2.1 Introduction

If we look at a cross-sectional "snapshot" of people who report that they currently have a chronic illness or disability, this will include people with very different experiences of disability. Some will have had a chronic illness or disability from birth, some will have developed a disability or chronic illness recently which will in all likelihood affect them long-term, and others will only experience that disability for a short period. We would expect the impact of these different patterns of disability on social inclusion to be very different. This is in essence why it is important to complement a cross-sectional picture with the analysis of the dynamics of disability over time and how it affects crucial aspects of societal participation.

In this study we will be using the longitudinal data obtained in the Living in Ireland survey from 1995 to 2001, and the chapter starts with a description of this panel dataset, which is unique in an Irish context. We then outline the analytic framework adopted in using these data to look at the impact of disability on key aspects of social inclusion on which the study is focused. This involves distinguishing those observed as experiencing an "onset" of disability during the period covered by the survey - and thus acquire that disability during adulthood - those experiencing an "exit" from disability, and those experiencing chronic illness and disability over the entire seven-year period of the panel survey. The chapter also describes the numbers observed in each of these groups and their characteristics, as background to the analysis presented in the rest of the study.

2.2 The Living in Ireland Surveys

The Living in Ireland Survey was conducted by the Economic and Social Research Institute (ESRI) each year from 1994 to 2001, as part of the European Community Household Panel organised by Eurostat, the Statistical Office of the European Communities. The first wave of the survey in 1994 obtained information for 4,048 households, where possible interviewing each adult in the household. (More detail is given in Gannon and Nolan, (2004) and a full description is in Callan et al, (1996).) What is critical from the point of view of the present study is that the initial sample was then followed in subsequent years for re-interview. All individuals in the initial survey were to be followed and household and individual interviews were to be conducted as long as the person still lived in a private household. (New households were included where a person already in the sample moved to another household.) The fact that the same people were interviewed each year, insofar as possible, means that it is possible to study changes in the characteristics and circumstances of those individuals and their households over time, the unique feature of longitudinal or panel data. The European Community Panel Survey ran up until 2001, when it was discontinued, and so the wave conducted in 2001 was also the eighth and last wave of the Living in Ireland survey.

By that point there had been substantial "attrition", in other words a substantial number of those included in the initial survey had dropped out over time, as is common in panel surveys. In the case of the Living in Ireland Survey, slightly more than half the individuals interviewed in the original survey were no longer in the panel by 2000. At that point the Irish panel was supplemented by 1,500 new households, also included in the base for the final survey in 2001, which helped in ensuring a more comprehensive and accurate picture of those years, but the analysis of dynamics has to rely on those who were in the panel for a longer period. In this study we focus on what is known as the "balanced panel" of individuals present throughout the life of the survey.

With attrition on this scale, it was obviously important to see whether it was substantially affecting the composition of the sample in terms of age profile, urban-rural split, household composition and socio-economic status. The overall impact in those terms seems to have been relatively small, and weights were developed to counteract such effects where they were detected. A comprehensive description of the response pattern in the Living in Ireland Survey over the full eight years, the supplementation in 2000, the problem of attrition and the weights developed to deal with it are given in Whelan et al (2003). Gannon (2005) analysed attrition specifically in terms of those reporting chronic illness or disability and its impact on the estimated impact of disability on labour force participation, using the procedure suggested by Wooldridge, and found no evidence to suggest that the results were affected by such attrition.

The Living in Ireland Survey included several questions directly focused on illness or disability, as described in Gannon and Nolan (2005). Respondents were asked:

"Do you have any chronic physical or mental health problem, illness or disability?"

Those who said they did were then asked

• "What is the nature of this illness or disability?"
• "Since when have you had this illness or disability?"
• "Are you hampered in your daily activities by this physical or mental health problem, illness or disability?" with response categories "severely", "to some extent", or "not at all".

