The Distribution of Human Capital in Sweden, ch. 6

6. Conclusion

A new method to obtain human capital estimates at the level of the individual, introduced by Abowd, Kramarz and Margolis (1999)has been applied to Swedish data for the first time in this paper. The human capital estimates are based on micro-level data from the Longitudinal Individual Database at Statistics Sweden between 1992 and 2005, covering roughly 300 000 individuals each year.

The method has captured the return to both observed and unobserved components of human capital by applying econometric techniques of estimation made accessible by the unique availability of panel data. More specifically, a fixed effects model has been applied to the data to explain variations in individuals’ wages over time, whereby the fixed effects were retrieved and interpreted as individually specific components of human capital. Consequently , it has enabled the relative importance of the different components to be analyzed.

The advantage of the specification is vast. Upon adding estimates of human capital to a baseline wage regression, the explanatory power of the model leaped from 0.5 to about 0.8. Furthermore , the unobservable component of time-invariant human capital was shown to be more important and more highly correlated with wages than the observable component.

Crude comparisons of human capital and productivity growth in different branches of industry mildly support previous findings that indicate that human capital is more important in explaining productivity in the knowledge intensive service sector than in the traditional manufacturing sector.

6.1 Assessment of the estimation strategy and future research

For all intents and purposes, the application of the estimation strategy has been successful. Though it is difficult to assess the extent to which all sources of variation that are not associated with human capital has been controlled for in our estimating model, there is nevertheless undoubtedly room for improvement. Specifically, the control variables industry, county and immigrant can be greatly enhanced by defining them in analytical units that are more suitable for the purpose of obtaining human capital estimates.

Starting with the industry variable, the definition of its levels can be optimized by choosing them in accordance with the wage-setting strategies of firms at different levels of industrial classification so as to capture as much of this heterogeneity as possible.

Similarly , the levels of the county variable can also be optimized (whereby a change of the variable’s name would be practical). But in this instance I believe there is a much easier solution, since SCB has already created Local Labour Markets (LA-regions) based on commuting statistics between municipalities. By merely adopting those regional definitions instead of using county borders, the quality of the control variable would likely improve.

Also, recall that the studies of labor market discrimination mentioned in section 4.1 (page 26) suggested that the effect of being an immigrant on labor market outcomes depends on which part of the world you are immigrating fromHence the immigrant variable should be expanded in levels to at least allow for different estimates of Asian, African and Middle-Eastern immigrants.

On a final note, despite all its apparent advantages, the estimation strategy in this study has an Achilles heel of sorts. It relies on the assumption that the fundamental compensation scheme on the labor market stems from the optimizing behavior of firms seeking labor and individuals seeking income as depicted in the classical labor economic literature. This implies that phenomena such as compensating wage differentials (Rosen, 1986), solidarity wage bargaining (Björklund et al. 2006) and tournament wage setting (Lazear, Rosen, 1981) will diffuse the interpretability of the estimates. This follows because the estimation strategy, as it stands, cannot directly distinguish between the variation in wages that is caused by these alternative compensating schemes and the variation that is caused by differences in productive capacities. Though some of the variation induced by alternative compensation schemes will be caught by the controls in the model, it is obviously difficult to determine precisely what effect they still have on our estimates. But however severe the problem might be, it ought to be alleviated by the use of percentile distributions when comparing levels of human capital, since the ranking poses no requirements on the accuracy of the distance between them.