If the elderly population were geographically immobile, then the largest concentration of elderly people would be in places with relatively low fertility and longer life expectancy at birth (United Nations Population Division, 2002). In fact, the elderly population is not immobile, and considerable numbers move. Although geographic mobility has long been of great interest to demographers and other social scientists because of its significance for population dynamics (Lee, Reference Lee1966) and its implications for issues such as economic change and development (Devillanova, Reference Devillanova2004; Parish, Reference Parish1973), family relationships (Boyle, Kulu, Cooke, Gayle, & Mulder, Reference Boyle, Kulu, Cooke, Gayle and Mulder2008), and community cohesion (Oh, Reference Oh2003), there are relatively few recent studies of the mobility status of older Canadians.Footnote 1
In the past, older Canadians were a relatively small proportion of the population, but this is changing. In 2011, 14.8 per cent of the population were aged 65 and older, compared to 7.9 per cent in 1971 and 11.1 per cent in 1991 (Statistics Canada, 2013a). Population projections prepared by Statistics Canada (2013b) forecast that 22.8 per cent of the population will be 65 years or older in 2031 and will reach 25.5 per cent in 2061. The relative and absolute increase in Canada’s elderly population over the next five decades means that a higher proportion of Canadian families are aging families. The residential mobility of these aging families will have important effects on neighbourhood composition, characteristics, and housing and other social needs. The purpose of this article is to examine trends in the residential mobility of seniors and to offer explanations for possible changes in elderly mobility trends over time.
Migration of the elderly population is an important topic for several reasons (see, for example, Edmonston, Reference Edmonston, Edmonston and Fong2011). Older Canadians move in large numbers, and this movement, coupled with the increasing proportion of the population that is older, has the potential to create large concentrations of elderly people in particular areas of the country. Such concentrations increase demands for health care and other social services and shift demands for housing type and associated services. For this reason, sound policy planning requires information about trends in population aging and the migration patterns of the older population.
A large body of research literature on migration exists, with excellent surveys of this literature (see, for example, Etzo, Reference Etzo2008; Greenwood, Mueser, Plane, & Schlottman, Reference Greenwood, Mueser, Plane and Schlottman1991). There are also many studies of elderly migration, although studies of elderly Canadian migration are fewer. We limit this brief review to studies of elderly Canadian mobility status. University of Alberta sociologist Herbert Northcott (1984, 1985, and 1988) was the first to publish studies of elderly mobility in Canada. He used census data on geographic mobility, which was first collected in the 1961 census, to examine the geographic mobility of elderly Canadians. Since Northcott’s pioneering work, relatively few studies of elderly mobility have been made. Available studies include Moore, Rosenberg, and McGuinness’s (Reference Moore, Rosenberg and McGuinness1997) 1991 Census Monograph which provides an excellent discussion of elderly mobility and related issues. Northcott and Petruik (Reference Northcott and Petruik2011) have offered a useful review of research literature and a list of needed research. More recently, since the first draft of this article, Northcott and Petruik (Reference Northcott and Petruik2013) published an article, using 1961 to 2006 census data, that examined the residential mobility for elderly Canadians.
This article makes three contributions to current empirical studies of the geographic mobility of older Canadians. First, it complements Northcott and Petruik’s (Reference Northcott and Petruik2013) recent study with updated analysis of mobility status, using census data on trends from 1971 to 2006. Second, it tests for temporal trends in mobility status with a multivariate model and offers possible alternative explanations. Third, it provides a multivariate analysis of mobility status for older Canadians that includes ethnic origin and temporal changes for the first time.
Although there is no unifying theoretical basis for integrating available studies of elderly migration, elderly migration is distinctive from general migration in several ways (Marr & Millerd, Reference Marr and Millerd2004). First, employment opportunities are less important for elderly migration because many elderly people are retired and do not seek employment. Second, older persons place greater weight on climate, health care, and other amenities in their migration decisions. Third, elderly migration is complex because of different characteristics and preferences of older movers (Bergob, Reference Bergob1995). Elderly migrants are not homogeneous, and different groups move for different reasons. Reasons for moving may also vary depending on the type of moves (for example, local, intraprovincial, or interprovincial).
Although there is no overarching theory for the residential mobility of elderly adults, the available research literature offers some guidance for factors affecting elderly mobility. Several perspectives on general migration are useful for studying elderly migration. One key aspect is the distinction between movers and stayers (Wiseman & Roseman, Reference Wiseman and Roseman1979). Elderly persons who choose not to move display several interesting characteristics: they are less likely to have had frequent residential moves in the past, more likely to be married, more likely to be homeowners, and have higher family income. On the other hand, elderly movers often cite one or more of several reasons for moving: (1) dissatisfaction with housing, (2) changes in health and the need to move to accommodate those changes, (3) desire for improvements in local amenities, (4) need to reduce residential living costs, and (5) preference to be closer to relatives and friends. Because of life course changes associated with retirement, many elderly adults consider migration when they are no longer attached by employment to a particular area. When retirement occurs, then, there is often a decision made about whether to stay or move.
We chose individual attributes available in public-use census data that are identified in the literature as determinants of geographic mobility. We consider elderly mobility to include two simultaneous decisions: the evaluation of the expected costs and benefits of moving, and the decision about the specific destination. The empirical application below models both of these decisions.
The relationship between individual characteristics and migration has been summarized in several typologies of retiree migration (Longino & Serow, Reference Longino and Serow1991; Wiseman & Roseman, Reference Wiseman and Roseman1979). These typologies note the heterogeneous preferences for elderly migration, which is a function of personal and demographic characteristics. For example, amenity-seeking retirement migration tends to include a higher proportion of younger retirees. The heterogeneity issues motivate our inclusion of demographic characteristics – such as age, sex, and marital status – in the multivariate analysis of elderly mobility.
