1 Introduction

The large-scale expansion of remote and hybrid work practices in the wake of the COVID-19 pandemic has significant implications for regional science. By reducing the need for daily workplace commuting, remote work has the potential to alleviate the spatial constraints that have historically limited residential mobility. As a result, individuals are afforded greater flexibility in choosing where to live based on their preferences for local goods and services (such as schools, hospitals, parks and entertainment facilities), rather than merely proximity to their workplace. The aim of our paper is to explore whether remote work enables sorting in the spirit of Tiebout by weakening limitations to mobility.

The seminal contribution by Tiebout (1956) posits that individuals “vote with their feet” by relocating to communities that best satisfy their preferences in terms of public goods and taxation levels. In the context of the USA, where metropolitan governance is highly decentralized and municipalities have a high level of public spending autonomy, people moving to their preferred location increase tax revenues to fund that community’s specific bundle of services, enabling local governments to provide different combinations of public goods and cater to heterogeneous preferences. This would lead to communities of optimal size and, arguably, increase the economic efficiency of metropolitan governance overall. This theoretical framework, however, relies on several (unrealistic) assumptions, including perfect mobility, zero moving costs, complete information, no work-related constrains, a large number of communities to choose from and no public goods spill overs from one community to the next.

Italy is characterized by significant territorial inequalities, between the affluent North and the less prosperous South, as well as between urban and rural areas, city centers and peripheries. Although economic opportunities and public service quality significantly impact migration patterns from the latter to the former, which aligns with the Tiebout model's predictions, we argue that the system is not efficient and market failures arise due to the assumptions of the model not holding.

Market failures include overpopulated areas being affected by congestion and unaffordable housing, while peripheral areas suffer from depopulation, aging populations and economic stagnation (De Renzis et al. 2022). Although the presence (or absence) of job opportunities continues to be central in people’s locational decisions, the role of non-economic factors has been increasingly acknowledged, as many of Italy’s less prosperous and more peripheral areas offer natural, cultural and social amenities at a lower cost of living, making them attractive for reasons other than labor market conditions (Sonzogno et al. 2022). Anecdotal evidence from the post-pandemic period moreover points to the so-called South-working phenomenon, where people originally from the South but living in the North return to their hometowns while continuing to work remotely for a Northern employer (Magliaro 2020).

The Tiebout model’s assumptions that are least likely to hold in the Italian context are perfect mobility and no work-related constrains, as the country has a strong in-presence working culture causing workers to commute to the office daily. Before COVID-19, people’s preference of living location was therefore strongly linked to their workplace location, but our hypothesis is that this preference does not necessarily hold if the limit to residential mobility represented by daily commutes is lifted by the possibility to work remotely more often. The reorganization of working practices could therefore lead to a new spatial equilibrium in worker’s living location.

Leveraging the large reach of COVID-19 containment measures, we explore how the newly opened possibility to work remotely changes Italian workers’ propensity to move. We overcome the unavailability of official data on remote workers in the Italian context by collecting primary data through a specifically designed original survey. The survey was launched in early 2022 on a sample of workers in 12 Italian metropolitan areas. We ask workers who expect to continue working remotely in the future about their moving aspirations and behavior. We also gather information on a wide range of socioeconomic, job and personal characteristics.

We focus on jobs that can be performed at home for at least part of the time (i.e., office workers). We thus compare the moving behavior and aspirations of workers who have the possibility to work remotely in the long term, with those of workers who could theoretically work remotely, but do not expect they will be doing so. We make a distinction between those who already moved (since the outbreak of the pandemic) and those with untapped moving “potential.” To limit sample selection arising when naively comparing people who have very different possibilities to move (e.g., access to a second house), we adopt a conditional logistic analysis in the framework of a case–control study, where observations with similar (observable) characteristics are matched in groups, and hence, the likelihood of the dependent variable to take a certain value is calculated within each group of comparable observations.

