Contractual conditions, working conditions and their impact on health and well-being

A two-stage stratified systematic sampling procedure was used to do the initial selection of households for inclusion in the survey. The procedure was designed to give each address an approximately equal probability of selection.

For further details see Taylor et al. [47].

For further details about this procedure see Hernandez-Quevedo et al. [25].

As a self-reported subjective measure of health, SAH may be prone to the measurement error often referred to as ‘state-dependent reporting bias’ (Kerkhofs and Lindeboom, [27]), ‘scale of reference bias’ (Groot [22]) and ‘response category cut-point shift’ (Sadana et al.[44]; Murray et al.[36]). This sort of measurement error occurs if the mapping of ‘true health’ into SAH categories vary with respondent characteristics, that is, if sample subgroups adopt different cut point levels in a systematic way when reporting their SAH, although they have the same level of ‘true’ health. To attempt to surmount this problem researchers could model the reporting bias making reference to more ‘objective’ indicators of true health (Kerkhofs and Lindeboom, [27]; Lindeboom and van Doorslaer [32]) or to ‘anchoring vignettes’ (Murray et al. 36). In the study we split the sample by gender before estimating the empirical models to take into consideration the differences in reporting behaviour across sexes shown in the literature (ie. Bago d’Uva et al. [5]). We do not pursue the potential issue of reporting error further.

To define if workers have a permanent or temporary job, they are asked “Leaving aside your own personal intentions and circumstances, is your current job permanent or non-permanent?”.

From waves 2–4 individuals were not asked information about some variables (not daytime, unions, payrise and promotion opportunities) if they were still in the same job as the previous year. For these cases, we assume that the value of these variables did not change from the last year it was recorded.

“We created a continuous variable by taking the midpoint of each category for each individual. For those who could not report the category into which their establishment fell, but were able to report whether it was above or below a particular value, we estimated their observation as a weighted average of the midpoints of the relevant categories. The weights used are the proportions of the relevant sub-sample which are in the relevant categories” (Contoyannis and Rice [15], p. 610).

This has been stressed, for example, by Bartley et al. [7].

Wooldridge’s [51] approach for dealing with linear panel models is also used by Hauck and Rice [24].

Note that the correlation between the individual effect α _i and h _it−1 (correlation that is implicit in the dynamic panel models) is broken by the fact that u _i is assumed to be independent of h _i1 , and therefore also h _it−1 .

Notice that the direction of the effect of the covariates on the probabilities of reporting the extreme outcomes (“poor or very poor” and “excellent” health, in our study) is unambiguously determined by the sign of the coefficients (Wooldridge [50]).

We attribute the mean value to the covariates that are continuous and the modal value to the covariates that are categorical. To make the partial effects meaningful, when we compute the partial effects of part-time*pref less hrs, part-time*pref more hrs and part-time*children the representative agent is assumed to have a part-time job, while computing the partial effects for temp job*high education and temp job*children this individual is assumed to have a temporary job.

The standard errors of the interactions are computed by applying the delta method (Norton et al. [39]).

For people satisfied with the number of hours worked or without children, a shift from full time to part-time increases the probability of reporting excellent health by 2.7% for women and 7.4% for men. If the person with the part-time job wishes to work less hours, the probability of reporting excellent health reduces by about 6% for both women and men, while if she/he wishes to work more hours the probability reduces by 1.9 and 4.8% for women and men, respectively. If the part-time workers have children, the probability reduces by 0.7 and 2.9% for females and males, respectively.

In fact, the partial effect for the probability of reporting excellent health for temporary workers with a high level of education is 0.048 for females and 0.029 for males, while the partial effects for temporary workers with children are 0.023 and −0.088.

The variation in the probability of reporting excellent health induced by any of these working conditions is smaller than 2%.

The value reported in the table is about 0.00005 for females and males, meaning that, for example, the presence of 100 more people at the workplace increases the probability of reporting excellent health by 0.5% for men.