If an elderly person or family decides to move, there is a complicated process of deciding about a destination. Previous research has analysed destination choices in two ways. One way is to consider all possible destinations: for example, interprovincial migration in Canada from any specific place to any of the 10 provinces (Liaw & Ledent, Reference Liaw and Ledent1988).Footnote 2 A second approach is to define various types of mobility, such as local movers and long-distance migrants. This has the advantage of linking the destination decision with the type of move. For example, Knapp, White, and Clark (Reference Knapp, White and Clark2001) examined household mobility in the United States with a study of three types of movers: intrametropolitan, intermetropolitan to central city, and intermetropolitan to a suburb. Their research focused attention on whether the household decides to move to another metropolitan area and, if so, whether they move to the central city or suburb. We follow this second approach in this article. Older Canadians first decide whether to stay (non-movers) or move. If they move, they have four possible choices to become movers within their local community (called local movers), intraprovincial migrants, interprovincial migrants, or external migrants who move from outside Canada.
Data and Methods
Data Sources
The largest set of data on Canada’s elderly population is the census. Except for 1976, data from the 1971 to 2006 censuses are currently available in public-use census microdata samples, which contain anonymous information on individuals. These data are particularly useful for migration analysis because every census since 1961 has asked respondents where they lived five years ago.Footnote 3 Replies to this question provide information on their mobility status (non-mover and type of movement) as well as the origin and destination of their move. We used seven census microdata samples for analysis in this article, as shown here.
Overall, the analysis included 502,275 elderly Canadians who answered migration questions in the 1971 to 2006 censuses.
On the basis of responses to census migration questions, mobility status can be defined as five mutually exclusive categories: (1) non-movers: persons who did not change their place of residence, (2) local movers: movers who changed their place of residence but stayed in the same municipality or geographic area, (3) intraprovincial migrants: migrants who moved from one municipality or geographic area to another but remained within the same province or territory, (4) interprovincial migrants: migrants who moved from one province or territory to another, and (5) external migrants: migrants who moved from outside Canada to a place in Canada. The sum of the last four categories is the total number of movers, and the sum of the last three categories is the total number of migrants.
A key advantage of census data is that data can be tabulated by socioeconomic characteristics of the population. In the next sections, we examine the relationship of mobility status with age, sex, marital status, ethnic origin, nativity, education, family income, homeownership, and province.Footnote 4 The use of census microdata for the study of residential mobility also has several limitations, however. First, public-use census microdata samples include only persons living in private households. Because the sample does not include persons living in institutions, there is no information on moves of elderly Canadians into nursing homes or moves from one institution to another institution. According to 2011 census data, 7.9 per cent or about 390,000 elderly Canadians, aged 65 or older, reside in collective dwellings such as nursing homes or other institutional facilities (Milan, Bohnert, LeVasseur, & Pagé, Reference Milan, Bohnert, LeVasseur and Pagé2012). A second limitation is that census data do not record Canadians who leave Canada because census questionnaires are not sent to Canadians who live outside Canada. Finally, census data do not ask individuals about their motivations or reasons for staying or moving. Different data are needed for the study of subjective factors.
Measurement
Mobility Status. The response variable for multivariate analysis was a categorical variable measuring mobility during the five years prior to the census, using the five categories as already described and coded from 1 to 5.
Determinants. We examined determinants of elderly mobility status, including individual characteristics, temporal trends, and contextual variables.
Individual characteristics were defined at the end of the migration period, as reported in the census. The eight individual variables included were as follows: (1) age, measured in five-year age groups, 65–69, 70–74, 75–79, 80–84, and 85 years and older, with the 65–69 group used as the reference group; (2) sex, measured as female or male, with females designated as the reference group; (3) marital status, measured as single or never-married, married or common-law, separated, divorced, or widowed, with “single or never-married” designated as the reference group; (4) nativity, coded as Canadian-born (a Canadian citizen at birth, regardless of whether they were born in Canada or outside Canada) or foreign-born (not a Canadian citizen at birth), with “Canadian-born” serving as the reference group; (5) ethnic-origin group, coded in 14 groups that are comparable for each census – these included Canadian, Aboriginal, French, British, Other West European (including Scandinavians and other North Europeans), East European, South European, Black/African/Caribbean and Latin American, Arab/West Asian, South Asian, Chinese, Other East and Southeast Asian, Other Single Origins, and Multiple OriginsFootnote 5 – with “Canadian” serving as the reference group; (6) education, measured in terms of the highest degree completed with five categories for less than high school degree, high school degree, non-university degree, university (bachelor’s) degree, and post-bachelor’s degree (master’s, doctorate, or professional degree), with “less than high school degree” designated as the reference group; (7) family income, reported family income (in $10,000s, 2006 dollars after adjustment for inflation) with the family income category of less than $10,000 serving as the reference category; and (8) homeownership, measured as a binary variable for “do not own” or “own” their house. The reference category is “do not own”.
Temporal trends were measured by coding each census year – 1971, 1981, 1986, 1991, 1996, 2001, and 2006 – as a categorical variable and including year in the multivariate analysis, with 1971 serving as the reference category.
In order to take contextual effects into account in the analysis, we included fixed effects for the province of origin. This was done by including the province where the respondent lived five years ago in the multivariate analysis, coded as the 10 provinces and a separate code for Northern Canada, which consists of Yukon, Northwest Territories, and Nunavut.
