. (2007). Scheiner (2010) and Vale (2013) apply a quasi-longitudinal methodology. However, at a given residential location, only grocery shopping trip distances can be easily varied depending on travel attitudes. For example, individuals with positive walking attitudes tend to choose grocery stores nearby so they can walk (Handy and Clifton 2001). Conversely, households with a car accept much farther distances.

Commute distances, in contrast, are relatively fixed and determined by the relationship between residential location and workplace (at least under the assumption of regular employment and a traditional workplace). Changes in commute distance are always connected to either a residential relocation, a job change, or a workplace relocation. Thus, the decision-making processes preceding these events are crucial to understand the factors underlying commute distances, while grocery shopping trips are usually adapted after residential relocations. In the following, we give an overview of distance-related aspects that play a role in residential location choice.

Transport-related criteria and proximity to destinations in housing search

First, travel attitudes may influence the choice of a residential location, resulting in travel-related residential self-selection (Cao et al. 2009). People might self-select according to their attitudes towards travel modes, which are related to certain neighborhood structures. While dense urban structures support the usage of active modes of transport or public transport, suburban or rural locations promote car travel (Cao et al. 2007). Dense and urban areas are further associated with shorter trip distances for commuting and grocery shopping due to a higher density of destinations (Surprenant-Legault et al. 2013; Boussauw et al. 2012; Smith et al. 2023; Gil Solá and Vilhelmson 2022). Therefore, urban areas offer additional benefits for active travel.

Proximity to the workplace or school was found to play into the decision-making for a majority of movers (Langlois et al. 2015). As long commutes are generally found to lead to dissatisfaction (Chatterjee et al. 2020), it could be expected that movers aim to minimize their commute distances. However, the relationship is much more complex than this and the actual importance given to this criterion is rather low (Giles-Corti et al. 2008; Chatman 2009). Vale (2013) found that decreases in commute time did not increase commute satisfaction if the commute time was already below the maximum commuting tolerance. Similarly, Ma et al. (2021) found that decreases in commute time did not increase satisfaction if the time spent commuting actually represented the ideal commute time. Thus, commute satisfaction would be similar for all job locations that can be reached within this time. Consequently, households value a short distance to the workplace in their residential location decision but do not try to minimize it. Instead, they evaluate neighborhoods within a predefined search perimeter around fixed locations (Bruns and Matthes 2019; Guidon et al. 2019). Clark et al. (2003) empirically found an approximate critical limit for the distance between home and workplace of about 8 miles (∼ 13 km) and between 12 and 16 miles (∼ 19–26 km) for two-worker households. The situation can be further relieved for those with the option of working from home regularly, as employees then need to commute less frequently, if at all (Zhu 2013).

Job characteristics also determine the role of commute distances in residential location decisions: full-time work and higher incomes lead to higher job utility, making longer commute trips worthwhile (Manaugh et al. 2010; van Ommeren and Fosgerau 2009). This is amplified by the fact that jobs requiring higher education levels and those in specialized sectors, usually connected to higher wages, are unevenly distributed. Consequently, the probability of finding a suitable job nearby is lower (Axisa et al. 2012; Zhu 2013). Studies investigating movers who start a new job at the same time as their relocation have led to diverging results. Cervero and Day (2008) found a negative effect of job changes on changes in travel time, analyzing a dataset of recent movers to the outskirts of Shanghai. Here, changing the residential location and the job simultaneously allowed respondents to find locations nearby while overall, commute distances increased. In contrast, Beige and Axhausen (2017) and Prillwitz et al. (2007) observed increases in distance after relocations connected to job changes.

As women, on average, perform a higher share of household duties, work more often in part-time jobs, and earn less than men, they tend to accept shorter commute distances than men (Chidambaram and Scheiner 2020; Hu 2021; McQuaid and Chen 2012; Shearmur 2006). Consequently, residential location decisions are more sensitive to the commute distances of female household members (Sermons and Koppelman 2001; Clark et al. 2003). These tendencies are even more pronounced in households with children. Overall, commuters in households with children travel farther, as these households tend to value certain criteria in residential location search that are easier to realize outside of agglomerations with a high density of jobs (Shearmur 2006; Schirmer et al. 2014). The relationship between age and commute distances seems to be non-linear and to follow a U-shape: first, distances tend to increase until the age of 30–44 and then they decrease again, which may be linked to life stage issues such as having children and following career opportunities in mid-life (Axisa et al. 2012).

Accessibility of shops plays a role in residential location decisions as well (Chatman 2009; Giles-Corti et al. 2008) and good accessibility is important for households regardless of preferred travel mode (Bruns and Matthes 2019). Furthermore, grocery shopping trip distances might interact and be combined with other regular trips, including commuting, forming trip chains (Van Acker and Witlox 2011). Overall, however, the empirical basis for considering the role of shopping trip lengths and changes after relocation is still scarce.

