Does frequent internet use increase the propensity to change address? UK evidence from Understanding Society on mobility preferences, expectations and moves

Long ‐ term declines in rates of internal migration have been widely documented in many developed economies. Accompanying this decline has been a proliferation in the everyday use of new technologies such as the internet. New communication technologies have been theorised to influence migration decisions, but the direction of this influence is ambiguous, with some studies finding that they decrease migration, others that they increase it. Using data from the UK Household Longitudinal Study, this paper assesses the relationship between internet use and preferences/expectations to change address as well as the decision to move. Although frequent users of the internet tend to be more mobile on average, longitudinal analysis offers no evidence that internet use influences migratory preferences and expectations or actual moves. This paper contributes to the literature by studying a developed country with high levels of internet penetration, by engaging not only with actual moves but also with preferences and expectations to move, and by drawing on longitudinal data to address unmeasured confounding in so far as possible. More broadly, it provides a reference point for future studies which might consider whether internet use plays a different role in the aftermath of the COVID ‐ 19 pandemic.


| INTRODUCTION
This paper seeks to understand how use of the internet influences address-change preferences, intentions, and then whether a move takes place (or not).There are two main rationales for this focus.
First, there has been a trend towards falling levels of internal migration in many countries (Champion & Shuttleworth, 2017).This has been seen not only in high-income countries such as Japan (Fielding, 2018), the USA (Cooke, 2011), and Australia (Bell, Charles-Edwards, et al., 2018) but also in lower-income nations in Africa and South America (Bell, Charles-Edwards, et al., 2018).There is no firm conclusion yet as to why there have been these falls; a nonexhaustive list of the possible explanations that have been advanced include population ageing, changing occupational and geographical labour market structures, a culture of 'rootedness', and new technologies that substitute for migration (Green, 2018).It is this last explanation we aim to assess given the context of Zelinsky's Migration Transition Model. Zelinsky (1971) hypothesised in a (then) future super-advanced society in Stage V of the model, that new forms of spatial mobility and new instantaneous modes of communication would substitute for the need to change address.Importantly, new modes of communication, such as the internet, would act as a stage effect in advanced economies but might also be a general period effect in all countries regardless of income level given their current global prevalence (World Bank, 2020).This possibility is recognised by Zelinsky's comments on countries missing out stages of the transition and the importance of the global context.It is therefore important to understand more about how the internet influences spatial mobility decisions and behaviour, especially given the impact of the COVID-19 pandemic, which hastened the use of technology for remote working among a large swathe of the workforce.Second, the impact of internet usage on migration is by no means clear and unambiguous in the literature.Some studies (Cooke & Shuttleworth, 2018;Shuttleworth et al., 2019), for instance, claim that individual internet use is associated with decreased propensities to move whereas others suggest a positive relationship between internet use and migration plans (Vilhelmson & Thulin, 2013).It is, however, by no means clear prima facie whether internet use would have a positive or a negative effect on moving.Green (2018), for example, concludes that the net impacts of the internet on shorter-and longer-distance migration are unclear with reasons to believe that it might have some positive but also other negative effects.Given this, there is a need for more evidence to assess the importance of internet use and the direction of its effect.
The paper seeks to contribute to the literature in the two areas identified above, investigating whether internet usage is associated with greater residential immobility and whether it might thus explain part of the internal migration decrease observed in high-income countries but also in some with low income.It does so by using Understanding Society, a UK longitudinal survey resource.This offers better prospects for this type of analysis than some of the other datasets used (e.g., Cooke & Shuttleworth, 2018;Vilhelmson & Thulin, 2013) as it is fully longitudinal with successive waves which collect information on preferences and expectations of changing address as well as subsequent address changes between waves.Moreover, information is not only collected on whether respondents used the internetyes/nobut also on their frequency of use.This is important because internet use has expanded through time.
Twenty years ago, internet users were a minority, and perhaps different from the general population, but now the internet is used by the majority of people, so they are more likely to be similar to the general population, being differentiated only in how often they use it.