Our results suggest that some people live where they do because of commuting needs rather than personal preference, and that remote work availability asymmetrically altered locational preferences for people with certain characteristics. This suggests that a degree of Tiebout sorting in the Italian context is held back by a strong in-presence workplace culture, as well as uncertainty.

This article is structured as follows: Sect. 2 outlines our theoretical framework and the existing background literature. Section 3 explains how we conducted our survey design and offers summary statistics. Section 4 discusses our methodology, and Sect. 5 presents our results. Section 6 concludes with some policy implications.

2 Literature background

The literature on remote and hybrid work has expanded rapidly, with empirical studies primarily concentrating on labor-economics aspects. Topics encompass the feasibility of remote work, with certain occupations being more amenable than others (Dingel and Neiman 2020; Barbieri et al. 2022). Remote workers typically exhibit higher socioeconomic status and earning potential (Mongey and Weinberg 2020). Productivity outcomes vary, influenced by task type and home environment (Bloom et al. 2015; Emanuel and Harrington 2023; Gibbs et al. 2023), although mentoring and on-the-job training impacts are often overlooked (Lee 2023). Well-being effects are also mixed, as they depend on parental status (Song and Gao 2020; Arntz et al. 2022) and suitable home offices (Möhring et al. 2021). Despite the downsides of remote work, such as isolation from colleagues and interruptions from family members, workers generally find the arrangement beneficial, particularly because of the absence or reduced frequency of commuting, which can place significant economic and psychological strains (Bloom et al. 2015; Barrero et al. 2021). Employers, however, hold a less positive outlook, with a notable disparity in desired remote workdays between employees (2 days/week) and employer plans (1.1 days/week) (Aksoy et al. 2023).

Studies on how remote work affects urban structures are mainly analytical (Delventhal et al. 2022; Davis et al. 2021; Brueckner et al. 2023) and find suburbanization tendencies, as working remotely reduces the frequency of commuting, which in turn facilitates living further away from the workplace. Empirical analyses have investigated the shift of housing market demand from centers to the suburbs, particularly in large US cities (Liu and Su 2021; Ramani and Bloom 2021; Gupta et al. 2022).

Most of the literature focuses on the USA, which is characterized by a higher rate of internal migration than other high-income countries, making the evidence hard to generalize. Due to differences in language, culture and institutions Europeans are estimated to be 15 times less mobile than Americans when comparing similarly sized regions (Cheshire and Magrini 2006). Reasons for migrating also differ, as natural amenities, such as a favorable climate, play a significant role in the USA (Graves 1980) but less so in Europe. In Italy, long-distance moves are predominantly motivated by economic reasons, while shorter-distance moves are also linked to quality of life factors (Biagi et al. 2011). Given these differences, country-specific evidence is crucial for effective place-based policies.

Testing the Tiebout Hypothesis in the Italian context has its limitations due to its highly centralized metropolitan governance system and the low degree of fiscal autonomy municipalities have, which accounts for less than 10% of total government expenditure (European Committee of the Regions 2024). This renders municipal-level tax competition to attract residents irrelevant, making the original interpretation of the Tiebout Hypothesis less applicable. However, the Italian context illuminates Tiebout’s core intuition that individual and household mobility leads to a more efficient allocation of population across space, while limits to free movement lead to spatial equilibrium distortions, with over or underpopulated areas.

