Residential mobility across the life course: Continuity and change across three cohorts in Britain Advances in Life Course Research

Although a buoyant literature has emerged examining residential mobility across sections of the life course, a full life course perspective has remained lacking. This paper exploits an as yet under-used data source – the English Longitudinal Study of Ageing – to achieve this. The lifetime residential mobility trajectories of older men and women in three birth cohorts born between 1918 and 1947 are compared, examining how these are associated with changes in cohort members ’ socio-historical contexts, and life course events in the domains of employment, partnership and fertility. Results indicate that change in residential mobility between cohorts is gendered, with persistent continuity between male cohorts, and marked change between female cohorts. Such gender differentials are particularly notable during young adulthood, highlighting the signi ﬁ cance of de-standardising pathways to adulthood and the changing role of women in society. Generalised mobility pathways from birth to age 60 for men and women are identi ﬁ ed using sequence analysis, and the paper discusses how these may be associated with contextual changes and life course characteristics. In conclusion, the research re ﬂ ects on the bene ﬁ ts of the life course perspective for understanding the complexities of residential mobility, and the importance of socio-historical context in understanding trends and patterns in this area.


Introduction
The widespread adoption of the life course perspective across the social sciences has given rise to a burgeoning literature examining partnership, family and occupational trajectories (Berrington & Diamond, 2000;Wu, Bumpuss, & Musick, 2001;Elzinga & Liefbroer, 2007;Hagestad & Call, 2007;Heinz, 2007;Sefton, Evandrou, & Falkingham, 2011). Residential mobility has become more prominent within life course research in recent years (Findlay, McCollum, Coulter, & Gayle, 2015). Some studies have identified the importance of particular life course events as 'triggers' and 'constraints' for residential moves (Clark, 2013;De Groot et al., 2011;Michielin & Mulder, 2008), while others have focused on the synchronicity of mobility trajectories with occupational and family careers (Clark & Davies Withers, 1999, 2009Mulder & Cooke, 2009). Like most other life course research, studies of residential mobility have tended to examine trajectories in a piecemeal way, focusing on sections of the life course. This has in part stemmed from the lack of appropriate data tracing individual residential mobility across the full life course.
To address this gap, this paper takes advantage of the English Longitudinal Study of Ageing (ELSA) life history module, which collected lifetime residential histories from respondents in 2006/ 2007. Drawing upon these data, residential mobility across the life course in three birth cohorts (born between 1918 and 1947) is compared. The paper explores how these dynamics may be associated with the socio-historical contexts of respondents and how residential mobility is interconnected with other life events (partnership, fertility and employment). Building on this, sequence analysis is used to identify how the individual lifetime mobility pathways of ELSA respondents aged 60 and over are clustered into common 'types'.
In examining cohort and gender effects across the whole life course up to age 60, we advance what has been achieved in more focused, single-cohort, studies of residential mobility across particular life course stages by shedding light on how the accumulation of experience, advantage/disadvantage, and changing socio-historical context across long periods of time can shape individual long-term pathways. The application of sequence analysis then enables us to examine within-population variation in residential mobility over the life course, including identifying those who stay put i.e. non-movers, as well as movers. This represents a significant advancement for residential mobility research using innovative longitudinal methods, drawing upon an underused data source. The paper begins with a review of preexisting literature examining the intersections between residential mobility and the life course, before outlining the theoretical framework for the study. It is recognised that the paper is essentially historical in nature, focussing on moves that have taken place from the 1920s-1990s. Nevertheless the findings have relevance for understanding lifetime mobility for both past and current cohorts.