Furthermore, given the range and sophistication of the data collected by the study, it is possible to not only adjust our statistical estimates for a host of observable confounding factors, but also to relate changes in internet use to changes in migratory preferences, expectations, and realised migration outcomes, enabling further adjustment for time-invariant confounders which cannot be accounted for in a cross-sectional analysis.The next section expands in more depth upon the issues identified above in a literature review of internet use and migration.After this, the Understanding Society data, its preparation, and the methods used for analysis are introduced.Following this, the key results are presented (with additional tables provided in the Appendix) and, finally, the findings are assessed to see what evidence they offer in conclusion.

| BACKGROUND
There is a growing literature on the impacts of the internet on migration.It is wide ranging, covering international migration, internal migration (the focus of this paper), actual moves, and migration intentions and plans.Theoretical expectations of the impact of internet usage are unclear with some reasons to believe that they may be positive but equally valid reasons to argue that they are negative (Green, 2018), and these contradictions continue into empirical studies, some of which find that it increases migration propensities (e.g., Vilhelmson & Thulin, 2013) and others that it has the opposite effect (Cooke & Shuttleworth, 2018).Besides the variety of migration outcomes considered, there are perhaps other reasons for these contradictory findings and expectations.One is that the internet influences different demographic groups in different and perhaps opposing ways.For example, the internet may influence migration decisions for people in employment through enabling remote working, permitting migration as the spatial tie to the workplace is weakened, but also may hinder migration through encouraging more local relationships within neighbourhoods thus holding people in place for workers and non-workers alike.A second is that the studies have been done in different world areas, with differing economic and social contexts, and varying degrees of internet penetration.A final and related reason is that as internet users have grown as a share of the population through time, they have become more like that population, instead of when they were a minority in the past which could presumably be more easily differentiated from the majority.Internet usage may influence migration intentions and decisions in at least three ways (Cooke & Shuttleworth, 2018).Firstly, it might enable access to work and education remotely.Second, it provides more information about places-where a prospective migrant currently lives and to where they might be considering a move.Thirdly, it could weaken or strengthen ties to the neighbourhood of residence.Each of these aspects will be discussed in turn but it should be noted that they will tend to operate at different spatial scales, the first two above mainly for longer distances, the third for all distances including shortdistance moves.
The balance of views on remote working and access to education via the internet tends to be that it will limit internal migration.The internet may make longer commutes more bearable but more importantly it allows people to work from home; they will thus not need in some cases to relocate closer to a workplace to minimise commuting (Ettema, 2010).Moreover, whilst there are some negative implications for work-life balance, remote working via the internet on the whole increases job satisfaction meaning that people will, everything else being equal, be less likely to change jobs, and thus less likely to move house given the relationship between life events and spatial mobility (Bellmann & Hübler, 2020;Felstead & Henseke, 2017;Nam, 2014).Set against this is the caveat, noted by Felstead and Henseke (2017), that the rise of remote home working may facilitate a flight to the country, for people shunning the commute and searching for better housing in a better environment.
This could increase migration for some groups.It is thus hard to assess the balance between immobility and mobility with crosssectional data-longitudinal data are needed.A one-off 'flight to the country'-or several moves to places of increasing environmental utility-in one period could be followed by sustained immobility afterwards, for instance, with internet usage having different impacts at different stages of the life cycle in various residential contexts.
Much of the evidence on the impacts of the internet on migration via information flows is that it has a positive impact, but again this evidence is not conclusive, with reasons to expect a negative effect.Lin et al. (2019) in the USA note that migration searches on the internet are good predictors of migration flows.