Within the Tiebout literature, our paper borrows from contributions highlighting the role of life cycle stages in locational preferences. Clark and Hunter (1992) find that people with school-age children are drawn to areas with better public schools and, therefore, higher property taxes. Older groups with no school-age children, on the other hand, are deterred from migrating to such areas, but are instead drawn to locations with better public services, including care, and higher levels of income rather than property taxes, as retirement makes them less affected by it (Conway and Houtenville 1998). In Italy, where local tax differentials are less relevant, families with school-age children are still drawn to locations with high-quality schools, but other factors are driving locational preferences more strongly—such as workplace location and grandparents location, as they are central childcare providers in Italy (Tomassini et al. 2003). Findings from Guglielminetti et al. (2021), however, suggest that, since the outbreak of the pandemic and the expansion of remote work, the locational preferences of Italians are shifting from centers to lower-density areas, as these offer larger properties at a lower cost, more likely to be single-family dwellings and to have outdoor space. This suggests a change in the pre-pandemic trade-off between accessibility to jobs and space and a flattening of the bid–rent curve in the Alonso model (Alonso 1964), as working remotely more often reduces commuting costs, and consequently, the premium people are willing to pay to be located near the city center decreases.

We moreover draw from contributions outlining the tendency of people with similar characteristics to colocate (Buchanan 1965; Schelling 1971; Heikkila and Coutin 2024) to examine a wide range of the socioeconomic, professional and personal characteristics of those more (or less) likely to move in the Italian context. Our results are somewhat surprising as they go against the common narrative of remote workers looking to relocate being single, independent and childless digital nomads. On the contrary, we find that singles are much less willing than couples and parents of young children to move away from centers, as they place a higher value in being colocated with other singles.

Despite a growing literature on the labor-economics aspects of remote work in the Italian context (Bonacini et al. 2021; Barbieri et al. 2022), its spatial implications have received scant attention, also due to the still limited availability of data. Empirical contributions include the aforementioned Guglielminetti et al. (2021) and Croce and Scicchitano (2022), who argue that higher living and congestion costs in cities could lead remote workers to relocate away from them. Tantillo and Zucaro (2024) moreover highlight successful case studies of local policies and technology investments in peripheral areas fostering remote working communities. We build on these contributions by shedding light on the potential residential mobility that a stable adoption of remote work could untap.

3 Data

3.1 Survey and data collection

As working remotely on a large scale is a relatively recent phenomenon, official data are currently unavailable and a representative statistical sample in terms of age, gender, living area of Italian remote workers cannot be obtained. To overcome this, we collect primary information through an original survey specifically designed to investigate this phenomenon. Participants are recruited through the panel of respondentsFootnote 1 of the market research agency SWG.Footnote 2 In January 2022,Footnote 3 we launched a web-based questionnaire focusing on 12 Italian metropolitan areas,Footnote 4 initially drawing from a sample representative of the working-age population in terms of age and gender. Employing a non-probability sampling design (Vehovar et al. 2016), individuals from this representative sample were invited to participate in the survey. To ensure a specific focus, screening questions are utilized at the survey's outset, narrowing the sample to workers in professions conducive to remote work. Our sample is thus tilted toward white-collar workers, but as 73% of the Italian workforce is employed in the service sector (European Commission 2023), our descriptive analysis remains informative on how remote work is affecting locational preferences for a large share of workers.

Our analysis focuses on the 26–61 age group, excluding full-time students and unemployed individuals. The survey, excluding respondents relocating or returning from abroad, yields a final sample of 2068 individuals. It captures personal and socioeconomic characteristics, job details and, crucially, insights into remote work experiences at different pandemic phasesFootnote 5—before the pandemic (when working remotely was rare), during its more acute phases and its more moderate ones. As for the post-pandemic period, the working culture of Italian organizations continues to place a higher value on working in person over remotely, so we expect that many workers will be required to be present at the office on a full-time basis, despite their role being compatible with remote or hybrid work.

Questions also explore moving behavior, inquiring whether respondents moved to work remotely from a different location than their pre-pandemic residence, for at least part of the time (e.g., from a second house). We expect those owning or having access to a second/family house to be much more likely to move than those not having access to one. We, however, account for the higher moving likelihood of these individuals through our methodology.

Additional queries delve into the respondents’ pre-pandemic living location and whether they were born or raised in the city they work in or migrated there. We expect many workers from the less affluent South to live and work in the country’s North, where most of the economic opportunities are located. We also expect these workers to feel a weaker connection to their living location relative to those born or raised there and, therefore, to be those with the highest moving propensity.

