Flora Macleod and Paul Lambe School of Education and Lifelong Learning St Luke’s Campus Heavitree Road University of Exeter Exeter f.j.macleod@exeter.ac.uk p.j.lambe@exeter.ac.uk Time invariant and time varying influences on the likelihood of participation in post compulsory formal learning opportunities A paper presented as part of the symposium ‘Dimensions of Diversity in University Learning’ at the Society for Research into Higher Education’s Annual Conference 12-14 December, 2006, Brighton, UK: Beyond Boundaries New Horizons for Research into Higher Education This working paper was produced as part of the Learning Lives Project. Copyright lies with the authors. If you cite or quote, please be sensitive to the fact that this is work in progress. The Learning Lives: Learning, Identity and Agency project (see learninglives.org) is a collaboration between the Universities of Exeter, Brighton, Leeds and Stirling and is funded by the Economic and Social Research Council as part of their Teaching and Learning Research Programme (TLRP) see www.tlrp.org Time invariant and time varying influences on the likelihood of participation in post compulsory formal learning opportunities Flora Macleod and Paul Lambe, University of Exeter Abstract Policy makers place emphasis on enabling more people to take up post compulsory education and training opportunities. However, there is little longitudinal evidence on who actually participates over time. This paper takes a life course perspective and uses event history frameworks to explore the power of time invariant and time varying predictors of whether and when a sample of 1997 initial phase leavers in England (n=463) return to adult learning for the first time. Our data source, the British Household Panel Survey (BHPS), allowed us to track these leavers over a six-year period (1997-2003). We found that the occurrence and timing of the first adult education episode was influenced by the time invariant individual characteristics of gender, qualifications, occupational status, household per capita income and household tenure upon leaving initial full time education and the time varying role transitions into marriage/cohabitation, unemployment and parenthood. The paper concludes by outlining how next phase of analysis is exploring the power of these early influences and the timing of role transitions for an individual’s learning trajectory. Key words: time-invariant predictors, time-varying predictors, event history analysis, panel data, gender differences, early adulthood, income status Introduction Returning to adult education or training is an event that will eventually occur for more than two thirds of us in the UK. The reasons why most of us return are various and have, from an adult education policy perspective at least, changed over time. Traditionally adult educationalists, notably Friere (1970), emphasised grander outcomes for adult education such as improving the lot of mankind through growth, development, empowerment and emancipation than the more mundane instrumental individual outcomes more prominent in current policy documents. These include getting a better job or being able to do our current job better. Education is no longer seen by policymakers as a phase completed in the early stages of life. Increasingly in the developed world participation in post initial education is seen by those in power not just as an option but a duty for which individuals must take responsibility as their employability and financial security 2 depend on the continual acquisition of new skills and knowledge in relation to the world of work (e.g. OECD, 2003). This shift has been recognised by economists who, in the past, were more interested in studying the economic rates of return from qualifications obtained in initial education (e.g. Bundell, et al 1999) are now beginning to study the economic benefits of education undertaken by adults (e.g. Jenkins, 2006). So it seems that, on the one hand, policy makers are pushing for continual skill (re) formation as a necessary response to a rapidly changing society and an ever-expanding skill need within the labour market and, on the other, there is a growing body of evidence that appears to suggest that participation in adult education leads to improved employment prospects and well being. Yet, in spite of all the attention being given to post compulsory education and training by politicians, economists and educationalists, the area remains undertheorized and under-explored using reliable longitudinal quantitative data. To the best of our knowledge there has been no systematic analysis on when the initial transition into adult education takes place, for how many and for whom the transition never takes place at all and identification of the factors that constrain or release an individual to make the transition. This area of investigation is important not least because standard explanations tend to be that determinants of participation in adult education predate the first adult education episode (Gorard and Rees, 2002). That is, the factors that determine whether an individual participates early in adult education opportunities also determine their participation (or non-participation) patterns later in the life course. This paper is an attempt to fill this gap by focusing on a cohort of young people in England and following them through for a six year period. The purpose of the paper is to answer two questions: (1) how long does it take leavers from initial full time education