Similarly, in the African context, Martin and Hayford (2021) and Grubanov-Boskovic et al. (2021) suggest that internet usage is positively related to migration desires and intentions for international and internal migration.In an analysis of Chinese data, Qin et al. (2016) comment that those who use the internet tend to have to move further than those who do not, while in the USA, Palm and Danis (2001) found that internet users had wider housing search radii than non-users.However, Winkler (2017) argues that more internet usage leads to decreased international migration and Cooke and Shuttleworth (2018) suggest that increased information electronic flows lessen the costs of not moving by allowing non-migrants to remain in contact with those who have left, and also to access a wide spatial range of opportunities.Moreover, better information about other places reduces the chances of failed migrations that 'test the water' elsewhere but end in return.
The final pathway through which the internet might influence migration decisions and behaviour is through its impacts on attachment to place.Place itself is a slippery and multiscale idea, involving ideas about housing quality and neighbourhood.Spatially it can range from the household and its immediate environs, through the neighbourhood (or community), to the larger spatial scale of the city (Massey, 1995); furthermore, it can also vary between different types of place.Conceptually, it might involve general feelings of topophilia (Tuan, 1974), interactions with friends and family (including notions of social capital and community strength), perceptions of the environment, and participation.In short, its measurement is difficult and multidimensional (Raymond et al., 2010).The general expectation in social science theory has been that place, place attachments, and local ties have been eroded by technological social and economic change in a modernisation process that started with the transition from traditional rurality to industrial and urban life (Bjarnason et al., 2021) and which has ended in a late Capitalist postmodern mobile world (Sheller & Urry, 2006).This seems to be logical with place attachments weakening through time under the assault of technical and social developments (of which internet usage is merely one).Despite this, however, there is some empirical evidence that internet usage may actually increase place attachment through increasing local social interactions and embeddedness in the community.Shklovski et al. (2008), for instance, claim that internet and ICT use increase local relationships, as do Birnbaum et al. (2021), who argue that digital media use and the internet mean that there is more information about the localities where people live and greater interaction, resulting in stronger ties to place.The concentration on all-address changes in the present study (for reasons of data counts) means that most emphasis is on place attachment as the most important mechanism in this case since better information about places is likely to be less important for local moves (although the possibilities of remote access to work and education remain in play to some degree as a causal mechanism).
The interpretation of this brief review is not conclusive or simple; there are ample theoretical reasons to assume that internet usage could have either negative or positive impacts on migration.Likewise, the empirical evidence is equally mixed.One reason for this is that internet usage could have differential and contradictory effects on migration desires, intentions, and actual behaviour across different demographic groups; much thus might depend on what is measured for which group, and the balance of factors in play in different study areas.Another reason is that the studies reviewed are based on evidence from various world areas and time periods.This means that in some older studies (or studies in lower-income countries), the internet is a minority pursuit, but in more recent analyses and in higher-income countries it is something for the majority; hence the selectivity of internet users varies as does the context and social implications of the internet.It is thus no wonder that different effects are observed in different analyses.
The analysis that is developed in this paper speaks to these debates and gaps.It adds to the evidence base drawn from higherincome countries and is therefore more comparable with other work in, for example, the USA and Sweden, and other countries with high internet penetration (the internet was used by 95% of the UK population in 2020 according to the World Bank, 2020).Specifically, it builds on empirical studies that have applied regression techniques to individual-level data to assess the relationship between internet use and actual moves.Instrumental variables estimation was applied to Northern Ireland data by Cooke and Shuttleworth (2018) and to US data by Cooke and Shuttleworth (2017) in an effort to deal with unobserved confounding factors.Both studies found negative effects of internet use on the probability of moving.Specifically, it aims to answer questions about whether a greater frequency of internet usage (a more meaningful question than whether the internet is used, yes or no, given its prevalence) leads to (i) stronger preferences to move, (ii) expectations of moving, and (iii) changing address before and after controlling for other factors.It considers these questions for the whole population, and then for the separate group of just those in employment to see if there are consistent effects.This latter group, of course, is of interest given the Three binary outcome variables measuring internal migratory behaviour, expectations and preferences were selected for analysis.