Residential mobility and the life course
Interest in the connections between residential mobility and the life course can be traced back to Rossi's (1955) pioneering work on intra-urban migration almost 60 years ago, but has gathered pace over the last 10-15 years as the life course perspective has become more prominent in geographical research (Mulder & Hooimeijer, 1999). This is evidenced by a proliferation of research on family migration (Cooke, 2008;Michielin & Mulder, 2008;Clark & Davies Withers, 2007), mobility and life course transitions (Finney, 2011;De Jong &Graefe, 2008 De Jong andGraefe, 2008), life course events as mobility triggers (Clark, 2013;Mulder & Wagner, 2010) and the synchronicity of mobility, employment and family trajectories (Clark & Davies Withers, 2009;Geist & McManus, 2008).
Before the emergence of the life course as a theoretical framework within mobility studies, age-mobility schedules were a common feature of research in this area. The pioneering work of Rogers and Castro (1981) in modelling migration age-schedules may be seen as an early attempt to introduce a life course perspective using cross-sectional data. Rogers and Watkins (1987) identified four mobility peaks associated with particular age groups and 'stages' of life: early childhood (0-3 years, indicative of family migration), early participation in the labour force (17-30 years), retirement (57-67) and late old age (80)(81)(82)(83)(84)(85)(86)(87)(88)(89)(90). More recently synthetic measures of lifetime migration using cross-section data such as gross migraproduction rates and migration expectancy have been developed, with estimates of the average number of moves people make in a lifetime being derived using age specific mobility rates in a similar manner to that in which total fertility rates calculated (for more detailed discussion see Rees, Bell, Duke-Williams, & Blake, 2000;Bell et al., 2002). As the life course perspective has become more prominent, there has been a shift in focus onto transitions and events. Warnes (1992) identifies specific transitions that can be associated with increased mobility including: leaving the parental home, career promotion, divorce, retirement, frailty or chronic ill health; and indicates an age range within which these transitions may occur. This is a helpful development for theorising the life course-mobility interface, but still relatively restrictive in the sense that it does not allow for life course diversity (i.e. individual pathways including many or few of these transitions, gender differences etc.).
The recent buoyancy in life course research on residential mobility has been fuelled by the increasing availability of longitudinal data and propelled by the application of longitudinal methods to examine the timing, sequencing and synchronicity of mobility and other life events (see for example Blossfeld, 2001;Blossfeld & Mills, 2001). Clark and Davies Withers (2009) defined 'windows' around events that trigger mobility (such as fertility and labour market changes including entering, leaving or losing employment, changing employer, changing place of employment) to examine their synchronicity, using the Panel Study of Income Dynamics from the USA, while others have used event history techniques to trace sequential change (e.g. De Jong & Graefe, 2008) Such studies reach similar conclusions about the complexity of residential mobility decision-making, involving interactions between trajectories in multiple areas of the life course. The value of the life course approach for revealing this complexity is highlighted by the role that other dimensions, or trajectories, of individuals' lives play in influencing mobility decisions (Mulder & Hooimeijer, 1999). This echoes an earlier call by Halfacree and Boyle (1993) for more in-depth analysis of the complexities of migration decision-making.
What these studies also have in common is that they tend to focus on sections of the life course-in particular the mid-phase where individual moves are made as part of a move affecting the whole family. This is often due to the limited timespan of longitudinal data sets. What sets our study apart is the use of individual mobility histories from birth to later life, enabling us to trace individual mobility pathways across the life course. This is achieved through the optimal matching of individual mobility sequences (see methods section for a fuller description), through which clusters of 'typical' mobility pathways are identified.

Theoretical framework
Outside of mobility studies, life course scholars have often framed their research in the work of Beck (2000), Beck and Beck-Gernscheim (2002), Beck and Lau (2005), Beck, Bonß, and Lau (2003) and Giddens (1991), who trace the transition from first to second modernity. Key changes include the growth of individualism; the de-standardising of normative pathways through the life course; the changing role of women in society; population restructuring and the changing structure of the life course; changing partnership and family formation; globalisation and increased mobility; and increasing social and spatial inequalities (Wadsworth & Bynner, 2011).
In the 1970s, mobility scholars explicitly used the transition to modernity as a framework for theorising mobility over the last century (for example Zelinsky's 'Mobility Transition', 1971). These theorisations are largely deterministic and have been widely criticised for this reason by contemporary mobility scholars (Woods, 1993;Cadwallader, 1993). More recently, the transition to modernity has implicitly underpinned the use of structuration theory to transcend the structure-agency dichotomy, instead stressing "the actions of contextualised individuals" (Boyle, Halfacree, Robinson, & Boyle, 1998: 81;Giddens, 1984). The emergence of the biographical approach (Halfacree & Boyle, 1993) built on structuration theory, allows us to reinterpret the life course approach for application within mobility studies.
We align with Halfacree and Boyle's (1993) biographical approach; accordingly we aim to understand the mechanisms behind mobility events across the life course, and how these relate to the socio-historical context of the respondents making decisions related to mobility. Below we draw upon pre-existing research to outline the main mechanisms for mobility that occur during each life course phase and identify some of the major socio-historical landmarks that are likely to have impacted on the mobility of the specific birth cohorts within the ELSA sample that are the focus of this paper, i.e. those men and women born between 1918-1947.

Cohorts
ELSA was designed to help understand the "unfolding dynamics of ageing and the relationships between economic circumstances, social and psychological factors, health, cognitive function and biology as people move through retirement into older age" (Banks, Nazroo, & Steptoe, 2012;p1).The survey collects data from a representative sample of the English population aged 50 and over relating to their health and disability, economic circumstances, social participation, networks and well-being (Marmot et al., 2015).The original sample of $12,000 respondents was drawn from the Health Survey for England. This paper analyses the life histories data, collected retrospectively at wave 3 (in 2006/2007). The analytical sample used in the paper includes respondents aged 60-89 at the time of interview, divided into three birth cohorts: cohort 1-born 1918-1927 (aged 80-89 at interview i.e. today's current older-elders); cohort 2-born 1928-1937 (aged 70-79 at interview i.e. today's current mid-elders); cohort 3-born 1938-1947 (aged 60-69 at interview i.e. today's current younger-elders).
The remainder of this section aims to build a picture of how the life events experienced by the ELSA respondents in these three cohorts may have aligned with their respective socio-historical contexts, giving rise to particular mobility pathways. Table 1 provides a summary of key socio-historical landmarks that have occurred during the sample's lifetimes, and indicates the age range of the individuals in the three birth cohorts when these occurred.