The question posed about illness or disability in the initial, 1994 survey was slightly different so we do not use that year's results, basing our analysis on the seven years from 1995 to 2001. This leaves us with a total of 2,727 adults who responded fully in each wave.

In studying the dynamics of disability one has to be aware that, in addition to panel attrition in general, there may be a specific concern about attrition due to chronic illness or disability. In other words, those who already have a chronic illness or disability may be more likely than others to drop out of the panel survey over time, and "onset" of such a disability may itself be associated with an increased likelihood of drop-out. We can look at the panel data to see whether more of the people reporting a chronic illness or disability in one wave have dropped out by the next one, and evidence from the European Community Household Panel suggests that there does seem to be some such effect across the participating countries. We cannot see whether, in addition, some others who dropped out experienced onset of disability after they were last observed in the panel and this contributed to their dropping out. The implications for dynamic analysis of those remaining in the panel are difficult to assess, since we do not know how similar or different those who dropped out are, but the fact that those who are available for such analysis might not be entirely representative needs to be kept in mind.

2.3 The Analytical Framework

We now outline the analytical framework we will employ in assessing the impact of disability on key outcomes of interest, the rationale for this approach and what it involves. This framework is used to address the core questions towards which the study is directed: what impact does disability have on an individual's labour force participation, his or her household income, the likelihood that this income will fall below conventional income poverty thresholds, and the extent of their social networks and participation?

As emphasised in our introductory chapter, it is particularly useful to be able to look at observed movements into chronic illness or disability - which following Jenkins and Rigg (2003) we label as 'onset' - because we can then see what happens to the outcomes we are interested in, such as labour force status and income, as disability occurs. In doing so it is helpful to first identify all those individuals who could potentially experience disability onset at a particular point, and see how many actually experience onset. We can then compare the outcomes of interest for those experiencing onset with those who could have but did not experience it. This comparison gives us the most direct measure of the impact of disability onset, via a "control group" who do not experience onset but are similar in other respects.^[1]

As well as analysing what happens when someone starts a period of disability, it is also valuable to be able to capture the extent to which labour force status, income etc. change when a disability spell ends. Adopting the same approach as with onset, we can compare those observed in the panel survey to "exit" disability with those who could have exited but did not do so. To be in a position to potentially exit, one clearly has to be reporting a chronic illness or disability; taking all those who could have exited, the comparison is between those who do and do not experience exit.

As brought out in our introductory chapter, looking at disability onset and exit, while very informative, will not capture the impact of how long a disability lasts. To explore the impact of persistent, long-term disability we will complement our analysis of disability onset and exit by also examining those who report a chronic illness or disability throughout the period covered by the panel survey, in each year from 1995 to 2001.

So the analytical framework to be adopted in subsequent chapters involves looking in turn at the impact of disability onset, disability exit, and long-term or persistent disability on the outcomes of interest in the study - employment, income, poverty and social participation. Before going on to that analysis, we look in the rest of this chapter at those actually observed in the panel survey to experience onset, exit and persistent disability, on whom we will be focusing in the remainder of the study.

2.4 Disability Onset

We now discuss how we define and measure disability onset, and the types of individual observed in the sample as experiencing it. Since only adults were interviewed, only onset after 16 will be included - disability acquired in childhood might have a rather different impact, since educational attainment might well be affected as already noted. One way of capturing onset would be to simply take all adults not reporting disability in any of the surveys from 1995 to 2000 as "at risk", and take every case where we subsequently observe a spell of disability starting as "onset". (The "at risk" population could only be measured up to 2000, since 2001 was the last survey and we could not observe subsequent onset). That would mean that a particular individual could count as being at risk up to a maximum of six times. An individual could also experience onset more than once - in fact, with seven year's data an individual could report up to three distinct "onsets".^[2]