In sum, a range of sociodemographic and spatial characteristics, as well as attitudes, have been found to influence residential location decisions, resulting in corresponding changes in trip distances after relocation. The question remains as to why theoretical considerations suggest a strong relationship between relocations and changes in commute distance, while empirical studies lead to contrary results. So far, little is known about how grocery shopping trip distances are evaluated in residential decisions and how they change after relocation, although it can be assumed that grocery destinations often change after relocation. Further, the roles of commute and grocery shopping trips in residential relocations have not been studied concurrently.

Conceptual model

Building on the findings of previous research and the identified research gaps, we developed a conceptual model to analyze changes in trip distance after relocation within a set of interrelated factors involved in residential location decisions (Fig. 1).

Fig. 1
figure 1

Conceptual model for the analysis of trip lengths before and after relocation

Full size image

Changes in trip distance after relocation are relative to the trip distance before relocation, which should be controlled for (Allison 1990). We allow for residential self-selection effects in two ways. First, we include a link from travel attitudes on residential preferences that, in turn, may affect the choice of residential location. The simultaneous occurrence of both relationships would indicate residential self-selection effects. We control for an effect of trip distances and the previous residential location on residential preferences (Van Acker et al. 2010). Second, we include a direct link from trip satisfaction before relocation to change in trip distance. As long trip distances were found to be related to low satisfaction (Ma et al. 2021; De Vos et al. 2019), we assume that lower trip satisfaction before relocation motivates movers to decrease trip distances after relocation.

Based on the findings of studies linking the built environment and trip distances, we assume that living in urban areas before relocation permits shorter distances (Buchanan and Barnett 2006; Manaugh et al. 2010; Næss 2005; Smith et al. 2023). Similarly, we assume that relocations towards more urban neighborhoods are overall associated with decreases in trip distances. Thus, we include a link from changes in urbanity to changes in trip distance. Especially for grocery shopping trips, we hypothesize that changes in trip distances mainly result from differences in accessibility levels between new and old neighborhoods. Further, the urban structure at the residential location might influence travel attitudes (residential determination) (Ewing et al. 2016). To allow for this effect, we also include a link from changes in neighborhood structure to travel attitudes. We hypothesize that travel attitudes towards active modes improve after relocations to more urban neighborhoods, as these areas usually allow easier travel by active modes (De Vos et al. 2021).

Travel behavior studies have traditionally focused on the effects of travel attitudes on travel behavior. In recent years, a discussion about the reverse effect of travel behavior on attitudes has emerged that can be based on Festinger’s theory of cognitive dissonance (1962). Several studies have found evidence for the existence of this effect (Kroesen et al. 2017; De Vos et al. 2021). To contribute to this discussion, we include a link from changes in trip distance to travel attitudes that were measured after relocation. We expect travel attitudes towards active modes to improve after decreases in trip distance triggered by relocation.

Similarly, we assume that favorable attitudes towards active modes and trip distances before relocation are negatively related (Boussauw et al. 2012; Smith et al. 2023; Handy and Clifton 2001). While this relationship might be bidirectional, we cannot infer the direction of causality in cross-sectional data. For the data collected before relocation, we focus on determinants of trip distances at a given residential location. Thus, we choose to investigate the link from travel attitudes to trip distances. For travel attitudes to exert an impact on trip distances, respondents might choose a grocery store that is accessible by the preferred mode. The connection is less obvious for commute trips, but travel attitudes might have played a role when choosing the (previous) residential location or the job location. Finally, we control for sociodemographic characteristics before and after relocation to account for further influences outside the focus of this study. As outlined in the preceding section, these factors were found to be associated with residential choices and trip distances as well.

Methodology

Data

Data was collected through a combined online / paper-and-pencil survey in Germany in 2021–2023. We contacted individuals who were looking for a new place to live through links to our online survey on real estate and housing websites. Additionally, a number of housing associations who supported the project handed out printed surveys with additional links to the online survey or sent links to the survey to future renters. As we do not have information on how many times the websites were accessed or how many invitations were actually distributed by the housing associations, we cannot provide a figure for the return rate. The housing associations involved in the survey are located in the cities of Berlin, Munich, and the Ruhr area. Additional contact channels, such as a nationwide website for housing search and online ads, allowed respondents from all over Germany to be included in the sample.

We surveyed one individual per household. Respondents were contacted again four months later using the contact channel they indicated (usually email, but letters or phone calls were possible as well). Respondents who had not yet moved could postpone the second survey. However, of the 3,953 respondents who completed the first survey, 1,838 did not provide any contact information, and at the time of writing 337 had not yet reached the four-month deadline at which they should be contacted for the second survey. Additionally, 1,315 respondents postponed the second survey or dropped out of the survey entirely. This may be attributed to the difficulty of finding a new residence. In many German cities, housing markets are tight. The housing markets in Berlin and Munich, where many respondents lived, are strongly supply-dominated. In the end, 463 individuals successfully participated in the survey before and after relocation. 20 respondents did not specify a valid postal code, essential information for our analysis, and were therefore excluded. Further, 8 respondents who did not answer many questions in the survey were excluded as well. The final dataset contains 435 individuals. Compared to the larger sample of all respondents who completed the first survey, the panel dataset includes more respondents aged 30–49 and less aged over 65, slightly more men, and more respondents with tertiary education. Other characteristics are similar between the samples.