Context
a) Mobility during childhood The perspectives of children and childhood experiences of residential mobility are relatively absent from migration studies (Dobson, 2009). However, literatures examining family (household) migration over the last 10-20 years have identified fertility as a common 'trigger' for residential movement (Clark & Davies Withers, 2009), particularly the birth of a family's first child (Clark, Deurloo, & Dielman,1994). Other work in Europe and the USA has found that the presence of older children is associated with less residential mobility, particularly during the ages children are going to school (Michielin & Mulder, 2008). This suggests that as children, the ELSA respondents were most likely to have experienced heightened mobility at young ages (0-5) while their family housing adjusted to fit a growing family (Clark, Deurloo, & Dielman, 2003).
Another main trigger for residential mobility is family dissolution, especially divorce. The three birth cohorts born between 1918 and 1947 experienced childhood (defined as age 0-18) across the period 1918-1965; five decades during which significant social, economic and political changes unfolded, including the Great Depression of the 1930s, the Second World War (WW2), the expansion of the modern welfare state in Britain, and the emergence of post-modernism and the new youth cultures of the 1960s (Wadsworth & Bynner, 2011). However despite these significant societal changes, the family context of childhood for the three birth cohorts remained relatively consistent. From 1921-1961 the marriage rate in Britain fluctuated around 45-55 per 1000 (unmarried women aged 16 and over), peaking at 63 per 1000 in 1940 as young couples rushed to marry before being separated by war (ONS, 2013), divorce rates were consistently low and relatively few children were brought up by single parents (Allan, Hawker, & Crow, 2001). It would be plausible to assume, therefore, that within all three birth cohorts childhood mobility due to family fragmentation was relatively low. The main exception to this would have been during WW2 when families were separated and some children may have been evacuated; mobility during times of crisis is further discussed separately below. b) Mobility during young adulthood Leaving the parental home Boyle et al. (1998) identify three main pathways out of the parental home during contemporary young adulthood: for entry to Higher Education (HE); to form a partnership or for first employment. The respondents in the three birth cohorts were in their early twenties between the late 1930s and 1960s, prior to the significant expansion of HE. Pre-1950 participation in HE was around 3%, rising to 8% in 1970, before a period of rapid expansion in the 1990's saw it increase to 20% (Dearing Report, 1997). It is likely therefore that only a relatively small proportion of the youngest cohort (born 1938-1947) largely limited to those from higher socio-economic backgroundswould have left home to attend University, while those current elders who experienced young adulthood during the 1930s-1950s, and those from lower socio-economic groups, were more likely to have left home to form a union or for first employment (cohorts 1 and 2).
Forming a partnership During the immediate post-WW2 period, when cohort 1 was in their 20s, large proportions of young couples began married life living in the parental home, reflecting housing shortages following the destruction of homes during the war (Holmans, 1981). This was a brief anomaly, however, in an otherwise fairly consistent trend since the pre-industrial period whereby marriage in England and Wales was associated with forming a new household, and thus a residential move (Grundy & Harrop, 1992). It would be plausible to assume then that the first marriage of ELSA respondents born between 1928 and 47 (cohorts 2 and 3) was accompanied by a residential move. Later home-leaving in the early part of the Century (cohort 1) may also reflect different decision-making processes as, according to Grundy (1992), during this period departure from the parental home was likely to have been determined to a great extent by the needs and wishes of parents, rather than the desires of their young adult children as would have been the case during the 1960s (cohort 3) (Boyle et al., 1998).
First employment Given the changes in the labour market across the last century, experiences of moving to first or new employment are likely to vary significantly by cohort. Boyle et al. (1998) note that young adults in the early part of the 20th century (cohorts 1 and 2) who did not leave home to marry were unlikely to form a new household but would lodge with landlords, employers or relatives, implying more unstable mobility pathways than those who left to form a new household with a partner. For cohort 3 (born 1938-1947), moving into shared accommodation with other unrelated young adults, or even forming solo households in the new place of employment was more of a possibility. Although young adult solo living was still relatively rare during the 1960s, by the early 1980s around 11% of young adults aged 20-24 lived in a shared household with unrelated others (Demey, Berrington, Evandrou, & Falkingham, 2011). This suggests that we might observe increasingly unstable mobility trajectories as this type of household is characterised by high occupant turnover. c) Mobility during mid-life Labour-motivated mobility The housing market and tenure constraints have been perceived as important factors in inhibiting labour mobility (Forrest & Murie, 1992). Housing policies were introduced during the 1980s and 1990s (such as the 'right to buy' in 1980) to free up mobility constraints imposed by social housing and encourage outmigration from areas of under-employment (see 'housing' column of Table 1). It has since been accepted that the link between tenure and mobility is far more complex, and the large-scale interregional flows of labour-migrants expected in response to this legislation did not materialise. However, it is likely that the increase in owner-occupation post-1980 encouraged some labour mobility, having the most pronounced effect on the youngest cohort (3) who were aged 33-42 when the 'right to buy' was implemented.
It is very likely that female cohort members pursued different employment-related mobility pathways to their male counterparts. Salt (1990) shows that relocation within the same company is far rarer for females than males, and is largely restricted to younger, childless women. Gordon (1995) quantifies this disparity, finding women are two and a half times less likely to be relocated in this way. The 'tied migrant' thesis suggests that the trailing female spouse adheres to the mobility pathway of her 'breadwinner' husband (Boyle, Cooke, Halfacree, & Smith, 2001). This is likely to have determined the mobility pathways of women in cohort 1, while women in cohorts 2 and 3 are more likely to show signs of 'untying' these bonds as dual career households became more prominent in the latter half of the twentieth century (Smith, 2004). Divorce As noted above, until the last quarter of the 20th Century, divorce was rare in England and Wales. The Divorce Reform Act in 1967 (implemented in 1971) marked a momentous change whereby legal separation could be achieved on the grounds of irretrievable break-down of the marriage as opposed to reasons such as adultery or unreasonable behaviour. This change witnessed a significant rise in the number of divorces from approximately 26,000 in 1961, when cohort 1 were aged 34-43, to just under 126,000 ten years later in 1971-when cohort 2 were aged 34-43. The divorce rate continued to increase significantly after the legislative change from 2.1 divorces per 1000 marriages in 1961, to 9.5 per 1000 in 1971, and peaking at 14.2 per 1000 in 1994 (ONS, 2013). Previous research has found that divorced and separated men and women have far higher mobility rates than those who are married (Feijten & Van Ham, 2007). It is likely, therefore, that members of cohort 3 will be more mobile during mid-life than members of cohorts 1 and 2, due to the timing of the Divorce Reform Act in their lives (see Table 1). a) Mobility during times of crisis Times of political and economic crisis during the first half of the 20th Centurymost notably the Great Depression of the 1930s, WW2, and the urban and economic restructuring that followedwould undoubtedly have influenced childhood and young adult migration. Latterly, recessionary periods in the UK during the 1980s and 2000s are likely to have impacted on ELSA respondents as adults in mid and later life, in particular cohort 3 who were in their thirties during the 1980s and thus were young enough to have experienced both recessionary periods during their working life. Grundy (1986) identifies two studies (Thompson, 1938;Goodsell, 1937) which find that constrained residential mobility during the Great Depression of the 1930s was a factor contributing to depressed fertility rates during this period (as housing consumption could not be adjusted to accommodate children), which may have affected cohort 1's childhood mobility, as they were aged 2-11 when the stock market crashed in 1929. Others, however, caution against underestimating the "strong unrelenting desire to move" which can propel residential mobility even during a global economic crisis (Forrest, 1987(Forrest, : 1617. Although this is most notable in higher socio-economic groups, it has also been evidenced that people move to escape unemployment (Böheim & Taylor, 2003), suggesting that ELSA respondents from across socioeconomic groups may have retained much of their propensity to migrate during recessionary periods, albeit for different reasons. WW2 changed the patterns and processes of residential mobility in Britain both in the short and long-term, by stimulating migration during war-time and broadening horizons and attitudes towards future mobility (Pooley & Turnbull, 1998). Movement motivated by conventional reasons (space-adjustment, employment etc.) was supressed during WW2; however, forced evacuation from Britain's cities introduced a 'synthetic' urban-to-rural flow, while bomb damage caused many who remained in urban areas to move residence. Many women migrated to war-time employment in munitions factories and agricultural settings, while some families pooled resources by sharing housing (Summerfield, 1989). When war-time ended, some families reunited and re-dispersed causing heightened mobility during the post-war period, while others who suffered the loss of a family member may have sought new accommodation better suited to their new circumstances. WW2 was experienced by cohorts 1 and 2, and a small proportion of cohort 3, causing disruption to childhood and young adult mobility trajectories (see Table 1). It is also likely that some male respondents from cohort 1 served in WW2, and hence travelled significantly around Europe and the UK; such experiences may have broadened horizons potentially inculcating a predisposition to be mobile later in life.

