«IZA DP No. 3067 How General Is Human Capital? A Task-Based Approach Christina Gathmann Uta Schönberg September 2007 Forschungsinstitut zur Zukunft ...»
4.2 Wages in Current Occupation Depend on Distance of Move If skills (i.e. those accumulated in the labor market, T Tiot; as well as those workers bring to the labor A M market, o Tit +(1 o )Tit ) are partially transferable between occupations using similar skills, we would expect wages at the source occupation to be a better predictor for wages at a similar target occupation.
This implies that the correlation of wages before and after an occupational move should decline in the distance of a move. Table 6 investigates this hypothesis. It reports estimates from a wage regression where the dependent variable is the log daily wage. All speci…cations include experience and experience squared as well as year and occupation dummies. Results are reported separately by education. As a benchmark for comparison, the …rst speci…cation (column (1)) estimates the correlation of wages for occupational stayers. Wages in the same occupation are highly correlated over time with the correlation being strongest for university graduates.
In the next speci…cation, we restrict the sample to occupational movers who start out with zero occupational tenure (column (2)). For all education groups, the correlation of wages is lower among occupational movers than among occupational stayers. Speci…cation 3 analyzes whether the impact of wages in the source occupation on wages in the target occupation varies with the distance of the occupational move. We add the distance of the move as well as the distance interacted with the wage at the source occupation as additional regressors. Indeed, the predictive power of the wage at the source occupation is larger for movers to similar occupations. Interestingly, the di¤erence in the correlation is strongest for the high-skilled workers, i.e. the group that is also most likely to move to similar occupations. For this education group, our estimates imply that the impact of the wage at the source occupation on the wage at the target occupation is 0.35 for the most similar move, 0.31 for the median move (0:351 0:103 0:354), and approaches 0 for the most distant move (0:352 0:939 0:354).
As a second test of skill transferability, we estimate whether tenure in the previous occupation matters for wages in the new occupation. In Column (1) of Table 7, we regress wages at the new occupation on occupational tenure at the previous occupation and the same controls as in Table 6.
Past occupational tenure positively a¤ects wages at the new occupation. This result is consistent with previous evidence that post-displacement wages depend positively on tenure in the pre-displacement job (e.g. Kletzer, 1989). Column (2) adds the distance measure interacted with past occupational tenure as controls. As expected, the predictive power of past occupational tenure is stronger if source and target occupations are similar. In line with our previous results, the impact of past occupational tenure declines more sharply with distance for university graduates. For this education group, the impact of past occupational tenure is 2.3 percent for the most similar move, but only 1 percent (0:023 0:103 0:072 = 0:010) for the median move.
Figure 4a relaxes the assumption that the correlation between wages across occupations declines linearly with the distance. The x-axis shows the distance with one being the most similar occupational moves and 10 the most distant ones, while the y-axis reports the coe¢ cient on the wage in the source occupation for each of the 10 categories. The coe¢ cient is obtained form a OLS regression (tobit regression for the high-skilled) that controls for actual experience, actual experience squared, year dummies, the wage at the source occupation, 9 dummies for the distance of the move and the 9 dummies interacted with the wage at the source occupation (see column (3) in Table 6). Two things are noteworthy: …rst, the …gure highlights that wages at the source occupation have a stronger explanatory power for the wage at the target occupation if the source and the target occupation have similar skill requirements. Second and in line with our results on mobility and wages, the decline is strongest for the high-skilled. For this education group, the partial correlation coe¢ cient between wages in the source and target occupation drops from 37 percent for the 10 percent most similar moves to around 15 percent for the 10 percent most distant moves. The drop is statistically signi…cant at a one percent level for all education groups.
Figure 4b provides a similar analysis for past occupational tenure. The y-axis are now the coe¢ cients on the 9 distance measure dummies from a tobit regression that also controls for actual experience, actual experience squared and year dummies. The correlation between past occupational tenure and wages in the new occupations is declining roughly linearly with the distance of the move. As before, the declining pattern is strongest for the high-skilled, particularly for very distant occupational moves.
We have performed a number of robustness checks. First, results for alternative distance measures are very similar. Second, our sample of movers contains both occupational switches between …rms as well as within the same …rm. The latter account for roughly 10 percent of all occupational movers. If some skills are tied to a …rm, internal movers would have more portable skills than …rm switchers. We therefore reestimated our speci…cations in Table 5, 6 and 7 using only external movers. The results exhibit the same patterns in mobility and wages which we observe for the whole sample of movers.
Finally, our original sample of movers contains everybody switching occupations irrespective of the duration of intermediate un- or nonemployment spells. To the extent that those remaining out of employment for an extended period of time are di¤erent from for example job-to-job movers, our results might not be valid for those with high attachments to the labor market. To account for this, we reestimated the results only for the sample of workers with intermediate un- or nonemployment spells of less than a year. Again, this does not change the patterns on mobility and wages.
4.3 Can these Patterns be Explained by Unobserved Heterogeneity?
The strong patterns in mobility and wages reported in the last section support our view that human capital accumulated in the labor market is portable across occupations, and the more so the more similar are the occupations. This section discusses whether our …ndings could be rationalized by individual heterogeneity. Note …rst that all results presented above are based on a sample of occupational movers.
The patterns in mobility and wages can therefore not be accounted for by a simple mover-stayer model, where movers have a higher probability of leaving a job and therefore lower productivity because of less investment in speci…c skills. To the extent that movers di¤er from stayers in terms of observable and unobservable characteristics, this sample restriction reduces selection bias.
