Who moves? A logit model analysis of inter-provincial migration in Canada

Abstract

This paper addresses the topic of inter-provincial migration in terms of the basic question: ‘Who moves?’. Panel logit models of the probability that an individual changes his or her province of residence from one year to the next over the 1982–1995 period are estimated using tax-based longitudinal data. It is found that moving is (i) inversely related to the home province's population size, presumably reflecting local economic conditions and labour market scale effects, while language also plays an important role; (ii) more common among residents of smaller cities, towns, and especially rural areas than those in larger cities; (iii) negatively related to age, marriage, and the presence of children for both men and women; (iv) positively related to the provincial unemployment rate, the individuals’ receipt of unemployment insurance (except Entry Men), having no market income (except for Entry Men and Entry Women), and the receipt of social assistance (especially for men); (v) (slightly) positively related to earnings levels (beyond the zero earnings point) for prime aged men, but not for others; and (vi) more or less stable over time, with men's rates declining slightly and women's holding steadier or rising slightly, indicating a divergence in trends along gender lines.

Acknowledgements

This research was made possible by the Applied Research Branch of Human Resources Development Canada. The Small Area and Administrative Data Division of Statistics Canada provided access to the LAD data (including the related human resources) upon which the study is based, related background file development and support, and comments on the analysis. The author also gratefully acknowledges a Social Sciences and Humanities Research Council Research Grant which was critical in earlier stages of the research with the LAD data. Useful comments were received from John Burbidge, Tom Courchene, Pierre Fortin, Gaétan Garneau, Stéphane Gascon, Michael Hatfield, Lars Osberg, Roger Sceviour, François Vaillancourt, and Allen Zeesman, with Zeesman providing the initial ideas for the general direction of the project. Gaétan Garneau and Roger Sceviour provided excellent research assistance.

Notes

Anderson (1966), Courchene (1974), Grant and Vanderkamp (1976), Lin (1995), Osberg, Gordon and Lin (1994), Robinson and Tomes (1982), Statistics Canada (1993), Stone (1969) and Vachon and Vaillancourt (1998) collectively provide evidence on gross outflows, gross inflows, net flows, and the specific province-to-province patterns of inter-provincial mobility, as well as the basic characteristics of movers and non-movers and the associated income patterns. The different ‘types’ of migration (primary, return, onward), are focused on in Hiscott (1987), Grant and Vanderkamp (1984, 1986), Newbold and Liaw (1990), Rosenbaum (1988, 1993), and Vanderkamp (1971, 1972); Hiscott (1987) and Hou and Beaujot (1995) concentrate on migration between Atlantic Canada and Ontario; and Day and Grafton (1998) look at students. Econometric studies of the determinants of inter-provincial mobility, including an important sub-literature on the role of fiscal variables (including relative wage structures), can be found in Courchene (1970), Day (1992), Day and Winer (1994), Dean (1992), Grant and Vanderkamp (1976), Lin (1995), Mills et al. (1983), Osberg et al. (1994), Robinson and Tomes (1982), Shaw (1986), Vachon and Vaillancourt (1998), and Winer and Gauthier (1982). The effects of inter-provincial mobility on provincial wage structures and related policy issues are focused on in Courchene (1974), Graham (1964), Rosenbluth (1996), Shaw (1986), and Vanderkamp (1988). Finally, the effects of migration on individuals’ incomes are addressed in Courchene (1974), Grant and Vanderkamp (1976, 1980), and Marr and Millerd (1980), as well as in more of a passing manner in Osberg et al. (1994) and Robinson and Tomes (1982). Important US work includes Gallway (1969), Greeenwood (1975), Laber and Chase (1971), Mincer (1978), Nakosteen and Zimmer (1980), Polachek and Horvath (1977), Schwartz (1973), and Sjaastad (1962).

This paper is one of a series on the topic based on the LAD data. Finnie (1998a,1999,2000) provide descriptive (non-econometric) analysis of migration patterns and longitudinal migration profiles and the associated income dynamics, while Finnie (1998b) exploits the longitudinal aspect of the LAD to estimate fixed effects models which control for pre-move earning levels and other individual attributes in order to isolate the effects of inter-provincial mobility on individuals’ earnings.