Military service
All of our cohorts were too young to have been involved in military action during the First World War; compulsory military service was terminated in 1920, when the older members of cohort 1 were aged just 2. Conscription was however reintroduced in September 1939 at the outbreak of the Second World War. Initially this was limited to all healthy men aged 18-41 years old, but by 1942 it had been extended to 18 and 51 years old men and all females 20-30 years old resident in Britain, thus potentially affecting both women and men in cohort 1. Following the end of the war, from 1949 onwards healthy males aged 17-21 years old were expected to serve in the armed forces for 18 months, a period known as National Service. National Service was, however, slowly phased out from 1957 and men born on or after 1 October 1939 were not required to undertake military service. As discussed above, some respondents in cohort 1 may have served in WW2 and members of cohort 2 may have been affected by National Service, but cohort 3 are largely been unaffected by military service unless they chose to join any of the armed services voluntarily. This section has outlined the primary mechanisms of residential mobility during each phase of the life course, drawing upon preexisting research, and has also outlined aspects of socio-historical context that are likely to have influenced the mobility trajectories of the three birth cohorts that are the subject of this enquiry. The remainder of the paper builds on this context to address the following research questions. First, how does life course residential mobility vary for men and women, and across cohorts? Second, how is life course residential mobility associated with 'trigger events' in the domains of fertility, partnership and employment amongst these three cohorts in twentieth century Britain? Finally, can distinct pathways of mobility be identified within cohorts?

Methods
As outlined above, our analyses begin by comparing aggregate rates of mobility over the life course between cohorts. This is followed by a second stage that uses sequence analysis to explore the clustering of individual mobility trajectories within cohorts. In both analyses, separate models are estimated for men and women, given that distinct patterns of mobility by gender are expected. The timeline is truncated at age 60 in order for all individuals in the sample to have a uniform sequence length and facilitate comparison between the three cohorts.

