One last piece of evidence on the validity of the specific training hypothesis is given by measuring the gain of the fine degree of segmentation used in this paper to the unsegmented earnings function or to a function which combines all previous jobs into one segment, PREVIOUS. These two-segment earnings functions are shown in Table 7 for the pooled sample and for the mobility patterns. When the results are compared to the full segmentation in the pooled sample (Table 4) the simpler two-segment earnings function does not fare badly. The R in the simpler equation is .223, while the explanatory power of the full segmentation is only slightly higher, .233.
Within mobility patterns, however, there are significant differences between the simple segmentation shown in Table 7 and the full segmentation in Appendix Table A-l. For example, no significant differences in explanatory power can be detected in the equations for Pattern 2 (where the current job is the longest).
Earnings functions Using Two Segments* Dependent = Ln(RATE), All Samples
The R for the unsegmented equation is .208; it Increases to .232 with the two-job breakdown, and to .234 with the full segmentation. Thus the introduction of the current job, where most investment took place, is the factor behind the increase in explanatory power. In Pattern 4, the results are quite different.
The full segmentation gives a much better fit to the earnings profile in this nobility pattern: the R for the full segmentation Is .184, while the explanatory power of the simpler equation Is only .142, and that of the unsegmented function is .140. Thus the Increase In R comes when we segment previous experience.
This finding suggests that the more “homogeneous” previous experience, the better the fit of the simpler (two-job) segmentation. That Is, In Pattern 4 we are combining the longest job and a series of short jobs Into one segment of previous experience. The results discussed earlier Indicated that some Investment took place In the longest job, but little Investment took place In the other previous jobs. If we combine these jobs Into a single category of previous experience, we lose the Information given by the relationship between job duration and the rate of growth In wages. Therefore the results point out the Importance of the longest job (regardless of when It occurred) In the determination of earnings.