to return to adult education? (2) Can we predict the likelihood of an early return from a selection of time invariant and time varying predictors? We consider both time variant and time varying predictors important as decisions to participate (or not to participate) must be seen alongside the complexities of the social context in which individuals find themselves at a given point and place as well as their personal and social background characteristics. By including time-varying predictors we can study how social role changes over time shape the likelihood of an early return to adult education. Theoretical framework Time invariant predictors are relatively straightforward to identify as they describe those stable characteristics of an individual like their gender or some other static status such as qualifications at a given point which remain immutable over time. The ones we use here are gender and various background characteristics such as 3 qualifications at the point of leaving and occupational status, household per capita income and household tenure during the first year from leaving. Time varying predictors are more complex as they are those values that may differ over time and are frequently subject to natural change such as moving from school to work. To help us identify these we take a life course perspective. Specifically we draw on Elder’s (1985) conceptualization of the life course as a series of interlocking trajectories of social roles over time. Movement through a social institution normally involves taking on an institutionally defined role such as being a student, an employee, a husband or wife, a mother or father with each role configuration giving an indication of the extent to which a particular individual is embedded in a given social institution over time. Trajectories are seen as longitudinal involvement in or connection to social institutions such as work, marriage and parenthood. Entry into (and exit from) these often ordered and age structured social institutions is normally characterized by an event such as getting married, getting divorced. The specific event that moves an individual into or out of a life course institutional context is referred to as a transition. Transitions indicate when a particular trajectory of a given social role begins, ends, and lasts. Trajectories and transitions are thus useful conceptual tools when it comes to identifying the value of a time-varying predictor at a given point as the life course unfolds. The ones we are interested in here are movement from paid employment to labour market inactivity, entry into marriage or cohabitation and procreation. Our intention here is to study how experiencing these role transitions impacts upon decisions to participate or not in adult education. Methodology and methods Data and sample Our source of data is the British Household Panel Survey (BHPS) waves 8 to 13 (1997-2003). The BHPS uses a representative sample (5000+) of UK household resulting in 10,000+ individual interviews. Our sample was made up of all BHPS respondents living in England who had left initial full time education at some point between wave 7 (1997) and wave 8 (1998). 463 individuals aged between 16 and 23 (at wave 8) in England who at the BHPS 1998 interview (wave 8) were recorded as not being in full time education but who were recorded as being in full time education at the 1997 (wave 7) BHPS interview were identified as eligible to enter our sample. Because the BHPS interview takes place between September and November each year, these 463 individuals can be said to have left initial full time education at some point between September/November 1997 and September/November 1998. Event history analysis 4 How long it takes a leaver to become a returner might seem a straightforward enough question. Yet it is a question that is fraught with methodological difficulty. Event occurrence represents an individual’s transitions from one “state” to another. In this paper state one is represented as “being a non-returner” and state two is represented as “being a returner”. The methodology does not permit definition overlap. So, in order to track the occurrence of an event and its timing, the state of being a non-returner and being a returner must be mutually exclusive such that each member of our sample can only occupy one of the two states at a given point in time. As our interest here is in whether a target event occurs at all during the period for which data are available and, if it does, when it occurs during that period, we must have a clear definition of our target event that will allow us to make a clear distinction between each state. This condition must be met so that we can pinpoint the timing of the transition if or when it occurs. Our target event The point of transition of central interest is moving from being a leaver to “being a returner”. As pinpointing this transition is primarily a measurement issue precision and clarity is essential. To achieve this we had to settled for a definition of ‘being a returner’ that would unambiguously indicate what constituted one state and what constituted another. We operationalised our target event as “return to adult learning for the first time” by using one BHPS item requiring a “yes” or “no” response which was asked at all waves since 1998. That question was: (Apart from the full-time education you have already told me about) Have you taken part in any other training schemes or courses at all since September 1st [the previous