The primary outcome indicates whether someone changed address postcode between survey waves, distinguishing movers from nonmovers.The second distinguishes between those who prefer to move house and those who prefer to remain where they are, while the third indicates whether someone expects to move house within the next year (vs.expects to stay).Most housing moves are short-distance so the analysis concentrates on all-address changes as the main metric.
Migratory moves of >20 km and >50 km were also analysed but cell counts become low in these cases for frequency of internet use so these results were less reliable.These results are therefore not reported but are available on request.
The effect of internet use on each outcome is estimated using multivariate linear regression analysis, enabling adjustment for multiple confounding factors correlated with internet use and migration behaviour.A potentially important confounding factor is age: because younger people tend to be more mobile and to use technology more than older people (Volkom et al., 2014), failing to account for the younger age profile of internet users would lead to an overestimate of its effect on migration (upward bias).Other factors besides age should also be considered according to previous studies (Green, 2018).As such, the regression equation used to estimate the effect of internet use is specified as follows, where subscript i denotes individuals and subscript t denotes time (wave): where i n = 1, …, and t T = 1, …, .
The binary dependent variable y it represents individual i's outcome at survey wave t, and ϵ it is an idiosyncratic error term.
The variables prefaced with 'Int_' are dummy variables capturing the frequency of internet use at the previous available survey wave, where the parameters δ δ , 1 2 and δ 3 represent, respectively, the effects of daily, weekly and monthly internet use relative to the reference category of never using the internet or having no internet access.These variables are lagged because address changes take place between survey waves, and it is necessary to ensure that internet use is determined before the decision to migrate; although this is not required for the 'expect to move' and 'prefer to move' outcomes, we retain this approach for consistency since the results are no different otherwise.The set of control variables is comprised of two groups: X it represents controls measured contemporaneously (dummies for age group, gender, ethnic group and survey wave) and X it−1 represents controls measured at the previous wave (dummies for marital status, health status, economic status, educational qualifications, household tenure, number of household residents, urban-rural location (where urban refers to living within a settlement with a population of 10,000 or more) and region of residence), where lags are taken here for the reason stated above.For example, region of residence reflects where people live before a potential move.
Although the set of control variables enables adjustment for measured differences between individuals, there may still be unmeasured differences causing the estimated effect of internet use to remain biased.To deal at least partially with this issue, we take advantage of the longitudinal structure of the data to include person fixed effects.In Equation ( 1 Likewise, conclusions are unchanged when alternative estimation approaches are used, for instance when estimating logit models or when using unlagged variables for the 'expect to move' and 'prefer to move' outcomes.These checks are not reported but are available on request.Finally, because we observe repeated measurements, implying that observations are not independent, standard errors are clustered by individual.

| All respondents
Tables 1 and 2 summarise the outcome and the main internet variable of interest across all waves.Table 1 shows that around 33% of the sample wanted to change address at one time or another, 11% expected to move house, but only 7% actually moved house during the analytical period.Internet non-users are in a minority of only around 8% with 93% using the internet either monthly, weekly, or daily (note: there is a rounding error).This is close to the 95% UK estimate cited earlier from the World Bank (2020).Table 2 gives information on moving desires, expectations, and behaviour for all four internet use categories (never, monthly, weekly, or daily) and shows there is a clear difference between the groups with daily users of the internet reporting (a) a greater desire to move; (b) a higher move expectation; and (c) more address changes than all the other groups with non-users having the lowest rates of all.This indicates that greater internet usage equates to more migration everything else being equal.However, everything else is seldom equal, and in this particular case internet users, and especially the more frequent users, might differ from nonusers (and less frequent users) in personal characteristics such as age and education that are known to be correlated with migration.In short, it might not be internet use itself that leads to more migration but the type of people that use the internet.To try to isolate the independent effect of the internet, therefore, the analysis now proceeds to estimate the impact of internet usage after controlling for selected personal and housing characteristics that are associated with migration.These model estimates are presented in Table 3  In Model 1 for moving address (Column 1 of the results in Table 3) the estimated coefficient for daily use of the internet is 0.047, indicating that daily users have a 4.7% point (pp) higher probability of moving address than non-users (the reference category).This is equivalent to the raw difference in moving between non-internet users and daily users in Table 2 (0.03 vs. 0.08).Model 2 adds a wide range of other explanatory variables.In this model, monthly and weekly use cease to be statistically significant-the other control variables mop up all the difference between these groups and non-users.However, daily internet usage remains statistically significant although the coefficient declines to 0.013; 73% of the raw differential is accounted for by the control variables but a significant unexplained fraction remains.For 'expect to move'-Columns 4 and 5-there are similar patterns with a raw 7.3pp differential falling to 1.5pp with all the control variables added -these 'explaining' 79% of the raw differential.The results differ somewhat for 'prefer to move' in Columns 7 and 8. Here, the internet use variables remain statistically significant even after other factors associated with internet usage and migration have been taken into account.