3.2 Summary statistics

After removing missing values in key questions, we have 1633 valid observations. There is no apparent bias among non-respondents, such as in gender or other factors, ensuring our final sample is comparable to the original one. Tables 1, 2 and 6 (in Appendix) provide summary statistics of our main questions on remote working.

Table 1 Summary statistics for the responses on moving propensity.
Full size table
Table 2 Summary statistics for the categories of moving propensity.
Full size table

Table 1 shows the distribution of responses in our sample regarding relocation during the pandemic. About 6.61% of participants chose option A, indicating a full-time move to a different location. A higher percentage, 10.29%, selected option B, indicating a move to remote work for part of their time while maintaining their original living location. Option C, indicating no move but an interest in doing so, was chosen by 21.19% of respondents. The majority, 61.91%, opted for option D, indicating no move and no interest in doing so.

In line with other studies on Tiebout sorting at the local level (Grassmueck 2011; Craig 2024), we observe that most migration took place locally, where the model’s more extreme assumptions—i.e., zero moving costs, perfect information and no work-related constrains—are more likely to hold. About half of those who already moved for at least part of the time did so within the same municipality in our full sample. Surprisingly, the majority of intra-city relocations (about 20%) took place in Bologna, a city of about 400,000 inhabitants, while relocations within the two most populous metropolises accounted for about 18% for Rome and 16% for Milan, and intra-city relocation for the other cities in our sample is marginal. A possible explanation for Bologna’s high intra-city relocation rate is its notably high cost of housing, which is higher than Rome’s in cost per square meter (Immobiliare.it 2024). Relocations outside the municipality suggest the presence of sorting at an inter-metropolitan scale, with the mean relocation distance being 225 km and the median 135 km. A general pattern of large metropolitan area residents moving to smaller cities or towns can also be observed, with more than half of those who left the municipality doing so from Milan (28%) and Rome (23%). These data suggest that some metropolitan area residents are willing to move or spend more time in small- and midsized cities, and that working remotely more often might enable them to do so. We are, however, cautious in making assumptions based on these preliminary data, as longer-distance moves in our data might still be held back by pandemic-related mobility limitations and uncertainty over future remote work availability. Moreover, moving is an aspiration that can take time to achieve given the legal, work-related and family constrains. We thus decided to focus on the likelihood of people moving, irrespective of moving location, leaving more detailed analysis to future research.

We utilized the responses in Table 1 to create three binary variables for our empirical analysis, summarized in Table 2.

The first dependent variable, “moved already,” indicates whether respondents actually relocated. It takes the value 1 if they selected answers A or B, and 0 if they chose answers C or D. We included those who moved for part of their time to capture the overall impact of the pandemic, irrespective of intensity. Even partial relocation signifies a behavioral change compared to the past.

The second dependent variable, “moving potential,” includes all respondents who either moved or express an interest in doing so. It takes the value 1 for answers A, B or C, and 0 for answer D. While “moved already” accounts for 16.9% of respondents, “moving potential” represents 38.09% of them, indicating untapped moving potential in the Italian context.

The third dependent variable, “untapped moving potential,” specifically captures respondents who did not move but express an interest in doing so. It takes the value 1 for answer C and 0 for answer D, with other observations resulting as missing. This variable enables us to isolate workers representing untapped moving potential and examine their characteristics.

Table 6 in Appendix provides summary statistics for our explanatory variables. Concerning our variable of interest—expected remote work frequency in the medium to long term—approximately 28% of respondents anticipate returning to full-time in-person work. Additionally, around 35% expect to work remotely for 1 or 2 days a week. These figures align with Italy's prevalent in-person workplace culture, suggesting limited potential for significant relocations away from office locations. However, 24% of the sample anticipates working remotely more frequently, specifically 3 to 4 days a week, while approximately 13% expect to work remotely full-time. Collectively, these respondents constitute about 37% of the sample, indicating a higher flexibility to move farther from their office locations. Analyzing whether they have actually relocated or express interest in doing so enhances our understanding of Italians’ relocation behavior and assesses whether increased remote work accessibility can lead to spatial sorting among workers.