year] or completed a course of training which led to a qualification? Please include part-time college or university courses, evening classes, training provided by an employer either on or off the job, government training schemes, Open University courses, correspondence courses and work experience schemes. (interviewees were instructed not to include leisure courses) In settling for this definition we fully acknowledge its limitations in that it represents only a very particular subset of all lifelong learning. However, we judged that this question represented a meaningful basis for our analysis because it tapped into a wide range of adult learning provision leading potentially to the full range of qualifications available to adults studying in England from the lowest to the highest. Measuring the passage of time All members of our sample were asked the above question in1998, 1999, 2000, 2001, 2002 and 2003. Our interest was purely in identifying the first time they 5 answered “yes”, or more precisely, the first BHPS wave at which they answered “yes”. For example at wave 8 (1998), our sample of leavers who had left school at some point between this wave and the previous one were asked “……..Have you taken part in any other training schemes or courses at all since September 1st 1997 …….” Thus they were being asked to reflect back on the year in which they had left initial full time education and say whether they had had any spells of adult learning formally organised on a part-time basis since they had left. Those who answer “yes” leave our sample at this point because they have experienced our target event and are thus no longer eligible to experience it for the first time again. Those who remain have their wave 9 (1999) responses examined. If they answer “yes” at wave 9 they leave, if “no” they remain in the sample and so on until we reach a wave where they answer “yes”. If they had not answered “yes” by the 2003 BHPS interview they were ‘right-censored’ meaning they left at the end of the period for which information is available having yet to experience the target event. The metric for recording the passage of time is BHPS waves, that is, one yearly discrete time intervals from September 1, 1997 or the point thereafter when they left initial full time education prior to the 1998 interview. This is because the “beginning of time” in this paper is the point at which sample members became eligible to experience the target event. According to how we defined our target event, it could not happen until after completion of full time initial education. There is thus no ‘left-censoring’ meaning everyone is in state one at the “beginning of time” because each individual in the sample is a non-returner until they become a returner. As we do not know precisely when between wave 7 (1997) and wave 8 (1998) respondents actually left full time initial education, this means that during year 1 (1st Sept 1997 – 1998 BHPS interview date) members of our sample could have had a longer or shorter period of eligibility to experience the event. For example, someone who left initial full time education on 1st October 1997 would have been eligible to experience the event from that date to the 1998 interview date, whilst someone who left initial full time education on 31st July 1998 would only have been eligible from that date to the 1998 interview date – the latter period being a substantially shorter period than the former and both being less than one full year. Also for this initial discrete time period the data do not allow us to differentiate between participation pre and post leaving, should the former have occurred. Models and statistical methods As the possibility of each sample member experiencing the target event, is in discrete 12 month intervals or periods of time (BHPS waves), a discrete event 6 history model is appropriate for our analysis. The application of regular statistical tools such as means and standard deviations is, however, inappropriate. This is because there are individuals for whom information on the occurrence and timing of the event is not available and excluding these cases would produce misleading results. To get round the problem of right censored cases (i.e. cases who leave at the end of the observation period without having experienced the event) three new statistical ways of summarising data are introduced: the hazard function, the survival function and the median lifetime (Singer and Willett, 2003). The terminology surrounding time duration methodology, such as survival and hazard functions, has mainly come from its two main areas of application, medicine and economics, where the first event is called the original event and the second is typically referred to as death or failure. Medical researchers are interested in how long patients survive after treatment. The main tool for describing event occurrence is the ‘life table’ which provides three key statistics: the median lifetime, the hazard function and the survival function. The median lifetime identifies the point in time at which half the sample members are estimated to have experienced the target event. The hazard function assesses the risk of the target event occurring among those eligible to experience it within each discrete time period. The survivor function assesses the probability that a given individual will survive from one discrete period to the next without having experienced the event. Findings For our methodology tabular and graphic displays are powerful ways of identifying and summarising trends over time. All 1997 Leavers The answer to our first research question is presented in table 1 which shows that the average time it took our cohort of 1997 leavers to become returners was just under three years. Average in a time table represents the median time, that is the time it took for 50% of the sample to experience the event. Table 1 also shows that at the end of the six year period of observation more than a quarter of the sample (27% n=124) had yet to return. 