Data Analysis – Multinomial Logit Model
Researchers studying migration can consider using several possible statistical models (McFadden, Reference McFadden, Manski and McFadden1979). Relatively few researchers examine simple mover-stayer binary data because there is more interest in the type of mobility or where the person moves to. For this reason, binary logit models are seldom employed in migration analysis. A key consideration for statistical models is whether the data include destination choices (Moss, Reference Moss1979). If alternative destinations are included in the model, then most researchers use either mixed conditional logit or nested logit models. If the analysis is limited to individual characteristics and does not include alternative destinations, then the most common approach is a multinomial logit model. We used the multinomial logit model for analysis because the data included temporal, individual, and contextual variables but not alternative destination choices. The analysis focused on the factors related to the choice of non-moving, moving within the local community, or moving to a different community.
The multinomial logit model examined the likelihood that a person (or family or household) stays or makes specific moves based on characteristics of the person, census year, and province/territory of origin. Although personal characteristics such as age, race, or education vary, the model assumes that individual characteristics are constant across locations. For example, a person’s age is constant across locations; however, age may have a different effect on the likelihood of selecting different mobility statuses. In other words, the multinomial logit models have regression coefficients that vary for alternative mobility statuses.
With a response variable Y that can have any of n categorical values, numbered 1, 2, …, n, let πij denote the probability that the ith observation falls within the jth category of the response variable. In other words, let πij ≡ Prob(Y i = j) for j = 1, …, n. Assuming that we have k regressors, X 1, …, X k, for the prediction of πij, using the multivariate logistic distribution for the modelling of the prediction, we have 
${\pi _{ij}} = 1 - \sum _{l = 2}^n{\pi _{ij}}$ for the baseline category; here, assumed to be the first category. And we have
$${\pi _{ij}} = {{{e^{{\gamma _{0j}} + {\gamma _{1j}}{X_{i1}} + ... + {\gamma _{kj}}{X_{ik}}}}} \over {1 + \sum _{l = 2}^n{e^{{\gamma _{01}} + {\gamma _{1l}}{X_{i1}} + ... + {\gamma _{kj}}{X_{ik}}}}}}$$for the j = 2, …, n categories of the response variable. This model is estimated for our analysis in this article, in which there are five categories in the mobility status response variable and 10 explanatory variables measuring temporal changes, individual characteristics, and contextual fixed effects. With some algebraic rearrangement, we have
$$lo{g_e}{{{\pi _{ij}}} \over {{\pi _{in}}}} = {\gamma _{0j}} + {\gamma _{1j}}{X_{i1}} + ... + {\gamma _{kj}}{X_{ik}}$$for j = 2, …, n. Regression coefficients in the multinomial logit model reflect the log odds of membership in category j relative to the baseline category.
An assumption of the multinomial logit and other similar logit models is that alternative outcomes are independent of one another. If outcomes are perceived by potential migrants as close substitutes, then unobserved factors that affect the choice of one outcome may also affect another. This violates the “independence of irrelevant alternatives” (IIA) assumption for multinomial and conditional logit models. If the IIA assumption is not met, these models provide inconsistent parameter estimates. For this reason, our analysis tested for whether the IIA assumption was met.
The multinomial logit model is estimated by the full-information maximum likelihood method, implemented for this analysis by the mlogit procedure in StataCorp’s Stata Statistical Software, Release 12. Estimates from the multinomial logit model are interpreted in a way similar to those from other types of logit models (see Long & Freese, Reference Long and Freese2006).
Interpreting the estimated coefficients in a multinomial logit model poses some difficulties because interpreting the sign and size of the coefficient depends on its comparison to the base outcome. For example, a negative sign for educational attainment does not necessarily mean that the predicted probability declines with education. If the predicted probability for the base outcome has a negative relationship with education, then a negative sign for the estimated coefficient means that there is a positive relationship with education. For this reason, interpreting the coefficients in a multinomial coefficient is best examined with use of predicted probabilities (Kohler & Kreuter, Reference Kohler and Kreuter2009).
Results
Elderly Mobility Patterns
Elderly spatial movement may be temporary (such as the seasonal movement of “snowbirds” from Winnipeg to Arizona) or permanent. Permanent changes of address are often short-distance local moves, such as when an elderly person moves from a single-family house to an apartment in the same city. Movers who cross geographic boundaries are called “migrants” whereas movers within the same geographic boundary are referred to as “non-migrant movers”. It is important to note that the demographic definition of migration is not based on the distance of the move but on whether the movement crosses a geographic boundary. Small movements may involve a relatively short distance but are defined as migratory if they cross a geographic boundary.Footnote 6
The proportion of the elderly population that moved or changed their place of residence in the five years prior to the census decreased from 30.6 per cent in 1971 to 20.0 per cent in 2006 (see the line for “All Movers” in Figure 1 or the first column of Table 1), a decline of more than 10 percentage points. One possible explanation for this striking decline in elderly movement is that the elderly population became older during this 35-year period and, hence, became less mobile. For this reason, it is useful to adjust mobility rates for the age distribution by standardizing the rates on the 2006 elderly age distribution.Footnote 7
Trends in mobility status for elderly Canadians: All movers, local movers, intraprovincial migrants, 1971 to 2006
Observed, age-standardized, and multinomial logit-predicted mobility rates by year: 1971 to 2006a
a Mobility rates are age-standardized on the 2006 age distribution.
When 1971 to 2006 mobility rates are age-standardized, as shown in Table 1, there are only modest changes in the rates. For 1971, all movers are 30.6 per cent for both the observed rate and the age-standardized rate. The largest influence of the age distribution is for 1981, when there is a modest 0.4 percentage point difference between the observed and age-standardized rates. For example, the proportion that are movers decreased from 25.8 per cent in 1981 to 20.0 per cent in 2006, a decrease of 5.8 percentage points; however, only 0.4 percentage points is due to changes in the age distribution while most of the change is due to actual decreases in age-specific mobility rates.
