5.5  Not participating (non-labour force participants)

5.5.1  Methods

For some people ill health may be the reason they do not participate in the labour force. Other studies have employed various methods to identify the group that may work in the absence of ill health. One example involved taking all those who do not work because of a health reason, or those who are not working and have a disability or chronic disease, as those who would potentially work in the absence of ill health (Davis et al., 2005). The analysis from SoFIE could have followed a similar approach, however, as well as the information on reason for not working not being available for all respondents; the impact of the health condition on preventing participation may be overestimated if other differences are not controlled for. In other words, even in the absence of the health condition not all of these people will participate owing to other factors. Therefore any estimate using this method would likely be an overestimate, or upper bound, of the impact. A modelling approach as used in Holt (2010), which identified a relationship between health and labour force participation, goes somewhere to control for these differences. This work provided estimates of the additional number of people who may participate in the absence of ill health. The standard binomial logistic regression model used in that report forms the basis for estimating the number of additional labour force participants and therefore the lost hours as a result of these people not working. As in the case of the analysis of labour force participants working fewer hours given in the previous section, this labour force participation analysis assumes that, in the absence of ill health, there would be demand for these hours to be worked and that the increase in the labour supply would not have impacted negatively on wages.

A binomial logistic regression model quantifies the relationship between self-rated health and labour force participation, while holding all other variables constant. The other control variables can be found in the full tables of results in Appendix G. A binomial logistic regression model is suitable as the dependent variable (L) is a binary response variable equal to one for those respondents who are participating and zero for those who were not participating (the latter was the reference category). The form of the equation can be seen in Figure 6. Maximum likelihood estimation was used to estimate the regression coefficients.[51]

Unlike the linear regression model that estimates the actual outcome, the binomial regression model estimates the probability of the outcome (that is, the probability of participating in the labour force). The probability of participating in the labour force (when all other variables are held constant at the mean of the sample) can be calculated for those in each health state. The probability of participating in the labour force can also be calculated for excellent health. The difference between the probability for each health state and that for excellent health is the marginal effect of that health state. Considering poor health, the probability of participating when in poor health minus the probability of participating when in excellent health, gives the marginal effect of poor health. If significant, the marginal effect for each health state is then applied to the number of people in that health state. This gives the additional number who may participate in this health state if they had in fact been in excellent health. The sum across all the significant health states is an estimate of the number who may participate in the absence of ill health (ie, if everyone had excellent health). To evaluate this in terms of lost hours it is assumed that each person would work eight hours a day for 260 days a year. These lost hours are then quantified in monetary terms using the average full-time hourly wage. The formula can be found in Appendix D.[52]

The model estimates the probability of participating for each person. The differences in probability can be used to estimate the additional number of people who may participate; however, this is done at a group level. It is not possible to know exactly which people may move into labour force participation in the absence of ill health. As such the wider characteristics of those who may participate in the absence of ill health (for example, highest qualification level or the presence of hospital inpatient appointments) is not known.

For similar reasons to those outlined in Section 5.4.1, this analysis will focus on basic logistic regression models rather than using panel models. It should be remembered that the results are therefore likely to form an upper bound of the cost. Again, future analysis could incorporate panel models into the analysis to try to better understand the costs associated with ill health.[53]

Those who are unemployed for the whole of the reference period are classified here as not participating. This is a very small group (around 0.4% of the population and 3.6% of those who are classified as not participating). If these people had been classified as participating then work hours lost would have been estimated in the absenteeism, presenteeism and working fewer hours sections despite them not undertaking any paid work. This section estimates the chance that those who are not participating would participate in the absence of ill health. For those who are continuously unemployed it is really estimating the chance a person would get paid work in the absence of ill health; implying that ill health may be one factor why they are continuously unemployed. The assumption is that, to some degree, health reduces the chance of the long-term unemployed obtaining paid work to the same extent as those who are inactive.