Transitions and Transition Probabilities
Table 1 gives an overview of the number of individuals and the number of transitions by gender and occupational category. As males have a higher labor force attachment than females, there are more males than females in the CWLS. The structure of the sample with regard to occupational category is skewed in favor of unskilled and manual work. This reflects the structure of the Spanish labor market, which is dominated by labor-intensive industries with a high demand for unskilled work (construction, tourism, personal services; see Bentolila et al. 2012b).
Table 1 Number of individuals and transitions by gender and occupational category.
Figure 1 shows the transition probabilities for key transitions based on data of the years 2012 and 2013. The first panel shows the probability of becoming employed conditional on being inactive by age and gender, whereby only findings for the main working ages of 15–64 are shown. The second panel shows the probability of staying employed conditional on being employed, and the third panel shows the probability of retiring conditional on being employed, i.e., of leaving the labor market.
The probability of becoming employed conditional on being inactive increases relatively steadily starting at age 15 and peaks at age 25 for females and at age 29 for males. For example, the probability of becoming employed at age 29 reflects the transition from being inactive at age 29 in 2012 to being employed at age 30 in 2013. Females have a higher probability of becoming employed than males for most ages up to age 29, except for ages 15 and 16. This difference reflects the fact that females spend more time in education and enter the labor market later than males (Dolado et al. 2000). The lagged peak and lower level among males may arise if significant numbers of inactive males are unemployed but are not receiving unemployment benefits, as the eligibility criteria are strict and a minimum amount of contributions is needed to qualify (Venn 2012). A strong break occurs for both males and females at age 30. This discontinuity is due to the estimation of transition probabilities by subsamples. Estimating transition probabilities using only one sample instead of multiple subsamples could lead to potentially biased estimates, as this would restrict results to a strictly smooth schedule, whereas our approach loosens this smoothness requirement and allows for breaks. While the resulting piecewise smooth schedule overall is less restrictive, it might slightly exaggerate discontinuities in the schedule. Up to age 54 the transition probabilities do not change much and then decrease rapidly.
The probability of staying employed is lower for individuals up to age 30 (\(85\%\)) than for individuals up to age 54 (almost \(90\%\)). This gap is attributable to the fact that younger individuals are more likely than older workers to be on a fixed-term contract. This pattern might also explain the higher probability of staying employed found for males than for females (Azmat et al. 2006). Starting at age 60, the probability of staying employed decreases rapidly due to retirement. This trend is mirrored in the probability of retiring, which increases sharply starting at age 60.
Results on Working Life Expectancy for the Total Population
Before we break down WLE by occupational category, we present findings for the total male and the total female population, which give an overview of the general trends in Spain. These results are based on estimates of transition probabilities that do not include occupational category as an explanatory variable. Findings are shown in Table 2. Note that due to our methodological approach, each row of the table relates to 2 years t and \(t+1\); e.g., 2004 and 2005. As the recession hit Spain in 2008, 2007/2008 can be seen as a “mixed” period, which arises from using year-to-year transitions, while the years before this point can be viewed as a pre-crisis period, and the years after this point can be seen as a period marked by recession.
Table 2 Remaining life expectancy at age 15 (in years) spent in employment, unemployment, inactivity, and retirement, for Spanish males and females by period.
All of these numbers can be interpreted similar to total life expectancy and represent the average time spent in each state (measured in years) by members of a hypothetical cohort who experience the transition probabilities of a specific period. Because of the method we apply to correct mortality, the total life expectancy is similar to the life table estimates from the HMD, albeit with marginal differences: e.g., according to our findings, remaining life expectancy at age 15 for females in 2012/2013 is 70.33 years, while the corresponding value reported by the HMD for 2012 is 70.41.
Because of the recession, WLE has decreased considerably for both males and females. WLE was roughly 38 years for males and 33 years for females in 2006/2007, but had declined to 26 years for both males and females in 2008/2009. This implies that males lost nearly 12 years of WLE while females lost 7 years due to the recession. WLE varied before and during the crisis, but these year-by-year differences are small compared to the impact of the recession.
The years of WLE lost due to the recession are mostly spent in either unemployment or inactivity. Unlike the time spent in unemployment and inactivity, the time spent in retirement has not been greatly affected by the recession for either males or females, and recent reforms do not show clear effects, at least yet. Women spent four more years in retirement than men, mostly due to their longer life expectancy.
Working Life Expectancy by Occupational Category
Estimates of WLE (measured in years) by occupational category and gender are shown in Fig. 2. These results only include years spent in employment. Additional tables showing life expectancy in all labor force states by gender and occupational category are provided in the supplementary materials.