Model 3 is like Model 2 but with person-fixed effects-gender and ethnicity are taken as temporally invariant and therefore omitted, while region is omitted because changes in lagged region for the same person would be equivalent to a lagged change in address.
Other controls, in contrast, are allowed to vary between survey waves and this includes internet use where people can transition from not using to using the internet as well as between the various frequency categories.In Model 3, the internet variables all cease to be statistically significant, the only exception being for 'prefer to move' where changing to weekly internet use has a negative effect.
Taken together, these results tend to suggest that becoming a more frequent internet user-for example, moving into the daily user category-does not increase the propensity to prefer to move, expect to move, or, in reality, change address.
Widening now the overview of the results to consider the sign and significance of the control variables, it is readily apparent (and reassuring) that the models make sense in terms of expectations generated by the internal migration literature.Older respondents are less likely to prefer to move, expect to move, and to move in actuality.Those in larger households are also less likely to expect to

| Employed respondents
Table 3 showed that unemployed and economically inactive respondents were not more likely to move than the employed/selfemployed reference group, although unemployed people were more likely to expect and to prefer to move.Despite these undramatic findings, this section focusses more closely on internet usage, place of work, and their interaction, for only those respondents who had been employed at any time given recent interest in the mobility of these groups after the COVID-19 pandemic.Table 4 presents the results for changing address-the analogous results for expecting or preferring to move are presented in Appendix Tables A2 and A3.
Because the number of observations is reduced due to the focus on employees only, a fixed effects version of these models is not estimated.
Model 1 (internet variable only) and Model 2 (workplace variable only) demonstrate that employed daily internet users are more likely to change address than non-users and that those who either work at home or whose workplace is mobile are less likely to move than the reference category of those with a fixed workplace.Model 3 includes both the internet and workplace variables and their interaction.The positive effect of using the internet daily is almost cancelled out by the negative effect of working at home although the interaction term between working at home and daily internet use is statistically (positively) significant with a p-value less than 0.05.Model 4 adds all the other control variables to Model 3. When this is done, working at home ceases to be statistically significant, the positive effect of daily internet use is much reduced, and none of the interactions are significant.In Table A2 (expect to move), in the final Model 4, the internet and workplace effects, and their interactions, are all statistically insignificant.In Table A3 (prefer to move), the same holds in the final model apart from the interaction between working in 'One+ Other Places/Other' and both weekly and daily internet use, where there are statistically significant positive effects.

| DISCUSSION AND CONCLUSION
The analysis is based on a fully longitudinal, widely used data set and because of this it adds to the internet/migration literature.In assessing the results, there is no clear and consistent evidence that internet usage either increases or decreases moving preferences, expectations, or actual moves.It is true that frequent users of the internet are more likely to move (or to prefer or expect to move) than non-users but these differentials lessen, and generally become statistically insignificant, when other variables are taken into account.