Most previous research on the characteristics of movers and the determinants of moving, including econometric models similar to those estimated here (e.g. Vachon and Vaillancourt, 1998) have used census data, which have numerous important limitations. First only the individual's current province of residence and that as of the previous census is known, resulting in incomplete, biased samples of all moves which occur over time (return moves and multiple moves which occur over the inter-census period are missed). Second, there is very little information on the earlier (‘pre-move’) situation, meaning that any analysis of the effects of individual characteristics or environmental factors is significantly limited to such basic items as the individual's age. Finally, the limit of the data to a given five-year inter-census period precludes any direct testing for shifts in the structure of inter-provincial migration over time. Alternatively, previous studies based on longitudinal data – notably the earlier Labour Market Activity Survey (Osberg et al., 1984; Lin, 1995) – have been limited by the small number of years over which individuals were followed and the relatively small sample sizes, which has limited the generalisability of the results in terms of the period of time covered and (again) precluded tests for shifts over time, while requiring the analysis to be carried out at an aggregate level which cuts across groups with distinctly different mobility behaviour (e.g. the pooling of men and women of different ages). Finally, previous research based on tax-based longitudinal files (the work by Courchene and Grant and Vanderkamp) is now simply dated in terms of the period covered and the methods used.

Atkinson et al. (1992) and OECD (1996) discuss the typically better coverage and lower attrition of administrative databases over survey databases. See Finnie (1998a) for evidence on attrition from the LAD and the relationship of this attrition to migration behaviour over selected intervals.

See Vanderkamp and Grant (1988) for a discussion of the various ways of identifying inter-provincial migration in different databases and the advantages of tax-based data in this regard.

This might be especially important in the case of Quebec, where the substantial proportion of allophones (close to the number of anglophones in the most recent census) will be included in one of the official language groups, depending on the language in which individuals decide to file their tax forms. The English–Quebec variable therefore includes anglophones and some allophones, while the Quebec variable on its own represents not just francophones, but also those allophones who file their tax forms in French – thus blurring the ‘French–English’ interpretation of the results to some degree. In other provinces, individuals who speak a third language are probably most likely to file their tax forms in English, leaving the French–ROC variable to reflect a purer minority (official) language effect.

Social assistance has been a separate item on individuals’ general T-1 tax forms (where it enters tax credit calculations) and the corresponding ‘T-5 SA’ forms have been sent out to individuals by the provinces only since 1992. From 1986 to 1991, social assistance was supposed to be included on T-1 forms (again affecting tax credit calculations), but in an income category which included other non-taxable government transfers (workers’ compensation, GIS) and T-5 forms were not sent out. Before 1986, social assistance was not included in any manner on individuals’ tax forms.

The models have also been estimated using a broader ‘market income’ measure. The results are generally very similar to those reported below.

These are calculated as the change in the probability of moving associated with each of the variables divided by the base line probability – that is, the percentage effect of ‘switching on’ each of the indicator variables individually in turn, or a change in earnings of $1000 for that particular variable. (The baseline probabilities are calculated using the omitted categories for the categorical variables and the means for each sex–age group for age, earnings, and the unemployment rate.)

Finnie (1998a, 1999) shows that the simple rates of out-migration from Alberta were the highest of all provinces in the earlier years covered by the data and then declined steadily thereafter to finish in the middle rank.

The total of the Quebec and English–Quebec coefficient estimates in the eight equations are: 0.383, 0.405, 0.413, 0.355 (men); and 0.278, 0.318, 0.315, and 0.146 (women).

See Finnie (1998a, 1999, 2000) for descriptive evidence regarding the specific directions of the flows and the classification of individuals according to their longitudinal profiles into one-time movers, multiple movers, and those who eventually returned to their province of origin.

The benchmark ‘single’ category includes never married, widowed, separated and divorced individuals, corresponding to the information directly available in the LAD data in this respect.

The specific province-to-province mobility patterns reported in Finnie (1998a, 1999) provide further evidence in support of this.

These assessments reflect the joint effects of the linear and quadratic earnings terms.

It would be interesting to know if these ‘social assistance leavers’ are moving into jobs or, as is sometimes alleged, simply collecting better social assistance payments in other provinces. The LAD data could certainly support such an analysis, but this is left to another project (as discussed in the conclusions).

The generally opposite signs of the male and female coefficients might indicate statistically significant divergences, even if the direction of the changes for each group, as represented in the standard errors and t-tests, are not.

Finnie (2003).

Finnie (2003).