In the models (see Sect. 3.3), we use two subsets of the total sample consisting only of respondents who commute/go grocery shopping at least occasionally and who indicate respective trip lengths in both surveys (commuting n = 339, grocery shopping n = 424). Similar to Næss (2005), Van Acker and Witlox (2011), and Islam and Saphores (2022), those with disproportionally long commute distances were excluded to decrease the influence of outliers that would lead to a stronger deviation from the normal distribution. In this study, the definition of Tukey’s fences was employed to detect outliers (Tukey 1977). Those whose commute trips exceeded 50 km either before or after a move were excluded (n = 20). Similarly, shopping trips over 10 km far exceeded Tukey’s fences and were classified as outliers, leading to a stronger deviation from the normal distribution. Thus, observations where shopping trips exceeded 10 km were removed (n = 14). Our final datasets include 319 observations for commute trips and 410 for shopping trips.

Both surveys before and after a move included similar questions about neighborhood satisfaction, travel behavior, travel attitudes as general dispositions towards different modes of transport, and sociodemographic characteristics of the household (see Sect. 3.2). In the first survey, we additionally asked respondents about their reasons for moving and the importance given to various criteria in their housing search, indicative of residential preferences.

In contrast to many studies dealing with only one type of move (urban to suburban or suburban to urban) and one distinct geographical region, our sample covers movers from all over Germany and a diverse set of neighborhoods. Moves were recorded in all parts of Germany, but there are clusters in Munich, the Ruhr area in the West of Germany, and especially in Berlin. There were a number of moves to Berlin from all parts of Germany. While many relocations occured in close proximity to the previous postal code or even within the same one (29.4%), some respondents moved between different regions of Germany. 8.3% of respondents moved farther than 100 km from their previous location. The mean distance between the previous and new postal code area was 34.5 km (median 4.4 km).

About 9.6% of the population in Germany move each year (Deutsche Post Adress 2021). As there are typical differences between movers and the general population, our sample could only be representative of the group of movers. We compared our sample to a subset of the representative German Socio-Economic Panel (SOEP) for 2020 of those respondents who moved in the previous year. Gender ratio, employment status, and the share of households with children in our sample are similar to the reference data (SOEP 2022; IAB-SOEP 2022; IAB-BAMF-SOEP 2022). Respondents in the age group of 18–29 are underrepresented, while respondents between 30 and 49 and single-person households are overrepresented.

Variables

In the questionnaire, we asked about frequency, usual modes of transport, and distance traveled for trips to the workplace/place of education and grocery shopping. To collect trip distances, we asked respondents to indicate the “distance traveled (estimated, one way only)”. Using a dropdown menu, the distances were collected in 25 categories that were small enough for us to assume that the mean values of each category can be used as continuous variables without considerable distortion. A number of features collected in the survey were found to be associated with trip lengths in previous research and are hence included in the analysis (Table 1). Spatial characteristics are determined based on postal code areas and supplemental data from OpenStreetMap, for the residential locations both before and after a move (OpenStreetMap contributors 2021). Change variables (trip distance, urbanity) were calculated as differences between values collected before and after relocation. Thus, positive values indicate increases in trip distance/moves to more urban neighborhoods. The two subsamples used for the analysis are similar to the full sample, except for mean trip distances that are considerably shorter due to the removal of outliers for both trip purposes. As respondents who do not commute are excluded in the sample for the commute trip model, the share of employed respondents is higher, the share of respondents aged 65 + years is lower than in the full sample.

Table 1 Key characteristics of the samples
Full size table

Modeling approach

We use SEM to analyze our data as this method allows the testing of causal hypotheses in a system of diverse interdependencies. This is especially important when analyzing longitudinal data like that in this study. It can be assumed that two variables reflecting the same property measured on the same individuals at two points in time are strongly related. It is also plausible that variables influence each other over time.

We model trip lengths for commuting and grocery shopping before a move, the change that occurred after a move, and the relationships with other possible influence factors based on our conceptual model (see Fig. 1). The change variables are approximately normally distributed, offering modeling benefits as standard SEM estimation procedures assume the dependent variables to be normally distributed.

The estimation of SEM parameters aims at approximating the modeled covariance matrix Σ(θ) to the empirical covariance matrix S as closely as possible. A chi-squared test statistic and fit indices such as the root mean square error of approximation (RMSEA), the comparative fit index (CFI), the Tucker-Lewis index (TLI), and the standardized root mean square residual (SRMR) are calculated to assess the fit. The models are estimated in R using the lavaan package. Additionally, we re-estimated the models in SPSS Amos 29, which confirmed the results.