Turning to other variables, with the exception of the under-30 group, respondents are fairly evenly distributed across age groups. However, our sample's age distribution diverges from that of the overall Italian population. As previously noted, our sample comprises only workers in professions conducive to remote work, leading to a relatively younger composition, given the digital skills often associated with younger individuals.

Regarding gender, our sample skews slightly male, accounting for about 54%. Geographically, respondents were predominantly located in the North-West (34.8%), North-East (11.2%), Center (27.4%), South (18.2%) and Islands (8.4%), mirroring the broader population distribution in Italy. Approximately 70% of respondents were born or raised in the city they lived in before the pandemic, while 30% relocated at a later stage in life.

Our sample diverges from the overall Italian population in terms of household income and education level. Sixty-two percent report a gross yearly household income between €26,001 and €75,000, and 49% hold a Master's degree or a PhD. These findings suggest a predominantly white-collar workforce, more likely to engage in remote work.

Most of the sample (73.4%) holds a permanent contract, 9% have a fixed-term contract, 4% are occasional workers, and 13.6% are self-employed, indicating a sample with relatively high bargaining power. Eighty-five percent continue to work for the same organization as before the pandemic, with 15% changing employers. Additionally, our sample leans toward employees in large organizations, with 41% working for entities with more than 250 staff, suggesting concentration of potential remote workers in larger firms, which generally have greater resources for remote work.

Regarding personal relationships, 71% of respondents cohabit with a partner, 5% have non-cohabiting partners, and 24% are single. In terms of family structure, 43% have no children, 47% have minor children under 18, and 10% have adult children. The majority own their property (75%), while 17% rent, and 7% are hosted by family, friends or a partner. 34% own or have access to a second home, while 66% do not.

4 Methodology

As a first step, we run a logit model to identify whether people who expect to work remotely more often in the long term are also more likely to have moved or be interested in moving in the future (Table 8 in Appendix). The finding of people having access to a second house being significantly more likely to have moved already is in line with our expectations. However, these estimates are potentially biased, as we are comparing people who have very different propensities to move to begin with. For example, working remotely more often would make someone who has access to a second house much more likely to move for at least part of their time than someone who has no access to a second house.

To limit these issues, we run a conditional logistic regression for matched case–control groups (Hosmer et al. 2013). While scarcely employed in social sciences, this approach is widespread in biostatistics, epidemiology and medical statistics (e.g., Pearce 2016; Kuo et al. 2018). Conditional logistic analysis in the framework of a case–control study differs from a regular logistic regression as observations with similar (observable) characteristics are grouped, and hence, the likelihood of the dependent variable to take the value one (i.e., moving in the present study) is calculated within each group of comparable observations. A key advantage of this approach is that the conditional probability of moving as a response to different levels of remote work frequency accounts for observable confounding factors by designing a data structure matching the demographic characteristics of the surveyed individuals.

Specifically, each group of observations is made of a “case” and a number of “controls.” Cases are identified based on the values of the dependent variable. For a value equal to one in the dependent variable, we have a case. To identify controls that enter the same group of a case, we adopt a propensity score matching procedure based on the probability that a respondent moves given his observable characteristics, except for remote working frequency. Subsequently, we generate groups where for each case we assign a number k of controls that fall within a caliper of a quarter of a standard deviation of the propensity score of the case. This is a rule of thumb meant to reduce the bias on all the covariates included in the propensity score and leads to quality matches (Rosenbaum and Rubin 1985).