However, Jenkins and Rigg's British study take a more restrictive definition that concentrates on more sustained spells of disability following longer periods of "no disability". They define "at risk of onset" as not having a chronic illness or disability for at least two years, and "onset" as then having a chronic illness or disability which lasts for at least two years. With that more restrictive definition one needs a window of four years to capture onset of chronic illness or disability - for example, individuals who had no chronic illness or disability in 1995 and 1996 are at risk, and if they become chronically ill or disabled in 1997 and remain so until at least 1998 they are defined as having an onset of chronic illness/disability. Our sample only covers 7 years, so by definition an individual can then only have one onset of chronic illness/disability. This "tighter" measure of disability onset is the one we adopt in the present study, because it allows us to focus on the impact of more sustained disability and reduces the possible impact of measurement error in reporting disability.^[3]

In Table 2.1 we look at the numbers observed as at risk and the frequency of "onset". A total of 2,727 adults were followed throughout the period from 1995 to 2001, so we have (2,727 x 7) = 19,089 observations in all. Out of these observations, the respondent reported having a chronic illness or disability in 3,948 cases or 21% of the total - so that is the average cross-sectional disability rate over the period. However, not all the observations will be included as "at risk of onset" - because some people reported disability throughout and thus were never "at risk" in that sense, for example. We see from the table that 2,331 persons were "at risk of onset" over a total of 8,169 observations when we use a two-year definition of "at risk". A total of 235 individuals are then observed to experience onset, in other words start reporting the presence of a chronic illness or disability and do so for at least two years in a row.

What sort of people are most likely to experience onset? We can first look at some characteristics and see the percentages reporting onset, in Table 2.2. This shows that onset of chronic illness/disability is more likely to be reported by older people and by those with lower levels of education qualifications. The probability of onset rises sharply for those over 55 years of age and is much higher for those with no educational qualifications beyond primary level than for those with higher attainment levels - which is in itself associated with age, since older people have lower levels of education on average than younger ones.

This brings out that simply looking at individual characteristics one by one can mask the true underlying relationships, so we now employ multivariate statistical analysis to more reliably estimate the effect of each characteristic holding the other ones constant. We estimate statistical models (logistic regression) that predict the likelihood that someone who was at risk of onset will then report a chronic illness or disability, depending on a set of characteristics that might affect that probability. The explanatory variables we use include age, gender, education, labour force status and household composition. For those with a disability onset we use values for the explanatory variables measured two years before onset (and thus not affected by it), while for those with no onset we use the values in the first year they are observed in the panel, i.e. 1995. The results from estimating this statistical model are presented in Table 2.3.

These results give the estimated probability of someone with the relevant characteristics experiencing onset of disability, relative to someone in the reference category - in this case taken as a single man aged 16-24, with post-primary or higher education, no children, living alone, and not in work. Only where the number has asterisks after it is it statistically significant, in other words the variation is not likely to be simply attributable to the random variation to be found in any sample survey. The first such number (or "coefficient") is 2.92 for the age category "45-54", which means that someone aged 45-54 is 2.92 times more likely to experience onset of disability compared to a man aged 16-24 with the characteristics described above, other things being equal. Someone aged 65 or over is eight times more likely to experience onset of a disability than a 16-24 year old man with the reference characteristics. ^[4]

** Statistically significant at 5% level * Statistically significant at 10% level

Reference category: male, education qualification beyond primary, not working, single adult household, no children, aged 16-24.

Apart from age, the other socio-demographic characteristics are not statistically significant in distinguishing those who experience onset - the small number of cases observed to do so reduces the likelihood of detecting such effects, but it does suggest that the apparent greater likelihood of onset for those with lower education in Table 2.5 was indeed associated with their older age profile.

The extensive research literature on health inequalities suggests that people already in poverty may be more susceptible to chronic illness/disability onset (Burchardt, 2003). When income poverty status before onset - measured by whether the individual was in a household falling below 60% of median income (see Chapter 4) - is included in the model it is also significant, with individuals living in such households 1.7 times more likely to experience onset of chronic illness or disability than others.