For males before the recession, both unskilled manual and non-manual workers had a WLE of roughly 38 years (2004/2005 value), while both skilled non-manual and manual individuals had a slightly lower WLE of 37 years. The impact of the Great Recession differed considerably by occupational category and reversed the gap between skilled and unskilled male workers. The changes in WLE between 2006/2007 and 2008/2009 amounted to 2.5 years for skilled non-manual work, 5.9 years for unskilled non-manual work, and close to 14 years for both skilled and unskilled manual work. For the most recent period (2012/2013), WLE is still below pre-recession levels for skilled manual, unskilled manual, and unskilled non-manual male workers, with values of 27 years, 27 years, and 34 years, respectively. For skilled non-manual workers, WLE is above the pre-recession level, with a value of 38 years.
The results for females roughly follow those of males, but mostly at a lower level. For instance, in 2004/2005 WLE was 37 years for male and 32 years for female skilled manual workers. Interestingly, the gender gap in WLE decreased considerably for manual work, while it increased slightly for non-manual work. The gender gap is calculated as the WLE of males minus the WLE of females. For example, the gender gap for unskilled manual workers was 5.6 years in 2006/2007 and 1.1 years in 2008/2009. For skilled manual workers, these values were 8.8 and 3.4, respectively. The recession thus had a greater impact on females than on males working in non-manual jobs, but a smaller impact on females than on males working in manual jobs.
Decomposition of Working Life Expectancy by Age Group
The WLE losses may be concentrated in specific age groups. For instance, we can speculate that because of high youth unemployment, most of the years lost are concentrated at younger ages. Figures 3 and 4 present a decomposition of the changes in WLE between 2006/2007 and 2008/2009 by age class for males and females, and for all occupational categories. The bars in each plot indicate how the differences in WLE are distributed among the following age classes: 19 or younger, 20–29, 30–39, 40–49, 50–59, and 60 or older. For instance, the total WLE of skilled manual male workers decreased by 13.5 years, and 2.7 of these years were lost in the age group 30–39. Detailed results are provided in the supplementary materials.
For neither males nor females, the age groups 19 or younger and 60 or older contributed substantially to WLE, and the losses are small. The same holds by occupational category. If we focus on the prime working age groups, we can see that for male skilled and unskilled manual workers, the losses are relatively evenly distributed among the age classes, although the losses are slightly larger in the age groups 20–29 and 50–59. For male skilled and unskilled non-manual workers, the losses are more concentrated in the younger age groups. Thus, for males the impact of youth unemployment on WLE seems to be limited to non-manual workers. Interestingly, for females the losses are more concentrated in the younger age groups for unskilled but not for skilled workers. For skilled non-manual female workers, the losses even show the opposite pattern and are concentrated at higher ages. The total loss of WLE is, however, small for this group.
Active Working Life Expectancy and Sullivan’s Method
Figure 5 compares the estimates of active working life expectancy (AWLE) based on Markov chains with the estimates obtained using Sullivan’s method; i.e., the Eurostat estimates and the estimates based on the CWLS, as described in Sect. 4.3. The Markov chain estimates are calculated by adding the columns “Employed” and “Unemployed” of Table 2. Note that for the estimates obtained from Markov chains, each data point in the figure relates to 2 years, but the axis labels only show the first year. For instance, the data point for 2012 actually relates to 2012 as t and to 2013 as \(t+1\). For the results based on Sullivan’s method, each data point relates to 1 year as usual.
For males, the differences between our estimates based on Markov chains and the Eurostat estimates are rather large. The biggest difference is found for 2008/2009, with the Eurostat estimates being nearly 6 years higher than our estimates. While our findings show a decline in male AWLE of roughly 12 years between 2006/2007 and 2008/2009, the Eurostat estimates actually show an increase of 0.3 years between 2006 and 2008, and the values are generally rather stable. As we discussed above, this discrepancy may be at least partly due to the fact that the Eurostat estimates are based on labor force participation rates, which also include individuals who are not receiving unemployment benefits but are looking for work, whereas the CWLS only allows us to cover individuals who are receiving unemployment benefits. The labor force participation rates calculated from the CWLS thus differ from the Eurostat estimates. AWLE calculated using CWLS data shows a decline for males, but the decrease is later and slower than in the Markov estimates; i.e., a decline of just 0.6 years between 2006 and 2008. Thus, this estimate differs from the estimate generated using the Markov chain approach, again by roughly 6 years.
For females, the results also differ strongly depending on whether the Markov approach or Sullivan’s method is used. Both our estimates and Eurostat’s estimates obtained using Sullivan’s method show an increase in WLE, at least from 2004 to 2008, while the Markov chain approach shows a decrease. After 2008, all of the estimates are roughly at the same level for at least some years, but the trends still differ: the estimates obtained using the Markov chain approach show a decline of roughly 1.6 years between 2008/2009 and 2012/2013, the estimates obtained using Sullivan’s method based on CWLS data show a decline of 1.1 years between 2008 and 2012, and the Eurostat estimates show an increase of 2.3 years. Overall, Sullivan’s method seems to miss both the levels and the trends of AWLE.