We consider different options for the number of controls k within each group. In our baseline estimation strategy, we allow each case to match one control, as this setting leads matches to be as similar as possible, though it limits the sample size and the precision of our estimates. In an extension of the analysis, nonetheless, we adopt a 1:2 matching scheme as a robustness check. Moreover, we run the matching alternatively with and without replacement, thus allowing each control to be matched to one case only, as well as be used as a control for several cases.

In terms of the covariates selected for the propensity scores, we included variables related to moving propensity, i.e., having access to a second house, whether they are renting or have a property home, their age group, gender, household income, level of education, whether they cohabit with a partner, whether they have dependent children, where in Italy they live, them being born or raised in the city of their employer, their contract type and the size of the organization they work for. In terms of the variable of interest (i.e., the expected remote work frequency in the long term), it could be related to covariates such as the respondent’s contract type and the size of their organization, which are already included in the matching. Another characteristic influencing whether a respondent expects to work remotely frequently in the long run is their specific occupation, which is a piece of information we did collect in our survey, but the responses are too heterogeneous to be codable, so we chose to not include this variable in the matching.

As having access to a second house makes respondents much more likely to have relocated since the outbreak of the pandemic, we imposed an exact matching on the variable indicating access to a second house (rather than based on the caliper defined above), to ensure that none of the respondents with no access to a second house is compared to those who have one. Moreover, we decided to impose an exact matching on categories of household income and education to generate groups with comparable economic and human capital attributes.

After having checked that these covariates are not collinear (see Table 7 in Appendix), we included them all to calculate the propensity scores, following Stuart and Rubin (2008) who underline the importance of including a large set of covariates to obtain quality matches. The covariates we selected are moreover not affected by either the variable of interest or the outcome, so their selection should not lead to bias (Frangakis and Rubin 2002; Imbens 2004).

After the matching, a variable Matches is created to identify matched cases and controls by giving them the same group number γ. We then proceed by running the conditional logistic on our matched sets. The empirical model we used is as follows:

$${\text{MovProp}}_{i\gamma } = {\text{ExpRemoteWork}}_{i\gamma } + {\text{ Personal}}_{i\gamma } + {\text{LocPrePand}}_{i\gamma } + {\text{Job}}_{i\gamma } + {\text{Living}}_{i\gamma } + \varepsilon$$

where MovPropiγ is a dummy indicating whether individual i part of group γ has moving propensity which, depending on the dependent variable we are running our regression on (listed in Table 2), can either mean he moved already, has moving potential or has untapped moving potential. ExpRemoteWorkiγ is our regressor of interest representing the expected frequency of remote work in the long term for individual i part of group γ. Personaliγ is a vector of personal characteristics including individual i’s age group and gender. LocPrePandiγ includes the area of Italy individual i used to live in before the pandemic’s outbreak, as well as a dummy indicating whether he was born or raised in the city. Jobiγ then includes variables relative to the individual i’s job status, organization’s size and a dummy on whether he changed employer since the outbreak of the pandemic.Footnote 6 Livingiγ includes information on the individual i’s living situation, including whether he cohabits with a partner, has minor children and whether he rents or owns his house. Although we already calculated the propensity scores on these covariates and matched cases and controls based on their propensity scores being similar, it is standard procedure to also include the covariates used to calculate the propensity scores as controls in the conditional logit (Pearce 2016). This is because similar propensity scores still entail a degree of variability in individual characteristics between matched cases and controls.

Finally, ε represents an idiosyncratic error term. For each covariate, the likelihood of the respondent i part of group γ having MovPropiγ = 1 will be estimated within-group γ.

4.1 Matching quality

After using the covariates to calculate the respondents’ propensity scores, and having matched cases to controls with similar propensity scores, we checked the quality of our matches. As shown in Table 3, for the “moved already” dependent variable, and in Tables 9 and 10 (in Appendix) for the “moving potential” and “untapped moving potential” variables, cases and controls have nonsignificant differences in means for all the covariates after the matching. We thus managed to significantly reduce the possible bias improving the accuracy of the estimates of our baseline logit model.