Why do changes in the age distribution have such minor effects on elderly mobility rates from 1971 to 2006? The effects of changes in the age distribution of the elderly population would be important if, in fact, there were strong shifts in the proportion of the elderly at different ages. Although there have been changes in the proportion of elderly adults in the total population, there have been relatively minor changes in the age distribution within the elderly population.
Because there is no evidence that changes in age composition influenced trends in mobility rates for all movers, the overall rate may be influenced by strong declines in rates for specific mobility types. But, returning to Table 1, we see there are decreases in mobility rates for all mobility types. Figure 1 shows trends for local movers and intraprovincial migrants, and Figure 2 presents trends for interprovincial and external migrants.Footnote 8 The proportion of local movers decreased from 19.0 per cent in 1971 to 11.2 per cent in 2006, or 59 per cent of the 1971 level. The proportion of intraprovincial migrants decreased from 8.4 per cent in 1971 to 6.6 per cent in 2006, or 79 per cent of the 1971 level. Interprovincial migration rates decreased from 1.7 per cent in 1971 to 1.2 per cent in 2006, or 71 per cent of the 1971 level. The proportion of external migrants declined from 1.5 per cent in 1971 to 0.9 per cent in 2006, or 63 per cent of the 1971 level. Although all mobility types decreased, the declines were larger for local movers and external migrants. The proportion of intraprovincial migrants decreased the least. It is clear that mobility rates for the elderly population diminished markedly from 1971 to 2006, dropping steadily in each census for all types of mobility.
Trends in mobility status for elderly Canadians: Interprovincial migrants and external migrants, 1971 to 2006
Next, we discuss results from the analysis of elderly mobility status using a multinomial logit model. Summary statistical results from the model are discussed, as well as a test to see if the IIA conditions have been violated, followed by results for local movers, intraprovincial migrants, interprovincial migrants, and external migrants.
Multinomial Logit Model Results
The multinomial logit model is estimated for 467,128 elderly adults. Overall, the model accounts for more than 15.5 per cent of the overall variance in choice of mobility status outcomes. The explanatory variables in the model provide a partial explanation of older Canadians’ residential mobility. Based on the Wald χ2 test, which has a probability equal to 0.0000, the hypothesis that all the coefficients are simultaneously equal to 0 can be rejected.
An important assumption of multinomial and other categorical logit models is that the outcome categories have “independence of irrelevant alternatives”, which means that the inclusion or exclusion of an outcome category does not influence the coefficients associated with explanatory variables for the remaining categories. Under the IIA assumption, we would expect no systematic change in the coefficients if we excluded one of the four outcomes from the model. There is no evidence that the IIA assumption is violated. Excluding any one of the four tested outcomes does not affect the coefficients for the remaining three alternatives, and the model appears to obey the IIA assumptions.Footnote 9
The estimated multinomial logit model includes non-movers as the reference group and estimates for four mobility categories (all estimated relative to the non-movers group): local movers (Table 3), intraprovincial migrants (Table 4), interprovincial migrants (Table 5), and external migrants (Table 6). Each of these tables shows the explanatory variable and category on the left-hand side, the multinomial logit coefficient, standard error, Z-value, and probability of the Z-value in the first, second, third, and fourth columns, and the predicted probability for the category value (holding all other factors constant) in the fifth column.
For ease of interpretation, we calculated the predicted probabilities for explanatory variables and their categories for the all-movers group, which is calculated here as 100 minus the predicted percentage of non-movers (see Table 2). Table 2 shows the explanatory variable and category on the left-hand side, the observed percentage for all movers in the first column, the multinomial logit predicted probability (holding all other factors constant), standard error, Z-value, and probability of the Z-value in the second, third, fourth, and fifth columns.
Observed and predicted probabilities for all movers
As noted in these tables, the estimated coefficients and predicted probabilities are statistically significant at the .05 level with only a few exceptions. Discussing each mobility type as well as the results for all movers would be overly complicated and lengthy. Instead, we interpreted results for each explanatory variable in the next sections.
Temporal Effects
About one-fifth or more of elderly Canadians changed their place of residence in the five-year period prior to the census, from 1971 to 2006. As shown in the first column of Table 2, the observed proportion of movers decreased from 1971 to 2006, from 30.4 per cent in 1971 to 19.8 per cent in 2006. Elderly Canadians were more likely to report moving in the 1971 to 1991 period than in more recent years. Overall, mobility rates for elderly Canadians in 2006 were about two-thirds of their 1971 levels.
Taking all other explanatory factors into account, the predicted probability for all movers changes slightly, as shown in the second column of Table 2. Although the observed percentage for all movers decreased by 10.6 percentage points during 1971 to 2006 (30.4 minus 19.8), the predicted probability declined by 9.0 percentage points. In other words, the composition of the elderly population changed slightly from 1971 to 2006, and accounts for about 1.6 percentage points of the overall 10.6 percentage decrease. Stated differently, there were considerable declines in the rate for all movers that are not explained by the explanatory factors included in the multinomial logit model.