The estimation process through maximum likelihood estimation requires the endogenous variables to follow a multivariate normal distribution. In travel studies based on survey data, however, the requirement of multivariate normality is often not met (Bagley and Mokhtarian 2002; Cao et al. 2007). Still, models based on this type of data could perform well, especially when using large datasets. Our data leads to high Mardia statistics indicating deviation from the multivariate normality assumption. We made an attempt to use the log of one variable with high kurtosis and removed observations with a high Mahalanobis distance five at a time until tests for multivariate kurtosis did not lead to significant results. However, during this process, a large share of observations was excluded from the analysis and the interpretation of the log-transformed variable is less intuitive. As SEMs using the reduced/transformed and the full dataset led to similar results, we retain all observations and use the measured values without transformation. To account for the deviation from the multivariate normality assumption, we adopt the robust maximum likelihood approach (MLR). Here, estimation is still undertaken through maximum likelihood, but robust standard errors and a robust test statistic are computed.

Our models include several ordinal variables reflecting the level of agreement, satisfaction, or importance given to certain criteria measured on a five-point Likert-type scale. The scales have been constructed according to Rohrmann (1978), who developed and empirically tested scales for the dimensions frequency, intensity, probability, and evaluation aspects that come as close to an interval scale as possible. This entails the wording of scale levels to be constructed and perceived by respondents as equally spaced intervals. Thus, usage as continuous variables in our models can be justified.

The survey enquired into commute trip lengths by asking about the distance to the workplace/place of education. Therefore, the sample might contain some trips to schools/universities. In our sample, the commute trips of those who indicated they were students were slightly shorter. We account for the specifics of students’ relocations by including a link to the change in urbanity, thus allowing an impact on residential location choice. For grocery shopping trips, the survey specifically asked about the typical trip distance to buy groceries. Thus, the question is concise enough to be understood similarly by respondents.

In Germany, there is no nationwide data available on the spatial distribution of workplaces. Therefore, we measure urbanity through an indicator that combines the number of shops and services within the postal code area (stores & services), the distance to the main station of a central city (main station), and the share of the postal code area within 300 m of a rail station (rail). The chosen variables resulted from correlation analyses between several measures for urbanity and represent different aspects of urbanity. Similar measures were used by Niklas et al. (2020) to classify urbanity for all postal code areas in Germany in a machine learning approach. They found the quality of public transport and the number of POI per km² conveyed more information than population density. The indicator is calculated as follows for all postal code areas in Germany, based on the normalized values for area i, i = 1,…., 8,169:

$${\text{urban structure}}_{i}= {\text{rail}}_{i}+ {\text{stores \& services}}_{i}- {\text{main station}}_{i}$$
(1)

The resulting variable is approximately normally distributed for the postal code areas in Germany (exemplary values for Berlin are included in Fig. 6 in the appendix). High values represent a high degree of urbanity. While there is an obvious link to shopping accessibility, our indicator of urbanity might also serve as a proxy for job density, as in dense urban areas job density is usually higher as well (Krehl and Siedentop 2019).

We construct a latent variable for the preference for urban structures in the search for a new residence based on the importance that is given to the accessibility of shopping opportunities, the workplace, leisure activities, and public transport. In a preliminary factor analysis of 13 neighborhood-related criteria in housing search, these variables were found to construct one of four factors, labeled pro-urbanity.

We collected attitudes towards car travel, public transport, cycling, and walking. In preliminary models, we considered using attitudes towards car travel or towards walking as these modes are related to an inclination towards shorter vs. longer trips: car travel is connected to longer trips, walking is associated with shorter trips and more urbanized structures (De Vos et al. 2021). While cycling is also associated with shorter trips, trip distances are less important than with walking. Public transport cannot be clearly connected to shorter/longer trips. Attitudes towards car travel before and after a move are extremely highly correlated, which led to Heywood cases (the occurrence of estimated variances smaller than 0) and hindered our model from producing reliable results. Therefore, attitude towards walking is chosen in our final models.

Due to the non-linear relationship between age and trip distances found in previous studies (Axisa et al. 2012), we first defined four age groups and performed t-tests for trip distances between these groups. The only significant differences occurred between respondents aged 50–64 (who were found to travel farthest for both trip purposes) and other age groups. Thus, we only include a variable in the model reflecting whether respondents belong to this age group.

We tested the influence of other variables, especially sociodemographic characteristics, on the model. Variables that were not essential in our conceptual model and only had an insignificant influence on the endogenous variables were removed from the model, following the procedure of Tao et al. (2023), De Vos et al. (2021), and Bagley and Mokhtarian (2002). We re-estimated our SEMs for a subset of the data, excluding respondents who moved within a single postal code area. We also performed a control analysis to determine whether the exclusion of respondents with long shopping trips leads to different results. For both SEMs, the significant effects found in the analysis are found to be stable overall, independent of the chosen subsample.