Table 3 Propensity score test results, matching on dependent variable “moved already”
Full size table

5 Results and interpretation

5.1 Conditional logit

Table 4 summarizes the estimates obtained through the conditional logit. The results are presented as odd ratios, with robust standard errors in parentheses. All the included variables are categorical; hence, the coefficient for each category can be interpreted as the difference in moving propensity relative to the reference group.Footnote 7

Table 4 Results of conditional logits on the “moved already,” “moving potential” and “untapped moving potential” response variables
Full size table

We observe that the expected remote work frequency in the long run is positively associated with having moved already (column 1) and moving potential (column 2). The respondents expected to work remotely 1–2 days per week in the long term have 93% higher odds of having moved already compared to respondents who expect they will mainly work in presence.Footnote 8 When they expect to work remotely 3–4 days per week, they are 3.7 times more likely to have already moved, while when they expect to work remotely full-time they are 3.1 times more likely to have already moved. Likewise, column 2 shows that when respondents expect to work remotely 1–2 days per week in the long term they have 50% higher odds of having moving potential (i.e., having either already moved, or being interested in moving) compared to respondents who expect to work in presence mainly. When they expect to work remotely 3–4 days per week (full-time), they have 90% (96%) higher odds of having moving potential.

As for those who moved already, the association between moving and remote work appears to be strongest for people expecting to work remotely 3–4 days a week. Being able to commute less often makes them more inclined to moving further away from their office. On the other hand, an interpretation for the stronger association among those expecting to work remotely 3–4 days, compared to those expecting full-time remote work, could be attributed to the significant lifestyle changes experienced by hybrid workers post-pandemic. In our sample, a notable connection exists between having worked remotely before the pandemic and the anticipation of full-time remote work in the long term (Table 11 in Appendix outlines the details), suggesting that this experience may be less novel for these individuals. Some may have already moved to their preferred location before the pandemic’s outbreak, making their results less strong—although still positive and significant—than those of frequent hybrid workers. Our hypothesis is that individuals experiencing the most profound change due to the availability of remote work were more likely to alter their locational behavior. Remote work expanded possibilities that were previously unavailable to them, broadening their choice set and potentially influencing their preferences for a different location.

In terms of age groups, workers under 40 years of age are significantly more likely to have already moved while moving potential significantly decreases for workers older than 45. Gender does not play a significant role in those who moved already, but most of those having moving potential are male. Workers with moving potential are much more likely to live in the North—which is unsurprising, as most of the economic activity is concentrated there—and those who already moved are much less likely to have done so from the South. Those who already moved are also more likely to not be born or raised in the city of their employer. These findings are in line with the anecdotal evidence of the so-called South-working phenomenon, which refers to people originally from the less prosperous South living in the North for lack of opportunities in their hometowns, who now wish to return home while continuing to work remotely for a Northern employer (Magliaro 2020).

Compared to workers with permanent contracts, those with temporary work are much more likely to already have moved, while self-employed are the least likely to have already moved, although having a significant moving potential. This finding is in line with the post-pandemic worldwide phenomenon of the “Great Resignation,” relative to a record-high number of people quitting their jobs in search of better working conditions and work–life balance after the pandemic ignited a reconsideration of life’s priorities. The outcomes of the next variable relative to whether the respondent changed employer since the outbreak of the pandemic supports this interpretation, as those who changed their employer seem much more likely to have moved already and to have moving potential. This is in line with the idea that people are giving more value to choosing their preferred living place rather than keeping their job.Footnote 9

In terms of organizational size, unsurprisingly, people from larger organizations are much more likely to have moving propensity, with workers in large companies with over 250 staff having the largest moving potential.