Local moves are the most common mobility type for elderly Canadians. Although there is considerable mobility among elderly Canadians, most moves are local, as shown in Figure 1. More than one-half of all moves are local moves for both males and females. For example, 19.8 per cent of elderly people moved in the five years prior to the 2006 census, and 11.3 per cent (or 57% of all moves) were local moves. Most of the decrease in overall rates for all moves is accounted for by declines in the rate of local moves by elderly Canadians. In 1971, 17.4 per cent of elderly Canadians reported that they were local movers (see the predicted probability column for local movers by year in Table 3). This rate decreased dramatically by 2006, when only about 11.3 per cent of elderly Canadians reported that they had moved locally in the five years prior to the census.
The second most common type of mobility pattern is intraprovincial migration, when an elderly Canadian moves to a different municipality in the same province. Intraprovincial migration rates decreased from 8.8 per cent in 1971 to 6.8 per cent in 1996 and have been steady since 1996 (see the predicted probability column for intraprovincial migrants by year in Table 4). In recent censuses, intraprovincial migration rates are about 80 per cent of their 1971 levels.
Interprovincial and external migrants are less frequent types of mobility. Interprovincial migration rates declined steadily from 1971 to 2006 (see the predicted probability column for interprovincial migrants by year in Table 5). By 2006, interprovincial migration rates were about 70 per cent of their 1971 levels.
Census data note whether Canadian residents lived outside Canada five years prior to the census. External migration rates have varied over the 1971 to 2006 period, with rates of 1.3 per cent in 1981, 1991, and 1996 (see the predicted probability column for interprovincial migrants by year in Table 6). Rates have been more modest, at 1.0 per cent or less, in 2001 and 2006.
Sex and Age
Younger adults are more likely to change their place of residence than elderly Canadians. Adults aged 25 to 64 were 2.6 times more likely to move over a five-year period from 2001 to 2006 than elderly Canadians.Footnote 10 Among elderly Canadians, the data show a steady decrease in the mobility rate for all movers by age. Controlling for all other explanatory variables, the predicted probability for all movers decreases from 24.8 per cent for those aged 65 to 69 years to 16.2 per cent for those aged 85 and older. Elderly males are slightly more likely to move (21.6 %) than females (20.4 %) after controlling for all other factors.
Older Canadians display modest variations in local mobility rates by age, and when controlling for other factors, local moves within municipalities decline with advancing age. As with all movers combined, males are slightly more likely to be local movers than females.
Among elderly Canadians, the probability of longer distance intraprovincial moves is highest for those age 65 to 69 years. Intraprovincial moves are less common with increasing age. Males are somewhat more likely to be intraprovincial migrants than females.
Interprovincial migration rates are highest for elderly Canadians aged 65 to 69 and decrease steadily with age. Elderly females have slightly higher interprovincial rates, taking all other factors into account, than elderly males.
External migration rates decrease markedly with age: external migration rates are more than three times higher for those aged 65 to 69 compared to those aged 85 and older. External migration rates are slightly higher for elderly males, compared to elderly females.
The consistent relationship of age and lower rates of geographic mobility is an important finding because it has implications (discussed in the concluding section) for future increases in the proportion of aging families. Comparing the 85-and-older age group with the 65–69 age group, local movers are 32 per cent less, intraprovincial migrants are 31 per cent less, and interprovincial migrations are 31 per cent less. As the proportion of older aging families increases (that is, families with elderly aged 85 and older), overall elderly geographic mobility will be markedly reduced.
Marital Status
There is great variation in mobility patterns by marital status. Regarding total mobility, older persons who are currently married or are never married are the least likely to move. Elderly persons who are divorced, separated, or widowed have higher rates for all movers.
The general pattern for all movers by marital status holds for all mobility types: currently married and never-married elderly individuals are less likely to be local movers, or intraprovincial, interprovincial, or external migrants. On the other hand, divorced, widowed – and, especially, separated – elderly adults have higher mobility rates. There are large differences for local movers, however, where married and never married elderly adults are relatively less mobile than elders in other marital statuses. Among external migrants, there are noticeably high rates for widowed elders, suggesting that there is a tendency for the elderly immigrant to move to Canada to join family members after widowhood.
Education
Educational attainment is one of the most commonly used indicators of socioeconomic status, which we measured in this analysis using the highest degree obtained. There is a general tendency for overall mobility levels to increase with higher levels of education, with the highest overall rates of mobility for those with university or higher degrees, and lower overall rates for those with a high school degree or less. Those with a university or higher degree are especially more likely to move from one province to another or to have moved from outside Canada.
Family Income
Family income is the second indicator of socioeconomic status used in this analysis, coded in categories ranging from $10,000 or less in 2006 constant dollars, $10,000 to $19,999, and so on with $100,000 or more as the highest category. After taking all other factors into account, this analysis reveals that mobility rates decrease as family income increases, except for interprovincial migration. Comparing the poorest to the richest family income category, all mover rates are 6.2 percentage points higher for the poorest category, local mover rates are 1.0 percentage points higher, intraprovincial migration rates are 1.6 percentage points higher, interprovincial migration rates are only 0.1 percentage points higher, and external migration rates are 3.6 percentage points higher. Overall, there is little variation in interprovincial migration rates by family income, ranging from 1.2 to 1.5 per cent. Family income effects are most noticeable for external migrants, with especially higher rates for elderly Canadians with family income of less than $20,000. For all movers, local movers, and intraprovincial migration, rates decrease with increases in family income, but not as markedly as for external migrants.
Nativity
Limited attention has been paid to differences in mobility rates for Canadian-born and foreign-born elderly adults in spite of the increasing contribution of immigration to Canada’s elderly population. Foreign-born elderly adults have higher rates of mobility, compared to Canadian-born elders. The higher levels for foreign-born elderly adults, however, are especially influenced by their higher rates of external migration: more than 3.6 per cent of foreign-born elders moved from outside of Canada, while only 0.2 per cent of Canadian-born elders did.Footnote 11
If other mobility statuses are examined, foreign-born elderly adults have higher rates of local movement, as well as intraprovincial and interprovincial migration. Stated differently, foreign-born elders, once they reside in Canada, are more likely to move within and between communities than Canadian-born elders.