Measure of mobility
Life history grids are increasingly being used to collect retrospective life course data, with the use of calendars enabling respondents to cross-reference certain life-events with others being found to reduce recall bias (Freedman, Thornton, Camburn, Alwin, & Young-DeMarco, 1988). ESLA included a life history grid in wave 3, with information collected on a range of items including children (natural and adopted), cohabiting and important noncohabiting relationships, housing and mobility and jobs and earnings (Ward, Medina, Mo, & Cox, 2009). Mobility is defined as any change in residence reported in the life history data, with changes recorded annually. The exact question working in the life history grid is 'For each accommodation lived in for more than 6 months: Date started lived in accommodation; Date stopped lived in accommodation' (Ward et al., 2009: p37). The classic distinction between long-distance (employment-related) and short-distance (housing-related, i.e. residential mobility) moves has not been adopted, as the data do not enable this division. The ELSA life history grid did collect information on the address for each residence, but this information has been removed to ensure data confidentiality (Ward et al., 2009). It should also be noted that within ELSA, there is a variable on type of residence which includes the armed forces as an answer category. We did not investigate this residence type explicitly, but can infer from the inclusion of armed forces as a reference category that residential moves associated with military service are included.

Analysis 1: cohort comparison
To compare the rates of mobility between cohorts, the annual data from the life histories are pooled in order to create a pairedyears, person-period dataset for use in logistic regression analysis. As described above, the outcome variable is residential mobility between ages 0 and 59, defined as any change in residence between two years. To account for clustering within individual trajectories of mobility, a multilevel model is estimated including a random effect at the individual level. The main independent variable is cohort, included as a categorical predictor (1918-1927; 1928-1937; 1938-1947). However, we are interested not only in comparing the absolute level of mobility between cohorts but also in differences in age-based patterns of mobility. Therefore, a main effect has been included for an individual's age band and an interaction between the age band and the cohort.

'Trigger' events
In addition to the residential histories, the ELSA life history data also include information on partnership, fertility and work over the life course. We can therefore investigate the extent to which mobility coincides with other life events in these domains. Five potential 'triggers' for mobility are included: 1. A child being born; 2. a new partner 1 ; 3. a union dissolution 2 ; 4. becoming unemployed or inactive; 5. starting a new job. 1 The retrospective histories do not distinguish between cohabiting and formal marital unions. However, prior to the 1970s cohabiting unions were uncommon.
2 Union dissolution includes the ending of a partnership for any reason and thus does not distinguish between separation, divorce and widowhood.

Other covariates
In addition, a number of time-varying and fixed covariates have been derived from the life histories and from the Wave 3 interview data. The time-varying measures, derived from the life histories and recorded annually, are marital status, economic activity, parity (number of children ever born) and housing tenure. Fixed measures from the Wave 3 interview comprise one's highest educational qualification, occupational social class (NS-SEC-3) 3 and the number of (living) siblings. Unfortunately we only know about the number of living siblings rather than the actual number; however, this measure gives an indication of family size during childhood, which is likely to influence rates of mobility in early life (Clark et al., 1994).

Analysis 2: individual trajectories
In order to investigate the extent to which individual mobility trajectories are clustered into distinct 'types', sequence analysis is applied, specifically optimal matching analysis, using the SQOM command in STATA 12. Optimal matching, introduced to the social sciences towards the end of the twentieth century (Abbott & Forrest, 1986), has subsequently been used to analyse a range of social and historical phenomena, from the structure of English folk dances (Abbott & Forrest, 1986) to patterns of work-life social mobility (Halpin and Chan, 1998) and economic activity (Stone, Evandrou, Falkingham, & Vlachantoni, 2015). The advantage of the technique is that it allows analysis of entire sequences of data simultaneously, with entire trajectories as the basic unit of analysis. In the present analysis, the data from all three cohorts are pooled and the age-based time-line collapsed into five-year bands (running from age 0-59 years). The number of moves occurring in each five year band is then counted, creating a sequence with 12 time-points for each respondent, each with a value ranging from 0 to 5.
Optimal matching applies an algorithm to calculate the 'cost' of transforming one sequence into another, based on the minimum number of insertions, deletions or substitutions required (Macindoe & Abbott, 2004). The user can specify the cost of substitutions between states using a substitution matrix; in this analysis we simply use the raw distance between values-e.g. transforming from zero to three incurs a 'cost' of three. The 'cost' of simultaneous insertion/deletion (usually referred to as the 'indel' cost) is set at two.
The output of optimal matching is a matrix that specifies the 'distance' between each pair of sequences. This distance matrix has been used as the basis of a cluster analysis using Ward's linkage, which produces clusters based on a weighted average of distances between variables. Following evaluation of the diagnostics from this analysis and examining a number of cluster solutions, the results for a six-cluster typology of residential mobility for men and a four-cluster typology for women are reported.

Results
The findings from the two stages of analysis are outlined below, starting with the comparison of residential mobility between three birth cohorts defined within the ELSA sample.