In the person fixed effects models the internet variables cease to be statistically significant.In other words, it is the type of people who use the internet frequently (rather than the internet usage itself) which 'explains' greater relative mobility of internet users seen in However, things are seldom straightforward, and there are limits to the analysis which cannot be dealt with given the data.It is possible that the absolute mobility of all groups has fallen through time partly as a result of the internet, and partly from the operation of other factors, but this is hard to detect in this analysis, and indeed others, which compare and contrast internet users with non-users and which measure a relative effect.To address this issue effectively, and to consider more fully the possible importance of Zelinsky's Stage V, data with a greater temporal depth would be required and a compositional analysis that more readily accounts for absolute and relative changes.Moreover, this analysis, with its all-address change outcome variable, raises questions about the causal mechanisms that are in play.Given that most all-address changes are moves over short distances, it is likely that many are motivated by neighbourhood, housing, and environmental factors (Niedomysl, 2011); of the three broad types of explanation identified earlier in the literature, this points to place attachment as being putatively the most important factor since short-distance address changes are seldom for educational or employment reasons, these being usually more significant for longer-distance migrations.From this, we argue that there is no evidence that frequent internet usage increases place attachment (and therefore holds down short-distance moving rates) and furthermore also no reason to believe that internet usage hinders longer-distance migrations.
The current analysis also considers group differences by concentrating on those respondents in employment, considering how internet usage interacts with workplace.This was done because of (a) the possibility that internet usage could differentially influence different social and labour market groups and (b) because of recent trends in home working and their uncertain implications for internal migration.
Of course, there are limitations in the data and the analysis.Just because someone uses the internet daily and works at home, for instance, does not necessarily mean that that respondent uses the internet for work purposes.There must thus be some degree of subjective interpretation and supposition in interpreting the results.
However, it seems that there is a clear (but not conclusive) trend for home workers to be less migratory than those with other workplaces but that this is partly cancelled out by a greater propensity to move for those who use the internet daily.This result, for home working, accords well with the literature that homeworkers tend to be less migratory (Bellmann & Hübler, 2020).However, when this theme was considered in more detail with the inclusion of workplace-internet interactions in the analysis, there is no evidence to indicate that those who use the internet and who work at home are any more or less migratory than anybody else.This was before the COVID-19 pandemic, and the expansion of home working during the pandemic, and this shock might have jolted behaviour, but if the patterns observed in these data are resumed in the long term, then it is likely that migration levels and trends will return to their pre-COVID-19 path.Further waves of Understanding Society will throw light on these questions, but those results will need to be compared with the pre-COVID-19 period as a benchmark.
Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/psp.2755by Queen'S University Belfast, Wiley Online Library on [16/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License possibility that internet usage has differential effects for different groups, and especially because of the increased prevalence of home and remote working hastened by the COVID-19 pandemic.In this situation, there are questions as to whether home workers (and home workers who use the internet) are more likely to be spatially mobile than others.3 | DATA AND METHODS The UK Household Longitudinal Study (University of Essex, 2022) data set (Understanding Society) was chosen for this study because it contains the required person-level internet use and mobility information as well as a variety of demographic and socioeconomic contextual information.The analysis focuses on waves 3-10 (2011-2019) of the General Population Sample for Great Britain.Wave 3 was chosen as the starting point because this was the first wave to contain address change records and a detailed question about internet usage; wave 10 was chosen as the end point to avoid overlap with the COVID-19 pandemic, when society quickly transitioned into a vastly different environment for home and work.The explanatory variable of primary interest is based on a question which asks, 'How often do you use the internet for your personal use'?Respondents are able to select one of seven responses ranging from 'No access at home, at work or elsewhere' to 'Every Day'.To simplify the modelling and increase cell counts, this variable was collapsed into four categories: (1) No Internet Access or Never Use; (2) Monthly (comprising: less than once a month, once a month, several times a month); (3) Weekly (several times a week); and (4) Every day.