These results provide useful background to the analysis in subsequent chapters of what happens as onset occurs. We also want to look at what happens to employment, income etc. when a disability spell ends, and so we now turn to how that is captured in the survey and how many people are observed as "exiting" disability.

2.5 Disability "Exit"

As with onset, we want to compare those observed in the panel survey to "exit" disability with those "at risk" of exiting who did not do so. We define "at risk of exit" as having a chronic illness or disability for at least two years consecutively, and "exit" as following such a period of disability by at least two years not reporting a chronic illness or disability. Once again this is a more restrictive definition than could be used, since we could simply treat anyone observed as having a disability in one year and not having it the next as "exiting", but for the same reasons as discussed in the context of onset we concentrate on this tighter measure of exit. As in the case of onset, it means that any individual can only be observed "exiting" disability once over the seven years of survey data.

We see from Table 2.4 that 138 exits are observed, out of a total of 1,251 "opportunities to exit" - occasions when someone in the sample had reported disability in the previous two years and was then observed for two more years.

Table 2.5 looks in a descriptive fashion at the characteristics of those observed as exiting disability, and shows the opposite pattern to that seen for those experiencing onset: the percentage exiting (as a proportion of all those at risk) seems to fall with age and with level of educational attainment. It also suggests women were more likely to exit than men.

However, once again the regression analysis which estimates the effects of specific characteristics when others are held constant, presented in Table 2.6, does not find significant education effects, suggesting those in the simple cross-tabulation mostly reflected age. The results do suggest that those aged 35-44, women, those who are in work prior to exit, and those with two or more children are more likely to exit disability. (Poverty status was also tested in the statistical model, but unlike the results for disability onset did not prove significant in predicting exit.)

** Statistically significant at 5% level * Statistically significant at 10% level.

Reference category: male, education qualification beyond primary, not working, single adult household, no children, aged 16-24.

2.6 Persistent Disability

As brought out in our introductory chapter, while looking at disability onset and exit is very informative it will not capture the impact on income or participation of how long a disability lasts. In defining onset and exit we have focused on individuals spending at least two years reporting disability, but there is clearly still likely to be a marked difference between someone who experiences disability for only two or three years and someone affected for a substantially longer period. To explore the impact of persistent, long-term disability we will complement our analysis of disability onset and exit by also examining those who report a chronic illness or disability throughout the period covered by the panel survey, in each year from 1995 to 2001. Table 2.7 looks at the numbers involved, and at where they fit in terms of the overall extent of reported experience of disability in the survey using a framework developed by Burchardt (2000) to categorise different types of disability experience or "trajectories".

Table 2.7: Persistent Chronic Illness/Disability, 1995-2001

We see that 180 individuals in the sample reported disability in each year from 1995 to 2001, representing almost 7 per cent of all respondents. At the other extreme, just under six out of ten respondents in the sample did not report chronic illness or disability in any year between 1995 and 2001. This means that 41% of respondents had at least one year when they reported such an illness or disability;^[5] those reporting disability in all 7 years account for 16% of those who reported disability at some point over the panel survey. While they represent what is in some sense an extreme in terms of duration, they allow us to assess the impact of such persistent disability on the outcomes of interest here, namely employment, income and poverty, and social participation.

The fact that we observe these individuals reporting a disability in each of the surveys does not mean that their disability "spell" lasted only 7 years. By their nature panel surveys only observe people over a specific period or observation "window", so we will not know when some of the "spells" of disability observed started or ended - they are what is known as "censored" in the data. Someone reporting in the first survey that they have a chronic illness or disability could well have had it before the panel survey started, but we did not observe it beginning - a case of what is called "left censoring". Similarly for those reporting an illness or disability in the last survey we do not observe when this spell finishes - a case of "right censoring". Those who report chronic illness or disability in all seven years are both left and right censored by definition - we do not know how long their spell of disability has lasted when we first observed them, nor how much longer it will continue after we last observe them.^[6]

Looking in a descriptive fashion at the characteristics of those experiencing disability throughout the panel, we see from Table 2.8 that persistent disability is most frequent in the older age groups and the lowest educational attainment categories.