Results

Descriptive analysis

Commute distance

The mean commute distance before a move amounts to 18.1 km (median 9.0 km, standard deviation 56.7 km), see Fig. 2. However, about a third of respondents (33.6%) commute less than 5 km before relocation, about a quarter of respondents commute between 5 and less than 10 km, and only 4.3% commute more than 50 km each way (see Fig. 2).

Fig. 2
figure 2

Commute trip length before and after relocation

Full size image

After moving, the average commute trip length decreases to 15.0 km (median 7.0 km, standard deviation 47.1 km). However, this is mainly driven by decreases in long trip distances, as the median change in trip length lies at +/-0 km. The shares of those with especially short commutes of less than 5 km and 5 km to less than 10 km slightly increase to 34.9% and 29.9%, respectively, while the share of commuters with trips longer than 50 km decreases to 2.0%. However, the share of commutes between 15 km and less than 30 km also increases visibly. There is a considerable difference between respondents who indicated professional/educational reasons for a move and those who did not. While the first group reduces commute trip lengths by 23.0 km on average (median 1.6 km), the latter group keeps trip lengths more or less constant on average, even increasing them by about 1.1 km (median 0 km). The difference between the two groups is statistically significant (t-test, p-value = 0.03).

Grocery shopping

The trip distance for grocery shopping averages 1.5 km (median 750 m, standard deviation 2.8 km) before relocation (Fig. 3). About two-thirds of respondents buy groceries within less than 1 km (64.3%). However, 5.6% travel for 5 km and more, 2.1% even for more than 10 km. After a move, respondents shorten their grocery shopping trips to an average of 1.2 km (median 620 m, standard deviation 2.2 km). While the median change lies at 0 km, there are considerable differences on an individual level. 46.5% of respondents shorten their trips, while 14.9% keep trip lengths constant and 38.7% increase trip lengths. 6.8% decrease and 5.9% increase their trip distances by even more than 2 km, respectively.

Fig. 3
figure 3

Usual trip length for grocery shopping before and after relocation

Full size image

Model results

The two SEMs estimated for commuting and grocery shopping trips fit the data well regarding the fit indices in Table 2. Results are structured by endogenous variables, presenting the estimates for all influence factors on each endogenous variable at a time. We start with the SEM for commute trips (Fig. 4), followed by the results of the SEM modeling grocery shopping trip distances (Fig. 5). Furthermore, the total effects as the sum of direct and indirect effects are indicated in Table 3 for the commute trip SEM and in Table 4 for the grocery shopping SEM.

Table 2 Model fit results, compared to values for an acceptable fit (Hu and Bentler 1999)
Full size table
Fig. 4
figure 4

Output of the SEM for commuting distance (SEM 1), standardized direct effects

* significant at α = 0.05, ** significant at α = 0.01, *** significant at α = 0.001, insignificant α = 0.05: dashed grey lines, in bold: variance explained

Full size image
Fig. 5
figure 5

Output of the SEM for shopping distance (SEM 2), standardized direct effects

* significant at α = 0.05, ** significant at α = 0.01, *** significant at α = 0.001, insignificant α = 0.05: dashed grey lines, in bold: variance explained

Full size image
Table 3 Standardized direct effects in the SEM for commute trips, total effects in parentheses
Full size table
Table 4 Standardized direct effects in the SEM for shopping trips, total effects in parentheses
Full size table

Commuting distance SEM

Commute distance

The change in commute distance after relocation is strongly associated with the commute distance before relocation, and with employment type. Movers tend to shorten longer trips, while those with short commutes also accept an increase in distance after relocation. In line with previous studies, working a full-time job before/after relocation seems to motivate respondents to accept longer commute distances before relocation and increases after relocation, respectively (Manaugh et al. 2010). Respondents working a full-time job before relocation tend to decrease their commute distance after relocation due to the higher distances traveled earlier. We also observe an association between the built environment and trip distances in the expected direction, although only for the cross-sectional data before relocation: respondents living in more urban neighborhoods commute over shorter distances. Indirectly, moving from an urban neighborhood is associated with increases in commute distance, as the overall shorter distances before relocation are more likely to increase. Changes in urbanity, however, are not systematically linked to either increases or decreases in commute distance. Contrary to expectations, lower commute satisfaction before relocation and a professional/educational reason for moving do not lead to decreased commute distances after relocation. Walking attitudes were not found to be associated with shorter commute distances either.

Trip satisfaction

As expected and in line with previous studies (Ye et al. 2020), longer commutes (before relocation) lead to lower commute satisfaction. Mediated by their respective effect on commute distances, full-time employment is negatively associated with commute satisfaction, while higher levels of urbanity are positively associated with commute satisfaction.