Findings from analysis 1: cohort comparison
The cohort comparison begins by presenting some descriptive statistics in Table 2, which show key life course characteristics by cohort and gender. The mean number of moves for men and women in cohort 3 is 6.2. This is remarkably similar to the gross migraproduction rate (GMR) of 6.26 computed for Great Britain 1990-1991 for ages 0-75+ from the 1991 Census (Rees et al., 2000). It is important to note that the results are for survivors only and thus may be affected by selection biases. Table 2 shows that, aside from expected structurally-driven changes such the rising age at leaving school across cohorts, the most striking cohort differences are apparent in the female sample. The oldest cohort (cohort 1) spent less of their working life in employment than cohort 3, reflecting changing female participation in the workforce post-WW2.They were also less mobile up to age 60 years, with an average of 5.4 residential moves compared to 6.2 in cohort 3. While female labour force participation increases across birth cohorts, the opposite is true for men, further reflecting changing gender roles across the twentieth century. Figs. 1 and 2 show the predicted probabilities of residential mobility by cohort across the life course for men and women. These are calculated from logistic regression analysis estimating the annual probability of mobility predicted by the interaction between cohort and age group, with no covariates. As would be expected from pre-existing analyses of mobility by age, the predicted probability of moving increases during young adulthood for all cohorts, then the likelihood of moving decreases during the late twenties/early thirties and continues to decline across mid-life until retirement. For men (Fig. 1), cohort differences in the predicted probability of moving are very small, despite being statistically significant at most ages. The main patterns indicate that the oldest cohort (1) was more mobile during adolescence (age 10-19), but less mobile than cohorts 2 and 3 during young adulthood (age 20-29). It is possible that the older cohort's increased mobility during childhood was associated with disruption caused by WW2 (which occurred when cohort 1 were aged Table 2 Key life-course characteristics of three cohorts, by sex.

Men Women
Means, age 0-60 Cohort  1  2  3  1  2  3  Born 1918-1927 1928-1937 1938-1947 1918-1927 1928-1937 1938-1947  3 National Statistics Socio-economic Classification (NS-SEC) is the primary social classification in the United Kingdom. It was adapted from the Nuffield class schema developed in the 1970s and is based on type occupation and the size of employer. The collapsed three category classification is hierarchical with 1 = higher occupations (managerial and professional); 2 = Intermediate occupations (e.g. clerical, sales, service, lower supervisory); and 3 = lower occupations (semi-routine and routine manual occupations).
12-21), and that cohorts 2 and 3 were more mobile during young adulthood in response to the destandardisation of pathways to independence. In contrast to the continuity observed across the male birth cohorts, Fig. 2 shows marked evidence of (statistically significant) cohort change in the mobility trajectories of women, particularly during young adulthood. The predicted probability of moving at age 20-29 increases from 0.17 in the oldest cohort to 0.23 in the youngest cohort, reflecting the effect of de-standardising pathways to adulthood and the changing role of women in society from the 1960s onwards. Table 1 highlights several sociohistorical landmarks that might be associated with the changing Fig. 1. Predicted probability of residential mobility by age and cohort: men. Fig. 2. Predicted probability of residential mobility by age and cohort: women. Fig. 3. Predicted probability of residential mobility by age and cohort: men (covariates: trigger events, NSSEC, housing tenure, marital status, parity, economic activity, educational attainment, number of living siblings). All covariates set at baseline values. role of women and the timing of these within respondents' lives, including the Equal Pay Act (1970), the Sex Discrimination Act (1975), and the introduction of the contraceptive pill (1961 for married women; 1967 for all women). These landmark events occurred when cohort 3 were in their late teens to early thirties (14-37) and thus are likely to have significantly shaped their young adult trajectories.
Figs. 3 and 4 show the predicted probabilities of moving across the life course after adjustment for all covariates. Although some of the trends are now ameliorated, these results demonstrate that there remain significant differences between cohorts, particularly for women. Table 3 shows odds ratios for the full logistic regression model from which the predicted probabilities in Figs. 3 and 4 are derived. Many statistically significant predictors of mobility can be seen that follow expected patterns À for example, owner-occupiers are the least mobile group, mobility decreases with parity, and being in education raises the odds for mobility as compared with being employed. However, for the purposes of the present paper, we are particularly interested to observe that all of the trigger events show a significant association with residential mobility even after taking all of these other predictors of mobility into account (although we make no assumptions about the direction of causality, therefore we interpret them as 'triggers' with some caution). More specifically, partnership appears to be a stronger influence than fertility or employment. This applies to both partnership formation and dissolution, but the former is by far the strongest predictor: forming a new partnership confers odds ratios for residential mobility during the same year of 21.17 and 25.08 for men and women respectively, compared with years during which a partnership was not formed.
Although a number of significant predictors of residential mobility are identified, the random effect (presented at the bottom of Table 3), indicates that there is significant residual variation attributed to individual-level differences that is not explained by this model. These differences are explored below, drawing upon optimal matching of individual mobility sequences. Fig. 5 shows the six mobility clusters (or 'types') that emerged from analysis of the male sample, and Fig. 6 shows the four clusters that emerged from the female sample. Tables 4 and 5 present key characteristics for each of the clusters respectively.

