APPENDIX A: DIFFERENCES AMONG SAMPLES AND THE EFFECTS OF OTHER INDEPENDENT VARIABLES

Tables A-1 and A-2 present the coefficients of sub-baccalaureate credentials and the benefits of varying amounts of postsecondary education among noncompleters for three samples: (1) individuals 25 to 64, including those still in school at some point during the year (the sample used in the body of this monograph); (2) individuals age 18 to 64, with a relatively higher fraction of individuals still in school; and (3) individuals 18 to 64, excluding all individuals who reported being in school at any time during the year. The coefficients for sub-baccalaureate credentials are relatively stable across these three samples. However, for men, the coefficients of the variables describing some college vary substantially: they are consistently lower, and statistically insignificant, in the second sample, which includes a higher fraction of students. Evidently these students who are over the age of 18 are more likely to be enrolled in college and to not have completed credentials. They are also likely to have poorly paid jobs--what counselors sometimes call "stay-in-school jobs"--designed only to get them through postsecondary education. The lower earnings of these temporary jobs are partly reflected in the coefficient of the dummy variable for those still in school; but even with this controlled variable, the earnings differences associated with larger amounts of postsecondary education are small, erratic, and insignificant--precisely what one would expect if "stay in school" jobs are menial jobs which do not use the skills college students have accumulated. However, when the sample is restricted to nonstudents, or to individuals 25 and older who are more likely to be in more career-oriented jobs, then the returns to one or more years of college prove to be significant. The coefficients are relatively stable between these two samples.

For women, the divergent samples shown in Table A-2 make much less difference because the effects of some college are almost uniformly insignificant. These findings buttress the conclusion that small amounts of postsecondary education do little to improve the earnings of women.

Table A-1

Effects of Varying Samples on Schooling Coefficients
Males, 1987 Sample

Table A-2

Effects of Varying Samples on Schooling Coefficients
Males, 1987 Sample

The coefficients for the racial and ethnic variables suggest that discrimination is more serious for men--and especially for African American and Asian American men--than for women. In addition, among Hispanics, there appears to be discrimination against Mexican-Americans and Puerto Ricans--though small sample sizes mean that the coefficients for Puerto Rican men and women are not statistically significant--but not for Hispanics of Cuban and South American descent, confirming the findings of DeFreitas (1991).

The effects on earnings of family background in this data is generally erratic. Efforts to include the occupation of the head of household when the respondent was 15 yielded consistently insignificant coefficients; only the education of the head of household provided any explanatory power. For men, the results are roughly consistent with the hypothesis that family background continues to influence earnings directly, as well as indirectly, through its influence on education. The effects of the household head having only elementary education are negative and the effects of a college degree are positive and significant; the other coefficients have the appropriate signs though they are insignificant. For women, however, the coefficients on the education of the household head are consistently insignificant. These results are consistent with a pattern in which high-status families favor sons over daughters: daughters in families of higher socioeconomic status may benefit from having higher levels of education, but there is no further direct effect of family background on earnings.[46]

Most other independent variables have the signs expected of them, and most are significant. Individuals still in school tend to have lower earnings, for reasons explored in the previous section. Those covered by union contracts have higher earnings, especially

Table A-3

Effects of Other Independent Variables on Earnings
1987 Sample

women. Earnings are higher in metropolitan areas and lower in the South and North Central regions. Marriage works quite differently for men and women, increasing the earnings of men but decreasing the earnings of women; and women with children have lower earnings than those who do not. Finally, those who report a disability have substantially lower earnings than others.

These results distinguish among three different forms of experience: (1) tenure on the current job; (2) experience on prior jobs that are related to the current job; and (3) experience on prior jobs unrelated to the current job. Because the SIPP data collection leaves unavoidable gaps in the employment record, these specifications also include age (and age squared), in part to pick up the effects of any omitted periods of experience and in part to see whether there are effects of age independent of experience.[47] If employers value most the kinds of job-specific learning that takes place on the job, then the effects of current job tenure should be the highest of these three forms of experience. Furthermore, related experience in another job should have greater effects on earnings than unrelated prior experience. Alternatively, experience may be important, not because it provides on-the-job learning, but because it allows employers to perceive the real abilities of employees; in this kind of signaling model, the effect of current job tenure should still be highest while the effect of prior unrelated experience should be lowest.

Indeed, the results are consistent with this hypothesis for both men and women. While the quadratic form means that the effect of a year of schooling declines over time,[48] these effects are still dominated by the linear term; this declines monotonically across these three types of experience, for both men (from .077 to .039 to .017) and women (from .152 to .075 to .021). It is worth noting that the effects of experience are consistently larger for women than for men. Finally, age still has a statistically significant influence even after these three kinds of experience are included, though it is impossible to know whether this reflects simply the effects of age, independent of experience, or periods of experience unmeasured by the other three experience variables.[49]

The remaining coefficients in Table A-3 reflect the influences of dummy variables which are included whenever certain values are imputed for the other independent variables. Several of these coefficients are significant; several of them are highly so. In general, imputations are made for relatively small numbers of individuals--ranging from 0.5 to 3.7% among the 5,452 men and 0.2 to 4.4% among the 4,952 women--and so their inclusion or exclusion does not change parameter estimates by much. These imputations are included to reflect the presence of unavoidable error when such imputations are made. The exception is that total experience is imputed as age minus education minus 6 for 31% of men and 25% of women; the negative coefficients of the associated dummies indicate that the experience thereby imputed is too high. By and large, the imputation dummies affect only the coefficient of the specific variable to which they pertain--for example, the dummy reflecting missing data on union coverage affects only the coefficient describing whether an individuals was covered by a union contract--and so their influence on the education variables that are of principal concern are slight. Therefore there is little chance that the results in this monograph are distorted by the need to include imputed values for certain independent variables.

[47] Several experiments with other variables which are designed to reflect missing spells of experience--particularly the one often included in human capital formulations, age minus years of schooling minus 6--proved to be inferior, in terms of both explanatory power and effects on other coefficients, to the simple inclusion of age.

[48] The maximum effects of experience come at levels of experience ranging from 17 to 39 years for men and from 21 to 24 years for women--all sufficiently high enough that the effects of experience never decrease in these samples.

[49] The positive effect of age reaches a maximum at 38 for men and 29 for women and becomes zero and turns negative at 76 for men and 58 for women. Over the sample used--those 25 to 64--increasing age generally decreases earnings.