Travel attitudes

As respondents’ attitudes before and after relocation strongly correlate, attitudes towards walking seem to be relatively stable. Further, the analysis confirms the expected link to the built environment, but only for the neighborhood before relocation. Thus, living in urban neighborhoods is associated with more positive walking attitudes before and, indirectly, also after relocation. Changes in urbanity or trip distance, in contrast, are not related to walking attitudes after relocation.

Change in spatial structure

We observe a significant and negative relationship between the level of urbanity before a move and the change in urbanity caused by a move, indicating an exchange between different environments among respondents. While students tend to move to more urban areas, respondents with a preference for urban areas as residential locations do not seem to be able to realize this in their searches for accomodation.

Reasons for moving & residential preferences

Respondents who are dissatisfied with their commute tend to indicate a professional/educational reason for moving more often. As commute satisfaction is lower for long commute distances before relocation, we observe a number of indirect effects mediated via trip distances. Thus, full-time occupation before relocation is connected to a higher tendency to indicate a professional/educational reason for moving, while living in an urban area before relocation is connected to a lower tendency to indicate a professional/educational reason for moving. In line with expectations, respondents with favorable attitudes towards walking tend to prefer urban neighborhoods in the search for new accommodation. Indirectly, living in urban areas before relocation is also associated with a greater preference for urban areas, mediated by more favorable walking attitudes.

Grocery shopping distance SEM

Grocery shopping trip distance

Shopping trip distances before relocation are negatively related to the change in shopping trip distances after relocation. In line with expectations, grocery trip distances are strongly related to the residential environment. Trips are shorter in urban locations and trip distances tend to decrease after relocations to more urban neighborhoods. Indirectly, relocations originating from more urban neighborhoods are associated with an increase in shopping trip distance, mediated via shorter distances before relocation and decreases in urbanity. Students and respondents aged 50 to 64 tend to decrease their shopping trip distance, mediated by increases in urbanity and longer trip distances before relocation, respectively. Positive walking attitudes before relocation are linked to shorter trips before relocation and, indirectly, to increases in distance.

Trip satisfaction

As expected, respondents traveling farther to go grocery shopping are less satisfied with the accessibility of shopping facilities before relocation. As respondents aged between 50 and 64 travel, on average, longer, they are less satisfied. In contrast, urban residents and those with a positive walking attitude travel shorter distances and are more satisfied.

Travel attitudes

Attitudes towards walking after relocation strongly correlate with attitudes before relocation. Respondents living in more urban areas before relocation tend to have more positive walking attitudes. Due to the high stability of attitudes, the effect extends to attitudes after relocation. Respondents experiencing increased trip distances tend to have less favorable walking attitudes after relocation. This effect causes a weak but positive association between shopping trip distance before relocation and walking attitudes after relocation: as respondents tend to decrease long trips, their walking attitudes improve.

Change in spatial structure

Urbanity before relocation and change after relocation stand in a significantly negative relationship. Thus, residents of urban neighborhoods tend to move to less urban areas, and vice versa. Respondents with a higher preference for urban areas tend to realize their preference and move to more urban neighborhoods. Students move to more urban areas more often, while households with children under the age of 3 move to less urban locations.

Residential preferences

Respondents’ attitudes towards walking are significantly related to their preference towards urban neighborhoods. A more positive walking attitude and, mediated by walking attitudes before relocation, higher levels of urbanity before relocation are associated with a higher preference for urban neighborhoods.

Discussion

Our cross-sectional analyses confirm the expected association between dense and mixed urban structures and shorter trips for both trip purposes (Ramezani et al. 2021; Scheiner 2010). However, individuals who move to more urban neighborhoods only decrease grocery shopping distances, while no effect is found for commute trips. Thus, dense, urban areas are not per se connected to shorter commute distances. Several factors might be responsible for the lack of evidence for a longitudinal relationship. Firstly, the impact of personal dispositions/long-term lifestyles or expectations towards trip distances that were formed at the previous location might surpass the impact of changes caused by relocation. Secondly, for moves that are associated with a job change, the temporal order of events might differ among respondents, leading to counteracting effects in the model. Thirdly, from a modeling perspective, the relationship between the two variables inherently defined as differences (between the value after and the value before relocation) is likely to be weaker than the relationship between the baseline values as more measurement errors can play into the equation. In contrast to commutes, trip distances for grocery shopping are more strongly related to the spatial distribution of opportunities. Even though respondents might still choose a certain grocery store according to their preferences (Handy and Clifton 2001), the broader choice in urban areas allows movers to urban areas to decrease trip distances.