Findings from analysis 2: individual trajectories
Cluster 1 for both men and women has been termed 'infrequent movers' to describe the lack of mobility evident in these groups across the life course (with the lowest average number of moves between ages 0-60 of all the clusters, at 3.9 for women and 4.2 for men). Despite their limited movement, the classic peak in mobility during young adulthood (at age 25-29) remains evident. Cluster 1 represents the most common mobility trajectory for men (37%) and women (39%), reiterating that migration events are rare occurrences in the life course (Cooke, 2011). The second cluster to emerge from the male and female samples (cluster 2, Figs. 5 and 6) is titled 'early childhood movers', since a peak in mobility is marked at age 5-9, in addition to the expected peak in young adulthood. Based on the average age of men and women in this cluster (70.9 and 71.1, respectively-see Table 5), this heightened mobility during childhood could be a result of disruption caused by WW2, as discussed earlier. Male cluster 3 also shows evidence of heightened mobility during childhood, although a greater consistency of movement is apparent across childhood and into young adulthood (reflected in the cluster name 'early life movers'), and peak mobility occurs at age 15-19 (as opposed to 20-29 in all other clusters). It is possible that the adolescent peak in mobility reflects a greater propensity among this group for movement out of the parental home for work (as opposed to partnership) on finishing education, perhaps to lodge with an employer.
Also showing evidence of high mobility across childhood and young adulthood are males in cluster 6, the 'persistent early life movers'. Respondents within this cluster are differentiated from the other early life movers by experiencing consistently high levels of movement from childhood through to their early thirties, with a particularly high peak in their twenties. This appears to be a socioeconomically advantaged group, with more than half (50.9%) in the highest social class at interview, and showing evidence of leaving education later than most other clusters (Table 4). This suggests some of this mobility may have been associated with moving away to attend university, then onwards to the labour market, and perhaps postponing union formation until later in adulthood than others-indeed, these men reported forming their first union at an average age of 26.2, compared with 25.3 in the total sample.
Female and male cluster 4 have very similar trajectories, which are characterised by a steep rise to peak mobility at age 20-24 of 2.2 moves, then returning to the normative trajectory from age 35 onwards, hence being termed the 'young adult movers'. They also left education later than most other clusters (the latest for female cluster 4 at an average age of 16.4). Interestingly, on average women in this cluster are younger than in the other three groups and they are most likely to be in the highest social class, suggesting increased opportunities for education and work for women in younger cohorts. More generally, cluster 4 is the second largest Fig. 4. Predicted probability of residential mobility by age and cohort: women (covariates: trigger events, NSSEC, housing tenure, marital status, parity, economic activity, educational attainment, number of living siblings). All covariates set at baseline values. *p<0.05 ** p < 0.01 *** p < 0.001 Table 3 Odds ratios for residential mobility over the life course.

Men Women
Odds ratio for residential mobility Odds ratio for residential mobility male and female cluster, accounting for around one-quarter of these samples, which underlines the conventional nature of raised mobility during young adulthood. Although we know from pre-existing research that mid-life is commonly a time of relative stability as opposed to mobility, cluster 5 ('persistent mid-life movers') from the male sample shows evidence of heightened mobility during mid-life. An equivalent cluster is identified in the female sample (cluster 3), with both male and female mid-life mover clusters showing the highest total mobility across the life course (means of 10.4 and 13.0 moves, respectively). Table 5 clearly shows that union dissolution is most common in the 'persistent mid-life movers' clusters, confirming that divorce is a key trigger for mobility during midlife. As might also be expected, the mid-life mover clusters are   Table 4 shows that for men, the distribution of the three birth cohorts is relatively consistent within each cluster. However, for women (Table 5) there are more noticeable differences. In particular, the oldest cohort is especially prominent in cluster 2 (early childhood movers), representing 40% of this cluster, while the youngest cohort is strongly represented in cluster 3, comprising 39% of this 'persistent adult movers' cluster. This further reinforces the potential impact of the Divorce Reform Act for the youngest cohort of women. More generally, the findings support the results from our cohort comparison, suggesting that cohort differences in mobility are particularly important for women.