Ethnicity
Different ethnic groups may exhibit variations in their mobility patterns. Some groups may have an affinity to live close to their co-ethnic counterparts while others may have a weaker attachment and move more often or to different communities. Because there has been relatively little attention paid to ethnic variations in elderly mobility patterns, these results are suggestive and may stimulate further studies. Taking all other explanatory factors into account, this analysis reveals that mobility rates for all movers are relatively high for several ethnic groups that include more-recent immigrants to Canada: Black/African/Caribbean, Latin American, Arab/West Asian, South Asian, Chinese, and East Asian.Footnote 12 These groups have rates for all movers that are 25 per cent or higher than rates for other groups. Comparing the extremes, 36 per cent of South Asian elderly adults report moving during the previous five years while only 17 per cent of Aboriginals, East Europeans, or South Europeans report moving. The higher overall mobility rates for these recent immigrant groups are due to higher rates for local movers. South Asian elderly adults, for example, report local mover rates that are about 2.5 times higher than for British elderly adults. The recent immigrant ethnic groups also have higher overall mobility because they have higher external migration rates. External migration rates are 1.6 per cent for Black/African/Caribbean, Latin American, Arab/West Asian, South Asian, and East Asian elders, compared to rates of 0.1 per cent for British elders.
Homeownership
We expect that homeownership reduces elderly mobility because it is more difficult to sell and buy a home than to move from a rental unit. Moreover, elderly persons may have a greater attachment to their own home than to rented units. This analysis shows that elderly Canadians who do not own their home are much more likely to move than homeowners. Mobility rates for all movers are 37.7 per cent for elders who do not own a home, compared to 14.4 per cent for elderly homeowners. Mobility rates for homeowners are relatively low for local movers, and intraprovincial and interprovincial migrants. External migrants are an exception, however, and a slightly higher per cent of homeowners report that they are external migrants.
Province
Older persons residing in the Atlantic provinces are less likely to have moved compared to residents of other provinces. Mobility rates for all movers range from 16 to 17 per cent for elderly Canadians in the Atlantic provinces, which is lower than other provinces and much lower than the 26 per cent rate observed for British Columbia. The lower overall mobility rates for elderly people living in the Atlantic provinces result from lower rates for local movers and intraprovincial migration. Interprovincial migration rates for elderly adults in the Atlantic provinces are higher than in Ontario but lower than in the Prairie provinces.
Most elderly Canadians live in either Ontario or Quebec. There are distinctive differences in elderly mobility patterns for these two populous provinces. Quebec elderly people have modest local movement and intraprovincial migration rates, which are slightly higher than those in the Atlantic provinces but lower than in other provinces. Ontario elderly adults have higher rates of local movement and intraprovincial migration than Quebec, but have the lowest interprovincial migration among Canadian elders. As a result, compared to Quebec, Ontario has a slightly higher rate for all movers.
Although the Prairie provinces are often grouped together for discussion, it is apparent that differences exist in elderly mobility rates for Manitoba, Saskatchewan, and Alberta provinces. Manitoba and Saskatchewan are more similar because both have lower overall rates of elderly movement, although Manitoba has relatively higher rates of local movement, and Saskatchewan has relatively higher rates of interprovincial migration. Alberta has comparatively higher elderly mobility, especially due to higher rates of local movers. All three Prairie provinces have similarly high rates of interprovincial migration (2.5% or higher), being sources of elderly provincial out-migrants in recent decades.
British Columbia has the highest overall elderly mobility rates, with especially high rates for local movers and intraprovincial migration. No other province has an older population as mobile as that of British Columbia. British Columbia’s elderly population is heavily influenced by the arrival of large numbers of retirees from other provinces who move within and between British Columbia’s communities after arrival. It is not uncommon for a new elderly arrival in British Columbia to initially rent an apartment, move to a different apartment in the same city, and then purchase a home for longer-term residence – which generates higher local mover and intraprovincial mobility rates.
Canada’s Yukon, Northwest Territories, and Nunavut differ from other provinces in many ways. Northern Canada has a small but relatively mobile population that is more likely to move locally and more likely to migrate to another province. Elderly intraprovincial migration rates are very low in Northern Canada, suggesting that most elderly residents move within their communities or, alternatively, migrate to a different province.
Summary of Results
Our results reveal several common patterns. For the four types of mobility, mobility rates are generally higher for younger elderly adults in their late 60s and early 70s. Males are usually more mobile than females. Non-married elderly people move more than married persons, with higher rates for separated and widowed elders, for example. Mobility rates are higher for better-educated elders. And, finally, mobility rates are higher for foreign-born elders.
It is apparent that mobility rates for elderly Canadians have decreased substantially since 1971, and that a part of the decrease is due to composition changes in the population, as reflected by the reduction in the predicted probabilities for all movers after controlling for the explanatory factors in the multinomial logit analysis. Likewise, prior research is supported by similar results for the relationship of age, marital status, education, and homeownership with mobility status. We present new results, however, for temporal effects and ethnic origin, which have not been previously examined.
Discussion and Conclusions
Elderly Canadians are less mobile than younger adults. Nevertheless, older persons display moderate geographic mobility, with about one in five people moving from 2001 to 2006. Overall mobility rates have declined substantially from 1971, when almost one in three elderly persons moved during the 1966–1971 time frame. This decline is genuine and not due to shifts in the age distribution of the elderly population. Decreases in elderly mobility occurred across all mobility types: from 1971 to 2006, there were decreases in elderly mobility for local movers, as well as for intraprovincial, interprovincial, and external migrants.