Table 2.8: Rates of Persistent Chronic Illness/Disability

When we analyse this more formally, the regression analysis presented in Table 2.9 suggests that those with low education and those aged 35 or over are indeed more likely than better educated and/or younger respondents to have experienced persistent disability - though surprisingly that probability is not highest at older ages. On the other hand, women, those who were in work when first observed in 1995, and those with two or more children have a reduced probability of experiencing persistent disability. (Once again we also tested poverty status in the statistical model but it was not significant in this case.)

Table 2.9: Logistic Regression Model of Persistent Chronic Illness/Disability

** Statistically significant at 5% level * Statistically significant at 10% level.

Reference category: male, education qualification beyond primary, not working, single adult household, no children, aged 16-24.

The final point to be made about the disability onset, exit and persistence measures we have described is they are based purely on whether the respondent says that they have a longstanding illness or disability. Our previous studies have shown the value of the additional information also obtained in the surveys about whether the person felt that their illness or disability hampered them in their daily activities severely, to some extent or not at all. In a dynamic perspective, this is clearly also likely to be very important: there may be a marked difference in the impact of onset of a chronic illness that hampers the person severely versus such a disability that does not hamper them at all. We incorporate this into our analysis of disability onset, and also to some extent in our analysis of persistent disability (although this is complicated by the fact that the reported degree to which the individual is hampered changes from year to year for some of the persistently disabled). However, the disability "exits" we observe are insufficiently spread over the severity categories to allow for meaningful analysis in that context.

2.7 Conclusions

In this chapter we have described the data on which the study relies, from the Living in Ireland surveys from 1995 to 2001, which is unique in an Irish context in following a large representative sample over time. We then outlined the analytic framework to be adopted in using these data to look at the impact of disability on key aspects of social inclusion. This involves distinguishing those observed as experiencing an "onset" of disability during this period, those experiencing an "exit" from disability, and those experiencing chronic illness and disability over the entire seven-year period of the panel survey. Subsequent chapters will be examining how employment, household income, poverty and social participation appear to be affected by these different disability experiences.

The chapter then looked at the numbers observed in the sample falling into each of these groups and their characteristics. The measure of disability "onset" to be used focuses on those who reported no chronic illness or disability for two years, followed by two years when they do report illness or disability. Their situation can be compared with those "at risk" of onset who did not actually experience it - in other words, those who reported two consecutive years without disability in the panel and were then observed for a further two years without disability. (Someone who reported disability throughout the panel is thus not "at risk" in this sense, because disability onset cannot be observed.) We observed 235 individuals in the panel experiencing "onset" defined in this way. Multivariate analysis of the characteristics associated with an increased risk of onset then showed that older people are more likely to become ill or disabled. Having been in a low-income household in the previous year was also associated with an increased probability of disability onset.

Disability "exit" was defined and measured in a similar way, as a situation where the person was observed in the panel reporting two consecutive years with a disability followed by two years without such a disability. They can be compared with those "at risk" of exit who do not experience it - those reporting disability for two years in a row but not then reporting two disability-free years. The number of cases observed with a disability "exit" defined in this way was 138. Among all those "at risk", women and those in work (prior to exit) seemed more likely than others to report exiting disability.

The final group to be studied is those reporting persistent illness/disability over the seven years of the panel survey. A total of 180 cases had that experience, accounting for 7% of all respondents and 16% of those who reported disability at any point over the life of the survey. Those experiencing persistent disability were seen to be disproportionately male, older and poorly educated.

We go on in the rest of the study to examine the relationship between these disability experiences and key socio-economic outcomes, starting in the next chapter with employment.