), this is represented by the form of the parameter α i , which is person-specific and time-invariant, and which implies that estimated coefficients are driven by within-person variation.The main advantage of this model is that, compared to a model based on between-person comparisons (i.e., were α i instead equal to a constant term, say α 0 ), a larger number of confounding factors-namely measured and unmeasured person-specific factorscan be accounted for, thus reducing omitted variables bias in so far as possible given the available data.Although this allows us to rule out a larger number of explanations for the results, this method comes at a cost of reducing the precision of estimates, especially when few within-person changes are observed in the data.To allay this concern, we show in Appendix TableA1 thatmany individuals within the analysis sample change the frequency of internet use between survey waves, ensuring sufficient numbers of changes with which to estimate coefficients.For each of the outcomes, three versions of Equation (1) are estimated, each built up in sequence and estimated as linear probability models.In Model 1, only the internet use dummies are included, while the constant term is assumed fixed across individuals; this model provides baseline (unadjusted) estimates of the effect of internet use.In Model 2, the full suite of control variables is added, providing adjusted estimates.Finally, in Model 3, person fixed effects are included, yielding the most comprehensive specification of the model.Here, gender and ethnic group are omitted since they are timeinvariant, and region of residence is omitted because changes in lagged region of residence reflect past changes in migration status (note, however, that including region in Model 3 does not alter the results).The estimation sample is restricted to individuals aged 18-74 years old with complete information on the internet use and control variables, resulting in an unbalanced panel of 133,861 observations (person-waves) composed of 28,781 individuals.Observations with missing values on outcome variables are retained where possible, toensure that all available information is used when modelling each outcome.We have verified that alternative sample restrictions-for example, including respondents from the Northern Ireland and ethnic minority boost samples or using a common sample of individuals with no missing values on any outcome-do not materially alter the results.
Descriptive statistics-Outcome variables and internet use.Effect of internet use on internal migration outcomes.Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/psp.2755by Queen'S University Belfast, Wiley Online Library on [16/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 2 Crosstabulation of outcomes by internet use.
The table shows estimated linear regression coefficients with standard errors clustered by person shown in parentheses.All outcome variables are binary.The categorical internet variable represents personal internet usage at the previous survey wave.All variables except wave, gender, age group and ethnic group are lagged by one survey wave.Model (1) is unadjusted; Model (2) is adjusted by all measured control variables; Model (3) is Model (2) with person fixed effects.In Model 3, gender and ethnic group coefficients cannot be estimated because they do not vary through time for any individual, while region of residence is omitted; sample size in Model 3 is lower because individuals who appear only once do not contribute to fixed effects estimates and thus are excluded.or to move, but they are no more likely to prefer to move.In contrast, those in rented accommodation and respondents with educational qualifications are more likely to move or expect to move (the results are more mixed for 'prefer to move').The wave dummy coefficients indicate that, in general, adjusted rates of actual moves, move T A B L E 4 Effect of internet use and workplace arrangements on the probability of moving address, with internet-workplace interactions, employees only.Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/psp.2755by Queen'S University Belfast, Wiley Online Library on [16/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

Table 2 .
When considering models 1 and 2 in the analysis of moving spatial scales that have substituted for internal migration.However, a descriptive reading of the results would suggest that there are no signs of internet usage reducing either the preferences for, expectations of, or actual migration of internet users relative to non-users.This would indicate that electronic communication and the internet have little influence on internal migration in the UK, and that they have not altered group relative behaviour.
other variables are controlled for; perhaps 10 or 20 years ago, the innovatory internet users of those days would be more sharply defined relative to the others.This latter point is relevant to wider debates; the UK could now be classified as being in Zelinsky's Stage V of the Mobility Transition with the dominance of technology, electronic communications, and other types of mobility at various temporal and The table shows estimated linear regression coefficients with standard errors clustered by person shown in parentheses.The estimation sample is restricted to employees, since workplace information was only collected from employed respondents.The workplace variable refers to the location where employees mainly work.Model (1) includes the lagged internet use variable; Model (2) includes the lagged workplace variable; Model (3) includes both plus their interaction; and Model (4) is Model (3) with control variables (coefficients not reported).