Most elderly moves are local, with more than one-half (55%) of all moves occurring within the local community. Another large share of all moves are those older adults who move from one community to another within the same province; this type of migration accounts for about one-third of all elderly moves. Interprovincial and external migration account for relatively few – seven and five per cent, respectively – overall moves. Both interprovincial and external migrants have unusual effects, however. Interprovincial migration adds greatly to elderly population growth in destination communities, which often receive a large share of interprovincial elderly migrants. Recent interprovincial migration is highly selective to a few areas, including southern British Columbia and coastal New Brunswick, and cities such as Calgary, Victoria, Edmonton, and Halifax. External migration is limited to a few large metropolitan areas, especially Toronto and Vancouver.
Mobility status varies with several individual and destination characteristics. Older movers are more likely to be recently retired, to be separated or widowed rather than married, and to have higher levels of education. Mobility rates are relatively low for the Atlantic provinces, Quebec, Manitoba, and Saskatchewan, but comparatively high for Alberta and British Columbia. Interprovincial elderly migration is coastward in direction, with higher numbers moving westward to Alberta and British Columbia and smaller numbers to the Atlantic provinces.
It is clear that elderly mobility is influenced by many individual and community factors. This requires analysis that distinguishes temporal, individual, and contextual factors that affect elderly mobility. Elderly movers are heterogeneous with respect to their preferences for migration and for possible destinations. As adults age, they change their preferences for the locational features when selecting a possible new place of residence. The explanatory power of various community factors, therefore, is likely to change with age.
These empirical findings confirm that provincial residence of origin has different effects on local movers, intraprovincial, and interprovincial migration. The Atlantic provinces have relatively low mobility levels for elderly residents, including less movement for local movers as well as intraprovincial and interprovincial migrants. On the other hand, Alberta and British Columbia experience higher levels of mobility, reflecting to a large extent the higher numbers of elderly that migrate to these provinces from elsewhere.
Overall, the relationship of age and residential mobility is important for discussion in this special issue of the Canadian Journal on Aging. Residential mobility among older adults is less than one-half the rate for younger adults. In addition, geographic mobility rates for elderly adults appear to be decreasing over time. This suggests that residential mobility rates for Canadian households may continue to decrease into the future, with overall declines of about 10 per cent due to the increased proportion of aging families.Footnote 13
In addition to the influence of the relative proportion of elders, there are effects from aging within the elderly population. As mentioned earlier in this article, the population aged 85 and older have geographic mobility rates that are about one-third less than those aged 65 to 69. Future aging within the elderly population is likely to further reduce the overall geographic mobility of Canada’s population, other factors being equal.
Two Complementary Studies
We had previously noted that Northcott and Petruik (Reference Northcott and Petruik2013) updated previous work on trends in the residential mobility of Canadian elderly adults (their article was published after we began our analysis and writing of this article). It is fortunate to have two articles that complement each other and offer different perspectives on trends in the residential mobility of elderly Canadians. The two articles share some purposes. First, both have examined trends in elderly mobility over time. Northcott and Petruik (Reference Northcott and Petruik2013) used published census data in their study of trends from 1961 to 2006. Because census microdata are not available for 1961, our analysis examines trends from 1971 to 2006. Second, both articles have examined the same four categories of mobility: local movers, and intraprovincial, interprovincial, and external migrants.
Both articles have reported similar trends in residential mobility for elderly adults. Northcott and Petruik (2013: Table 1) reported that mobility rates were high in 1971 and 1976 and, based on published tables from Statistics Canada, declined for each of the four mobility components after the 1970s. In this article’s analysis of individual census microdata, results confirm the decline in elderly mobility rates, and further shows statistically significant temporal effects for all mobility components.
Northcott and Petruik’s (Reference Northcott and Petruik2013) analysis of published tables examined age and sex differences, and we also included age and sex in our analysis. Both studies document that mobility rates are higher for elders in the younger age groups, 65–69 years and 70–74 years. While Northcott and Petruik (2013: Table 5) reported somewhat higher mobility for elderly females, our multivariate analysis showed higher male mobility rates for all movers – 21.6 per cent for males and 20.4 per cent for females in Table 2’s predicted probabilities. Based on a closer examination of our analysis, we note that a higher proportion of females are widows among the elderly and that widows have higher mobility rates. As a result, multivariate analysis including sex and marital status reveals that differences in marital status by sex account to a great extent for the higher mobility rates for females. When marital status (together with other explanatory variables in our model) is included in multivariate analysis, females have somewhat lower mobility rates than males.
Northcott and Petruik (Reference Northcott and Petruik2013) noted that census microdata would be useful because they allow researchers to examine the effects of education, marital status, and other individual and household characteristics. This article includes these additional explanatory variables in addition to age and sex. Results confirm the decline in elderly mobility rates since the 1970s reported by Northcott and Petruik and provide additional insights into the role of gender, marital status, and other individual and household characteristics on elderly mobility.
Interpreting Temporal Trends
Findings show that overall elderly mobility decreased from a peak of over 30 per cent in the late 1960s and early 1970s to levels around 20 per cent in recent years. Such a trend may seem counterintuitive because many assume that elderly Canadians, like other Canadians, would have become more mobile over time, given the popular belief that Canadians are increasingly mobile. The empirical evidence, however, is that elderly adults are moving considerably less now than 35 years ago.