Individuals cohabiting with their partners are significantly more likely to have already moved compared to single. The latter might prefer to reside in a location offering more opportunities for socialization, facilitated by both the office and the city center. This finding supports the Tieboutian clubs literature’s argument that, while choosing their living location, people are not indifferent to the characteristics of those in their same location (Sandler and Tschirhart 1997).

Respondents with non-cohabiting partners show the highest moving potential, which is unsurprising, as they likely aim to reunite with their partners. In contrast to respondents with no children, those with young children up to 18 years old exhibit high moving potential. Conversely, respondents with adult children generally show lower moving potential overall, but among those who have already moved, they were the most likely to have done so. Regarding housing arrangements, those renting (rather than owning) their living spaces are significantly more likely to have already moved and have moving potential.

Discussing overall moving potential, however, does not provide insights into specific groups interested in relocating but facing obstacles. It is essential to understand these specific groups to inform policymakers about whom they might empower to relocate, if they chose to facilitate remote work. To this aim, we added a column titled “untapped moving potential” to our analysis, from which we can draw the takeaway that most untapped moving potential is represented by parents of minor children. This suggests that many families are raising their kids in a place that is not to their liking and are reconsidering their preferred living location. Overall, it appears that working remotely grants people more freedom in planning their personal and family lives.

5.1.1 Robustness checks

We test the robustness of our estimates by rerunning the nearest neighbor matching in different settings. In particular, we increase the maximum number of controls per case from 1 to 2 and allow matching with replacement, i.e., controls are allowed to serve as matches for more than one case. Matching cases to controls 1:1 leads to the highest number of cases finding a good control to match with and to the least number of cases remaining unmatched, but leaves us with fewer observations to run our analysis on. As in standard matching procedures, a compromise between matching quality and the precision of the estimates should always be made when running a case–control study. Table 5 presents and compares the results for the different combinations of settings for the “moved already” dependent variable. Tables 12 and 13 in Appendix outline the results of the robustness checks for the “moving potential” and “untapped moving potential” variables, respectively. We can see that, although the magnitude of the correlation and the level of significance might slightly vary by adding controls or changing the with/without replacement setting, the substantial interpretation of the results remains unchanged.

Table 5 Robustness check conditional logits on the dependent variable “moved already”
Full size table

6 Conclusions and policy implications

This paper provides insights from a European country's perspective into the extensive literature on Tiebout sorting processes. Our focus is the Italian context, marked by significant territorial disparities and a prevailing culture of in-person workplace attendance, which limits the possibility to live farther away from the office. We shed light on the role of remote work in reducing the mobility constrain of daily office commutes, enabling sorting in the spirit of Tiebout.

To address the challenges of analyzing this phenomenon amid a lack of official data in a still-developing context, we collected primary data through an original survey we specifically designed for this purpose, in which we ask respondents hypothetical questions on their moving behavior in relation to remote work availability. We shed light on the complexity of the factors influencing moving at the individual level by adopting a conditional logistic regression for matched case–control groups. This approach matches observations with similar observable characteristics into groups, calculating the likelihood of the dependent variable taking a specific value within each group of comparable observations. This method accounts for individuals’ differential possibilities to move.

Our results suggest that some people live where they do because of commuting needs rather than personal preference. Individuals most likely to move are those under 40, living in Northern cities, not born or raised in the city of their employer, cohabiting couples rather than singles, those not cohabiting with their partner, parents of minor children and renters rather than homeowners. Overall, these findings suggest a willingness of younger workers and families living in economic centers to relocate.

By introducing original data on a context underexplored by the literature, our paper brings fresh evidence to assist legislators in drafting more informed place-based policies. Our findings have a number of implications. First, encouraging remote work would empower more people to decide where (and how) to live and raise their family. Second, uncertainty about long-term in-presence office attendance requirements should be limited as much as possible, as it holds back many in their locational decisions. This suggests the need to properly regulate and stabilize remote work so that both managers and employees have clear guidelines and standards to conform to.