7 Table 1: Life Table Showing the Proportion of 1997 Initial Phase Leavers Returning to Learning between 1998 and 2003 Time interval 1998 1999 2000 2001 2002 2003 Respondents at risk of event 463 306 221 184 160 142 ↑RISK SET Event = 0 Event =1 Proportion Event =1 Proportion Event = 0 306 221 184 160 142 124 157 85 37 24 18 18 0.3391 0.2777 0.1674 0.1304 0.1125 0.1268 ↑HAZARD FUNCTION 0.6609 0.4773 0.3974 0.3456 0.3067 0.2678 ↑SURVIVOR FUNCTION This shows that the average time it took our cohort of leavers to become returners was just under three years. Average here represents the median time, that is the time it took for 50% of the sample to experience the event. Table 1 also shows that just under a third of our sample (31%, n=142) had yet to return. The hazard function for the whole sample is shown in figure 1a and the survivor function in figure 1b. Figure 1a is diagrammatic representation of the percentage of the sample at risk of event of returning happening within each time period (BHPS wave). The slope of the graph’s line shows that the risk of becoming a returner was at its highest in the time period immediately after leaving but by three years after leaving (2000) if a leaver had not yet become a returner their chances (risk) of returning was greatly reduced and reached in lowest point in 2002. In figure 1b, which shows the survivor function’s trajectory over time with the median lifetime plotted across the trajectory, shows the percentage within each time period that ‘survived’ without experiencing the event of returning with 27% surviving beyond the period of observation. 8 Figure 1a: Hazard Function: Lifetime: All 1997 Leavers (n=463) Figure 1b: Survivor Function and Median All Leavers (n=463) 0.35 Hazard Function Sample 0.30 0.25 0.20 0.15 0.10 1998 1999 2000 2001 2002 2003 Year Gender When controlled by gender, figures 2a and 2b, the shape of the hazard functions for female and male leavers were different. Although the probability of participation within the first year after leaving is almost equal for males and females, as the divergence of the hazard plot lines in figure 2a show, the probability for females returning diminishes and within two years of leaving full-time education the estimated probability of females becoming returners is almost a half of that of males (see also tables 2 and 3). The relative level of hazard differs significantly by gender, with the hazard function consistently higher for males than for females over five of the six years after leaving. This difference evidences that the conditional probability of becoming a returner is greater for the males in our sample than for the females, and that this differential persists over time. In Figure 2b the intersection of the median lifetime and the plot lines for males and females indicates that by the end of the second year after leaving half of the males and half of the females in the sample had returned, however thereafter the cumulative 9 probability of female returning is persistently lower than that of males. Gender is thus an important predictor in the probability of returning to adult learning. Figure 2a: Hazard Function Contrasted by Gender Median Lifetime (females = 223 females, males = 240) Figure 2b: Survivor Function and Contrasted by Gender, (females= 223, males =240). Table 2: The Probability of an Early Return to Adult Education among Female 1997 Initial Phase Leavers (n=223) Time interval 1998 1999 2000 2001 2002 2003 Respondents at risk of event 223 149 110 92 82 71 ↑RISK SET Event = 0 Event =1 Proportion Event =1 Proportion Event = 0 149 110 92 82 71 63 74 39 18 10 11 8 0.3318 0.1748 0.1636 0.1087 0.1341 0.1127 ↑HAZARD FUNCTION 0.6682 0.4933 0.4126 0.3677 0.3183 0.2825 ↑SURVIVOR FUNCTION 10 Table 3: The Probability of an Early Return to Adult Education among Male 1997 Initial Phase Leavers (n=240) Time interval 1998 1999 2000 2001 2002 2003 Respondents at risk of event 240 157 111 92 78 71 ↑RISK SET Event = 0 Event =1 Proportion Event =1 Proportion Event = 0 157 111 92 78 71 61 83 46 19 14 7 10 0.3458 0.2930 0.1711 0.1522 0.0897 0.1408 ↑HAZARD FUNCTION 0.6654 0.4625 0.3833 0.325 0.2958 0.2542 ↑SURVIVOR FUNCTION Our other explanatory variables Above we have showcased our analytic approach using the gender variable but space does not permit a similar tabular and graphic displays for our other explanatory variables even although tables and graphs are the most powerful ways of identifying and ummarising trends over time. So, in summary, further analysis controlling by qualifications on leaving, employment status at wave 8 (1998), parent’s (i.e. main earner’s) occupation at age 14, household per capita income at wave 8, household tenure at wave 8, urban/rural dwellers at wave 8, those who experienced a role transition into marriage/cohabitation during of observation contrasted by those who had not, those who experienced a role transition into parenthood contrasted by those who had not, those who had experienced a role transition from employment