Overall, both longer commute trips and longer grocery trips tend to be shortened by a move, while respondents with short trips before relocation also accept an increase. Prospective movers with especially long/short trips might compare their trips to those of others and thus be more inclined to decrease/increase their trip distances. The observed changes in trip distances might also indicate convergence towards respondents’ ideal commute time. Counteracting effects of increases and decreases are likely responsible for the median change of +/- 0 for both trip purposes after relocation. Respondents might also maintain similar trip distances after relocation if they had already realized their ideal commute time at the previous residential location. All this supports the notion suggested by previous studies that commute trip lengths play a role in residential location decisions as soon as they exceed a maximum commute tolerance, but movers do not aim to minimize them per se (Bruns and Matthes 2019). Another possible explanation for the negative association of changes and baseline values is regression to the mean, a statistical phenomenon that could also be observed in previous longitudinal studies (Tao et al. 2023).

We observe that only grocery shopping trip distances are subject to travel attitudes. As shorter trips can be undertaken on foot, respondents with more positive walking attitudes actually travel shorter distances to buy groceries before relocation. In contrast to commuting distance, which is usually fixed to a single location, it is possible to choose between several shopping opportunities, allowing trip distance to be adapted in line with attitudes in everyday decisions.

We observe the assumed relationship between residence in urban neighborhoods, positive walking attitudes, and urban preferences in residential location search from a cross-sectional perspective (Handy et al. 2006). Hence, travel attitudes are actually translated into search preferences in residential location searches. However, our models are inconclusive regarding whether a higher preference for urban surroundings actually leads to a significant increase in urbanity after a relocation. This finding challenges the idea that attitude-driven residential self-selection plays a major role. We cannot rule out the possibility that some respondents needed to compromise in their residential decisions, as housing markets in many German cities are tight. Only a few respondents who assessed accessibility of the respective destination as important in their residential location search, are dissatisfied after moving (10% with the accessibility of their workplace/school, 11% with the accessibility of shopping facilities). Thus, the majority found a new residence that fits their preferences regarding the factors under consideration, but a small share was unsuccessful. The observation that only about 17% of respondents increased their commute distance by more than 5 km suggests similar findings. It should be noted that we control for urbanity before relocation when estimating the effect of search preferences on the change in urbanity. A preference for a certain type of neighborhood might have already been realized at the previous residential location. Thus, the dataset only allows analysis of the impact of travel attitudes on the difference between the previous and the new location, not the impact on the level of urbanity overall. Generally, there is a weak indication of travel-related residential self-selection.

Travel attitudes are surprisingly stable in the models, despite natural variation between the responses to the questions at two points in time. In retrospective surveys, an adjustment of previous attitudes to the current ones can be expected (Goethals and Reckman 1973). De Vos et al. (2021), for example, observed considerable changes in walking attitudes after relocation. Given the panel study design, however, the measured consistency is noteworthy. The stability of attitudes is also higher than in similar studies using panel data, possibly attributable to the shorter time span of our panel compared to these studies (Kroesen et al. 2017; van de Coevering et al. 2021). The lack of a relationship between changes in the built environment after relocation and changes in attitudes indicates that a longer timeframe is necessary for attitudes to react to external influences. In addition, the stability of attitudes might indicate the significance of personal lifestyles (Van Acker et al. 2014). Walking attitudes after relocation are negatively associated with increases in grocery shopping trip distance, suggesting that attitudes may be endogenous to behavior. This finding lends support to the theory of cognitive dissonance, but is contrary to widely applied behavioral theories such as the Theory of Planned Behavior, which assume a one-directional causal relationship and treat attitudes as a cause for behavior. If increases in trip distance prevent respondents from walking to the grocery store, cognitive dissonance between attitudes and travel behavior arises if walking is the preferred mode. This leads to a desire to reduce the dissonance by adapting walking attitudes so they subsequently become more negative (Festinger 1962). Unlike grocery shopping trips, even shorter commutes may still be too far to walk. Hence, changes in commute distances usually have no behavioral implications relevant to walking attitudes. Our analysis suggests residential determination in the form of adjustments to travel attitudes only applies with changed trip distances for grocery shopping, but not directly in response to changes in urbanity. The distance traveled to buy groceries might also describe the accessibility of destinations in daily life. This may be more relevant for the walkability of a neighborhood than changes in the overall character of the larger postal code area.

As expected, commute and shopping trip lengths are strong predictors of respondents’ satisfaction with the accessibility of the workplace and shopping facilities, respectively, from their residence (Chatterjee et al. 2020). Lower commute satisfaction, in turn, causes respondents to indicate professional/educational reasons for moving, thus triggering relocations in the first place. In contrast to the descriptive analysis (that may also have been affected by outliers), satisfaction before a move and a professional/educational reason for moving are not found to significantly affect the change in trip length, as trip length before a move is accounted for in the SEMs. A study by Tao et al. (2023) led to the same finding for respondents who moved and switched jobs simultaneously. We assume that counteracting effects might be responsible for the reason for moving being insignificant. Residential relocations motivated by professional/educational reasons may lead to decreased commute distance especially when respondents were long-distance commuters before relocation and succeeded in finding a residence close to their workplace. In other cases, movers might change cities and accept longer distances to their workplaces within the new city. These effects might also explain the diverging effects found in other studies (Cervero and Day 2008; Beige and Axhausen 2017; Prillwitz et al. 2007). Overall, objectively measured trip lengths before relocation are found to be of higher importance than individual perceptions, which underlines their reliability and suggests a need for planning.