Is the decreasing trend in overall mobility genuine, and, if so, what explanation can be offered for declining overall mobility for elderly Canadians? Our article’s initial explanation for decreases in elderly mobility is that the elderly population may have been getting older. Because the oldest old, aged 85 years and older, have lower mobility rates, the possible aging of the elderly population may lead to declines in overall mobility. To address this possibility, we standardized mobility rates on the 2006 age distribution. As shown in Table 1, there is a modest influence of the age distribution on the overall mobility rates for the elderly population. There is a small difference in 1981 data, but otherwise the observed and age-standardized rates are similar. The age distribution has only a minor influence within the population 65 years of age and older from 1971 to 2006.Footnote 14 Therefore, decreases in mobility rates for elderly Canadians are not due to aging of the elderly population.
We then considered a second explanation, whether the overall decline in mobility rates may be due to declines in particular components of mobility (that is, local movers, intraprovincial movers, interprovincial migrants, or external migrants). If there were sharp declines in one or two components, this might have resulted in substantial decreases in overall mobility. Our analysis, however, reveals that there were decreases in each of the four components (shown in Figures 1 and 2). This suggests that analysis of temporal changes needs to investigate factors associated with each of the four mobility components, and this requires a multivariate model with temporal, individual, and contextual factors.
A third explanation for declining elderly residential mobility is that the population changed over time, and these changes have affected mobility levels. We investigated this explanation with a multinomial logit model that predicts the components of mobility over time. From this model, we calculated the predicted probability for overall mobility for each census year, holding constant all other explanatory variables. This shows temporal trends that take into account all other influences on elderly mobility. There is an observed decline in overall mobility of 10.6 percentage points from 1971 to 2006. The predicted rates from the multinomial model show a 9.0 percentage point decrease, which implies that 1.6 percentage points are accounted for by the influence of explanatory variables in the model. The variables in the model, therefore, account for about 15 per cent (1.6 divided by 10.6) of the temporal decline in overall mobility rates.
This third explanation is helpful because it tells us that the 2006 elderly population, compared to the 1971 population, had some characteristics associated with lower overall mobility. Nonetheless, the multivariate model does not account for most of the downward trend in elderly mobility.
If there are temporal effects that are not included in the multivariate model that reduce elderly mobility and these factors have been increasing over time, they would contribute to a decrease in overall elderly mobility. For example, if elderly Canadians are more satisfied with their current living arrangements than in previous times, then they may be more likely to remain in their present place and not move. This is an example of a possible temporally related factor that is not included in the multivariate model. To investigate this type of explanation, future research needs to explore new ideas with different data. For now, we can only speculate on alternative explanations that suggest avenues for further study. We briefly discuss four such possibilities.Footnote 15
First, one important factor influencing elderly mobility is health status, but Canadian census data do not have such information, and our model is missing this crucial variable. It is apparent, however, that health status affects elderly movement. For example, elderly people may move to an institutional setting because of poor health. Such moves would not be captured in public-use census microdata that exclude persons in non-private households.
A second important factor is missing information on the location of family and relatives (such information is also missing in census data). If an elderly person wants to maximize proximity to relatives, he/she may remain in the same house if a relative is nearby. This would reduce elderly mobility. However, an elderly person may move to a different community if relatives live there. It is also possible that some elderly adults move to distance themselves from relatives because of personal dislike. Both possibilities would increase elderly mobility.
Third, there have been major changes in transportation during the past 35 years, including greater ease and lower cost of travel. Elderly Canadians can more easily drive to nearby communities, or fly quickly and more cheaply to places farther away. This suggests a third possible explanation for decreases in elderly mobility rates, one that cuts both ways. Easier and cheaper transportation may lead elderly Canadians to remain in place because they can travel to visit family and friends. At the same time, they can move and still travel back to visit family and friends. In short, easier and cheaper transportation may decrease or increase elderly mobility. This is a difficult possibility to explore. Researchers would require data on the location of family and friends before and after migration, as well as data on the relative transportation costs for visiting family and friends.
Finally, a fourth possible explanation is the influence of birth cohorts. The highest elderly mobility rates were in the 1970s. Most of the elderly population in 1976 would have been aged 65 to 75, and would have been born between 1901 and 1911. These cohorts grew up during the economic depression of the 1930s, were young adults during World War II, and completed their childbearing in the 1940s and 1950s. They enjoyed a period of post-WWII prosperity and began to retire in the 1960s and 1970s. Were these birth cohorts unusually mobile, having moved more before, during, and after WWII? Did their post-WWII prosperity influence their later mobility? Upon retirement, were they more motivated to move to a different home or community? These are intriguing questions for further study. Methods for examining age-period-cohort effects (Yang & Land, Reference Yang and Land2013) are potentially promising to explore possible birth cohort effects on elderly mobility.
While the findings from this temporal analysis of elderly residential mobility raise many questions for further study, the overall decline in elderly mobility rates in recent decades has important implications for Canada’s aging families. As more aging families remain in their communities, aging-in-place seniors may benefit from greater community stability and cohesion, and stronger social ties to the community. In turn, more stable communities that have more growing numbers of seniors could respond by increasing amenities that appeal to seniors, including more sidewalks for walking (and wider smoother sidewalks to accommodate walkers and scooters), and situating more health care and other services in the community.
Multinomial logit analysis for elderly mobility status: Local movers (base outcome is non-movers)
Multinomial logit analysis for elderly mobility status: Intraprovincial migrants (base outcome is non-movers)
Multinomial logit analysis for elderly mobility status: Interprovincial migrants (base outcome is non-movers)
Multinomial logit analysis for elderly mobility status: External migrants (base outcome is non-movers)
a All external migrants are from outside Canada.