However, other sources of unobserved heterogeneity could bias our results. First, one might argue that the similar moves in the data are voluntary transitions, while distant moves occur because of layo¤s from the previous job. If wages at the source and target occupation are more strongly correlated after a quit than after a layo¤, then the distinction between quits and layo¤s can explain why wages are more highly correlated across similar occupations. It would also explain why past occupational tenure has a higher return in a similar occupation. However, the distinction between voluntary and involuntary movers does not explain why voluntary movers move to similar occupations in the …rst place. While we cannot directly distinguish between voluntary and involuntary job changes, we checked whether our results di¤er between job-to-job and job-to-unemployment transitions. Job-to-job changes are more likely to be voluntary, while moves into registered unemployment are more likely to be involuntary. We …nd that patterns in occupational mobility are similar for the two types of moves. This results makes it unlikely that voluntary and involuntary occupational moves are responsible for our …ndings.
Second, suppose that the sample of movers di¤ers in their taste for particular tasks. Some individuals prefer research over negotiating, while other prefer negotiating over managing personnel etc. Taste heterogeneity can explain why we see similar moves in the data. If individuals choose their occupations based on earnings and preferences for tasks, individuals would want to move to occupations with similar task requirements. However, a story based on taste heterogeneity alone cannot explain why wages are more strongly correlated between similar occupations. If there are compensating wage di¤erentials, we would actually expect the opposite: individuals would be willing to accept lower wages for a move to an occupation with their preferred task requirements.
Finally, suppose that high ability workers are less likely to switch occupations. This could account for the fact that the time spent in the last occupation has a positive e¤ect on wages in the current occupation, as past occupational tenure would act as a proxy for unobserved ability in the wage regression (see Table 7). However, unobserved ability per se cannot explain why the e¤ect of past occupational tenure should vary with the distance of the move or why individuals move to similar occupations at all.
This discussion highlights that a simple story of unobserved heterogeneity cannot account for all of the results presented above. Our theoretical framework however also makes clear that occupational movers are not a random sample of the population of workers. Individuals choose to switch occupations and they can also choose the distance of their occupational move. The next section outlines an estimation approach to quantify the importance of task-speci…c human capital for individual wage growth that accounts for the endogeneity of occupational mobility.
5 Task-Speci…c Human Capital and Individual Wage Growth To estimate the contribution of task-speci…c human capital to individual wage growth, we start from
ity in the return to the three types of human capital accumulation.
5.1 Least Squares Results We …rst present least squares estimates of the wage regressions of (3) as our benchmark. To account for censoring in the wages of university graduates, we estimate censored regressions for this group. Columns (1) to (4) in Table 8 reports the results for two di¤erent samples: the whole sample of …rm movers and stayers, and the sample of workers starting a new …rm. The …rst speci…cation (odd columns) displays results from a wage regression that ignores task-speci…c human capital, while even columns includes task tenure as an additional regressor. If human capital is indeed partially transferable across occupations, we expect that the inclusion of task tenure will lower the estimated return to both occupational tenure and experience. In addition to the variables reported, all regressions include occupation, region, and time dummies.
The results show several interesting patterns. Returns to task tenure are sizeable and exceed those of occupational tenure for all education groups. Based on the estimates for the whole sample, returns to occupational tenure decline by about 20 percent for the two lower educated groups, and about one-third for the high-skilled once we include task tenure. Returns to experience also decline by about 20 percent for the low- and medium-skilled and by about 30 percent for the high-skilled for a worker with ten years of (actual) labor market experience. The patterns are very similar for the sample of workers starting at a new …rm, i.e. workers with …rm tenure equal to zero, shown in columns (3) and (4). Compared to the full sample, the return to general experience is lower while the returns to occupation and task tenure are higher for all education groups.
is that individuals select into occupations based on the returns to their skills ( 1o ; 2o ; 3o ). We would generally expect the average return to experience, 1; to be upward biased. This is because with time in the labor market, workers …nd occupations that best use their task productivity through on-the-job
but also wage growth due to job search (see also Topel, 1991; Dustmann and Meghir, 2005).
In contrast, the return to occupation- and task-speci…c human capital, and 3; may be upward or downward biased. On the one hand, workers who are well matched are less likely to switch occupations or move to a distant occupation. This implies a positive (partial) correlation between occupation and task tenure and the match quality, and thus to an upward bias in the return to occupation- and taskspeci…c human capital. On the other hand, workers may have switched occupations or moved to a distant occupation because they are particularly well matched with the new occupation. Hence, workers with low levels of occupation and task tenure may have particularly high task matches, leading to a downward bias in the return to occupation- and task-speci…c human capital. However, workers may also move to a new and/or a distant occupation because of a higher return to human capital, which provides another reason why the (partial) correlation between occupation and task tenure and the error term in (3) cannot be signed.
To get an idea of the magnitude of these biases, we next estimate wage regression (3) for a sample of workers who were exogenously displaced from their job due to plant closure (see Gibbons and Katz, 1991; Neal, 1995 and Dustmann and Meghir, 2005 for a similar strategy).10 Displaced workers di¤er from voluntary …rm switchers because they are willing to accept a new job o¤er if its value exceeds the value of unemployment, as opposed to the value of the old job. Displaced workers lose some of their ‘search capital’ which should lower the upward bias in the return to experience. For the high-skilled,, the return to experience is indeed about 25 percent lower in the sample of displaced movers than in the sample of all …rm movers, while there is little di¤erence for the other two education groups (Table 8, columns (5) and (6)). This suggests that job search plays a more important role for the high-skilled.
Task tenure remains an important source of individual wage growth for all education groups.