to unemployment contrasted by those who had not, those who had moved and most were found to be important predictors. Informed by these analysis we were able to specify statistical models to test the power of each of these predictors by quantifying their effects. We hypothesised that leavers will have different hazard functions if they have different values on observed predictors. As discrete time models use binary variables (1- event occurrence, 0=event non-occurrence) logistical regression can be used to fit models to the data. In this analysis the closer the logit coefficient, that is the odds 11 ratio of event occurrence, is to 1.0 the more likely it is that the event will happen with equal parity in the two groups being compared e.g. males/females. For example if an odds ratio is 1.0 then the odds are even for both groups. If the odds ratio is greater than 1.0 then the event is more likely to happen and less likely if the odds ratio is less than 0.1. The findings of this analysis are summarized below: The estimated odds of females returning to adult education in any one of the six discrete timeframes were 84% of those for males. The estimated odds of those who had left initial phase education with no or low qualifications were 51% of those who left with qualifications. The estimated odds of those whose status was recorded as unemployed in 1998 returning were 54% of those who were recorded as employed. The estimated odds of those whose household per capita income was recorded as ‘poor’ in 1998 returning were 74% of those recorded as ‘not poor’. The estimated odds of those whose household tenure in 1998 was recorded as non-owner/occupier returning was 70% of those who were recorded as owner/occupiers The estimated odds of those who experience a role transition into marriage/cohabitation during the period of observation returning was 66% of those who did not experience this role transition The estimated odds of those who experienced a role transition into parenthood returning was 37% of those who did not experience this role transition The estimated odds of those who experienced a role transition from paid employment to labour market inactivity returning were 50% of those who did not experience this role transition. Discussion These findings support the view that decisions to participate (or not to participate) in formal learning opportunities must be considered alongside the complexities of the social context in which adults find themselves at a given time and place as well as their personal and social background characteristics. They raise important questions about the link between temporal patterning found in early adulthood and subsequent life paths followed. Our analysis is ongoing and this paper represents only a sample of our emerging findings. The next step in our analysis, which is now well underway, is to explore the extent to which we can accurately predict an individual’s education trajectory on the basis of characteristics largely known by the time an individual leaves initial full time education and thus influence an early return to education. If we find this to be the case it does not imply that human 12 agency or life crises cannot impact on trajectories but rather that human agency and experience occur within a framework of opportunities, influences and social expectations that may or may not be determined independently. Indeed the next step in our analysis will contribute to an understanding of human agency and its interplay with society’s institutions in the shaping of lives over time. Trajectories, that is long term patterns of stability and change often involving many transitions, require a different methodology to event history frameworks. In order to explore more fully the ways in which an individual’s adult education trajectory is embedded in social relations more widely and over time we need to identify patterns of typical trajectories that effectively encapsulate the complexity of individual learning biographies. For this we will use latent class modeling (Macmillan and Eliason, 2003) which allows us to effectively model the life course as an institutional whole by conceptualizing it as probabilistically distributed paths through age-graded role configurations. References British Household Panel Survey (BHPS) User Documentation and Questionnaires www.iser.essex.ac.uk/ulse/bhps/doc/vola Bundell, R., Dearden, L., Meghir, C., and Sianesi, B. (1999) Human Capital Investment: The Returns from Education and Training to the Individual, the Firm and the Economy, Fiscal Studies, 20(1), pp 1-3. Elder, G. H. (1985) Life Course Dynamics: Trajectories and Transitions 19681980. Ithaca. New York: Cornell University Press. Friere, P. (1997) Pedagogy of the Oppressed. London: Penguin. Gorard, S. and Rees, G. (2002) Creating a Learning Society? Learning careers and policies for lifelong learning. Bristol: The Policy Press. Jenkins, A. (2006) Women, Lifelong Learning and Transitions into Employment. Work, Employment and Society, 20 (2), pp 309-328. Macmillan, R. and Eliason, S.R. (2004) Characterizing the life course as role configurations and pathways. J.T. Mortimer and M.j. Shanahan (Eds.) Handbook of the Life Course. New York: Springer, pp 529-554. OECD (2003) Beyond Rhetoric: Adult Learning Policies and Practices. Paris: Organisation for Economic Co-operation and Development (OECD). Singer, J., and Willett, J., (2003) Applied Longitudinal Data Analysis: Modelling Change and Event Occurrence. Oxford: Oxford University Press. 13