Discussion and conclusions
In this final section we draw out four main themes from our findings for further discussion: the gendered nature of continuity and change in residential mobility across the life course; the significance of extending/destandardising transitions to adulthood and the changing role of women in society across the twentieth century; the importance of considering socio-historical context as well as life course events when interpreting life course data; and the added value of taking a full life course perspective on analyses of residential mobility.
Our findings identify significant evidence of change in female residential mobility behaviour between cohorts 1, 2 and 3: namely higher levels of mobility during young adulthood in younger cohorts (2 and 3) compared with the older cohort (1). There are a number of trends associated with life course destandardisation that can help to explain these results. Although leaving to form a union was still the most common pathway out of the parental home in the 1970s, participation in HE had more than doubled since the 1950s (to 8%), implying that more young adults were leaving home to attend university. Others will have followed an emergent pathway out of the parental home into a cohabiting household. We find more unstable mobility trajectories in female cohort 3, who were in their twenties and thirties during the 1960s and 1970s, compared to female cohort 1, who were young adults during the 1930s-1950s. Thus young adulthood appears to be the life course phase where women's lives have witnessed the most significant change, and the life course events which occur during this phase of life (e.g. leaving the parental home to form a partnership, to attend HE or to first employment, space adjustment following the birth of a child) appear to be strongly associated with changing residential mobility behaviours in women.
Our findings show that partnership formation and dissolution are the events most strongly associated with residential mobility, although partnership formation has a far stronger association with mobility than dissolution. Although divorce is a rarer event than partnership formation, it is one that has been found to bring about multiple moves before and after the divorce. For example, partners may move from the marital home at different times (Feijten & Van Ham, 2007), form new cohabiting unions, re-marry and move to new homes (Boyle et al., 1998). Although some studies have found that women fare worse than men following divorce because of the male-female wage gap and the greater proportion of women who do not independently earn a wage (Jarvis & Jenkins, 1999), others have found that divorced men are more mobile than divorced women (Feijten & Van Ham, 2007), since the partner with custody of any children (usually the woman) is less likely to leave the marital home (Mulder & Wagner, 2010).
It is therefore surprising that such marked continuity is found in male cohort residential mobility trajectories, and that a more pronounced effect of divorce is not evident in male cohort 3 compared to cohorts 1 and 2 (Mulder & Wagner, 2010). The two older cohorts would not have experienced the same exposure to union dissolution due to the timing of the Divorce Reform Act in 1971; cohort 3 were aged 26-35 and thus have been able to obtain a divorce with relative ease for most of the period. We might also have expected to see some long-term disruption to cohort 3's mobility trajectories later down the life course as a result of divorce (Feijten & Van Ham, 2007).
It is clear from the literature referenced above that divorce is a key determinant of residential mobility during mid-life, but it is still rare enough that it may not reveal its effects in studies of aggregate trends, such as the cohort comparison in this paper. However, the individual trajectory analyses identified a sub-group of men and women with heightened mobility during mid-life that is to some degree associated with union dissolution. This highlights the benefit of combining between-cohort and within-cohort analysis of residential mobility, and of examining individual mobility pathways.
The relatively low total levels of mobility across mid-life apparent in the cohort comparison, and also in most of the mobility clusters identified for men and women, are indicative of the overall rarity of residential mobility events across the life course. With this in mind, we draw attention to the importance of non-moving, or 'staying put', and caution against overstating the significance or prevalence of mobility. Indeed, findings from the sequence analysis indicate that for both men and women, the cluster accounting for the highest proportion of respondents' mobility sequences is one involving minimal residential mobility across the life course, with only a modest peak during young adulthood. However, the findings  (born 1918-1927) 33.4 29.3 36.9 32.1 29.7 26.2 % in cohort 2 (born 1928-1937) 43.4 41.5 38.4 38.9 44.8 48.7 % in cohort 3 (born 1938-1947) 23.2 29.2 24.7 29.0 25.5 25.1 also identify heterogeneity in the lifetime residential mobility trajectories of the ELSA respondents, and this significantly advances what we know from pre-existing research. By presenting a typology of residential mobility and drawing out how the types are associated with different life course characteristics and changing socio-historical contexts, we enlarge the usual lens of enquiry in residential mobility research from life stages to the full life course. In doing so our findings shed light on how changing social, economic and political landscapes both early in life, and further down the life course nearer the occurrence of a mobility event, might impact on the decision to move. A key example of this emerges from our female cohort comparison analysis where it appears that significant changes to women's young adult lives in cohort 3 were brought about by major societal changes during the 1960s and 1970s, reflected in a raft of policies implemented between the ages of 14-37 (e.g. the Equal Pay Act (1970), the Sex Discrimination Act (1975), and the introduction of the contraceptive pill (1961)). However, these cohort differences appear not to be sustained through the rest of the life course, as residential mobility trajectories for cohorts 1, 2 and 3 realign as respondents enter midlife. This interpretation echoes the conclusions of life course studies of other later life outcomes which find it is important to consider changing circumstances and context near the time of an event as well as the accumulation of advantage and disadvantage across the life course (e.g. Naess, Hernes, & Blane, 2006). Such longterm reflections on the impact of changing socio-historical contexts are not possible where studies draw upon datasets tracing respondents across shorter sections of the life course.
The study has a number of limitations, most notably the lack information on the distance of the residential moves. Moreover it is not possible to distinguish residential mobility as a choice as opposed to other circumstances such as an eviction, although such 'forced' events remain quite rare. 4 It is also important to bear in mind that the findings may be affected by selection bias, as necessarily life history data are only available for survivors and survivorship is not independent from other characteristics such as partnership status, education and social class. Notwithstanding these limitations, the results have underlined the value of considering socio-historical context as well as life course events and reflecting on change across the full life course. Although the life history data required to achieve this type of analysis are rare and can be beset with data quality issuesespecially when collected retrospectively thus subject to memory recall and other biases (e.g. Nelson, 2010) the added value of taking a full life course perspective for advancing conceptualisations of residential mobility is clear. Although residential mobility scholars have embraced the life course perspective, it is suggested that greater engagement with the socio-historical context of survey samples, and taking fuller advantage of life history data (for example ELSA, and the European equivalent the Survey of Health, Ageing and Retirement in Europe (SHARE-Life) would reveal more about the complexities and nuances of residential mobility patterns and processes. Although this paper has been primarily historical in focus, it highlights that recognising the diversity of migration history types both within and across cohorts could usefully inform the way that future studies conceptualise mobility.