Limitations

The panel survey of households before and after relocation allowed us to analyze the changes in trip distances after residential relocations, relying on data collected at the respective time – which has rarely been done before. However, there are still limitations to the analysis. One reason for the weak association between changes in urbanity and changes in trip distance might lie in the size of the spatial unit chosen for analysis and the lack of information on workplace distribution. Postal code areas in Germany can be quite large and cover areas with different levels of urbanity. Additionally, they are permeable, as people can move within or visit opportunities in other nearby postal code areas. Despite the correlation between the density of opportunities and job density, our indicator of urbanity cannot incorporate the locations of large employers, vital information for the commute trip distance SEM.

A regular assumption in studies analyzing the travel impacts of residential relocation is that movers are able to realize their preferences (Chatman 2009). However, housing availability, prices, and the spatial distribution define residential choices (Chatterjee and Scheiner 2015). In our sample, many respondents searched for accommodation in regions with tight housing markets. In these settings, movers might face the impossibility of finding a place to live that meets all requirements and then need to compromise. Therefore, some respondents might not have been able to choose a location freely.

As we model individual trip distances and their relation to relocations as joint household decisions, households involving more than one commuter might additionally consider the partner’s commute distance, which could not be captured in the model. Furthermore, due to the design of our study, our sample contains only recent movers who were surveyed a few months, at most, after their move. Changes that occurred subsequently, such as a job change or deciding in favor of a different grocery store, could not be captured.

Recommendations

The results of this study shed light on the interdependencies between the spatial structure of neighborhoods, travel attitudes, and residential preferences, on the one hand, and trip distances for commuting and grocery shopping, on the other hand, both before and after relocating. The findings are of interest to transportation planning and policy. The cross-sectional analysis confirmed that there is a strong connection between urban neighborhood structures and shorter trips, further encouraging the design of dense, urban neighborhoods to allow for shorter trips that can be made on foot or by bike.

Specifically, in urbanized areas with tight housing markets, an improved supply of housing in urban locations is needed to meet the high demand and to allow movers to reduce distances to frequently visited destinations. Not all respondents with a preference for urban neighborhoods move to more urban areas and not all trips were shortened by relocation. Thus, our sample may include some movers with a desire for an urban lifestyle with shorter trips who were forced to compromise.

The stability of movers’ attitudes in our sample further underlines the need for dense, urban neighborhoods with a potential for short trips. As attitudes were found to react to changes in trip distance, the results suggest adopting measures targeted at travel behavior. For example, offering housing options in urban neighborhoods with a higher density of opportunities or improving opportunities in existing or new housing developments might encourage shorter trips that do not necessitate car travel and might also improve walking attitudes.

Conclusions

In this study, two SEMs were estimated using a panel dataset of movers before and after relocation to determine the impact of residential relocations on trip distances (commuting and grocery shopping). Strong relationships between travel attitudes, spatial structures, and trip distances can overall be found for short-term travel behavior, while other factors determine residential location choice. Grocery shopping trip distances decrease after relocations into more urban neighborhoods, while commute distances do not change systematically. There is also an association between positive walking attitudes and short grocery shopping trips. A strong indirect association between the neighborhood structure at the previous residential location and change in trip distances indicates the importance of long-term relationships, which are closely connected to the observed stability of attitudes. The results suggest that travel-related residential self-selection does not play a major role. Thus, even though a connection between urbanity, positive walking attitudes, and a preference for urban areas can be found before relocation, this preference has a limited impact on residential location choice. In summary, urban areas are found to support shorter commute and shopping trips, while grocery shopping trips are also influenced by attitudes. Residential preferences and attitudes do not exert a strong enough impact on residential location choice to result in shorter trip distances after relocation.

We can make several recommendations for future research based on the findings of this study. As residential location decisions seem to be largely affected by the housing markets in Germany, future studies could look into which types of movers are able to succeed in realizing their search criteria and which types need to compromise. This also leads to the question of whether results differ if the study is set within less constrained housing markets. It may be that stronger built environment effects are found in areas where housing markets are not so tight because there is less spatial mismatch between travel attitudes and travel behavior. Furthermore, the panel study covers not more than a few months after relocation. It would be interesting to extend the time span to determine whether a longer duration of residence leads to more pronounced changes in attitudes and travel behavior.

Appendix

Fig. 6
figure 6

Values on the variable urban structure for postal code areas in Berlin (data: OpenStreetMap contributors 2021)

Full size image