Onset of disability and life satisfaction: evidence from the German Socio-Economic Panel

For more information on the GSOEP data, see, for example, Wagner et al. [27].

The most used multiple-item scales are the Positive and Negative Affect Schedule (PANAS) scales [30] and the Satisfaction with Life Scale [31].

For a comprehensive review on the measurement of subjective well-being, validity, reliability and methods see, for example, Diener [34], Myers [35], Loo [36], and Kahneman and Krueger [32].

The use of work limitation as a measure of disability is controversial. For example, Hale [46] points out that the results obtained from the work-limitation population in the Current Population Survey (CPS) are not representative of the fuller population of disabilities. However, many studies have shown that self-reported measures of work limitation are highly correlated with both objective assessments of health and clinical measures of disability (Bound and Burkhauser [47] for a review of these studies; Stapleton et al. [48]). In addition, the trends in the prevalence of disability and employment rates for people with work limitations are similar to those obtained with other surveys using different measures of disability. In any case, self-reported disability questions must be used with caution because the changes in employment status, environment, family structure or economic well-being may affect the individual’s answer to the work-limitation question (Chirikos and Nestel [49]; Kreider [50]).

In 2001, the work limitation question in the GSOEP was: Aside from minor illnesses, does your health prevent you from completing everyday tasks like work around the house, paid work, studies, etc.? (Yes/No) To what extent? (Not at all/Slightly/Greatly). This question is the closest one to that included in the CPS. In the CPS every working-age individual in the household is required to answer (in March of the survey year) the following question: “Do you have a health problem or disability which prevents work or which limits the kind or amount of work you can do?” (Yes/No). Those individuals who respond “Yes” are considered disabled.

The German Institute for Economic Research (DIW Berlin) provides an algorithm to impute the missing values of this variable in the year 1986, 1990 and 1993. For further information, see: http://​www.​diw.​de/​english.

Initially, to test the relationship between satisfaction with health and satisfaction with life, we have calculated the correlation coefficients by disability status. For disabled, the correlation coefficient is 0.316, whereas this coefficient is 0.415 for non-disabled. Although there is a positive correlation between both variables, this correlation is even higher for non-disabled as compared to disabled.

The number of individuals with more than one onset within the panel is very low [35 individuals out of the total number of individuals becoming disabled (562) in the estimation of the adaptation equation and 23 individuals out of the total individuals who are at risk of becoming disabled and finally enter into disability (496) in the anticipation equation]. To check the effect of these individuals with repeated onsets of disability on the estimation results shown in the next section, we have re-estimated the adaptation and anticipation equations but now excluding these individuals. The results are very similar to those shown in Table 1 and the main conclusions remain valid. These regressions are available upon request.

This includes those individuals who remain disabled throughout all of the years and are observed within the panel. In this case, we cannot identify any transition between disability and non-disability status.

This total number of males who experience the onset of disability within our panel (479) is greater than that observed in other longitudinal studies on disability. For example, 280 individuals (males + females) experienced disability onset in the work of Jenkins and Rigg [44].

In contrast, the model used by Oswald and Powdthavee [2] weights each previous year of disability (variable “Past disability from t − 1 to t − 3”) in the same way.

Clark et al. [4] include in Eq. 1 dummy variables from D_0it to D_5it.

Jenkins and Rigg [44] conclude that those individuals who become disabled, compared to those who do not became disabled, are older, have lower household income, are less likely to be employed and have lower educational levels.

We have replicated all calculations using balanced panels with different sizes (from t − 2 to t + 1, t + 2,… t + 6). Logically, the total number of males observed increases when we use shorter balanced panels (from 194 to 216 males). The results obtained from each balanced panel are very similar to those shown in Fig. 1. Similar figures are included in the work of Oswald and Powdthavee [2], but their balanced panel is shorter (from t − 2 to t + 2) and the total number of individuals (males + females) available is only 61.

We have to bear in mind that, due to definition of our disability measure, all individuals with a satisfaction with health of at most 2 are considered as disabled. For this reason, Appendix Table 2. shows zeros in the health satisfaction scores 0,1 and 2 for non-disabled individuals.

To check the validity of our disability measure, we carried out additional tests. Using the work-limitation question included in the GSOEP but not asked in all its waves (asked from 1988 to 1991 and in 1993 and 1994 but not at all from 2002 onwards), we have crossed-referenced health satisfaction and work-limitation measures for those years where this work-limitation question is available. The results are extremely similar to those obtained when we cross-reference health satisfaction and our disability measures. We also compared the evolution of the variables health satisfaction and life satisfaction before, during and after onset. Although health satisfaction drops as the individual becomes disabled, after onset health satisfaction increases again but the levels are lower than those reached before onset (around 1.8 points) In contrast, the levels of life satisfaction in Fig. 1 show a recovery to initial levels of life satisfaction observed before the onset of disability. To corroborate the different evolution of both variables, we have re-estimated our anticipation and adaptation equations but now using the level of health satisfaction of individual as a dependent variable (the remainder of independent variables remains the same).The results show no adaptation to disability if health satisfaction is used as the dependent variable. The anticipation equation shows greater drops in health satisfaction for all years before the onset. These additional tests contribute to supporting the work of Burkhauser and Schroeder [3] on the validity of the disability measure used in our study.

We confirmed that disaggregating the dummy variable D _7it (7 or more years) into D _7it, D _8it, D _9it and so on leads to the same results being obtained as those shown in Table 1 (i.e. all estimated coefficients of these disaggregated variables continue being less negative and not significant).

If we include more dummy variables such as, for example, becoming disabled 4–5 years hence (D _−5it) and 5–6 years hence (D _−6it) and re-estimate our anticipation Eq. 2, the coefficients of D _−5it and D _−6it are statistically the same as the coefficient of the dummy variable D _−4it at 5% level.

If we use the household income before taxes and government transfers (in logarithms), the coefficient on this variable is very similar to that obtained by Oswald and Powdthavee [2].