• Defining School-Leaving Groups
• Sample Restrictions
• Return to School After School Leaving
• Alternative Approaches to Defining School-Leaving Groups

Most of the empirical work that follows uses the National Longitudinal Survey--Youth. This Department of Labor-sponsored panel survey began with 12,781 young people (in the civilian sample) aged 14 to 21 in 1979 (Center for Human Resource Research, 1988) . Blacks, Hispanics, and disadvantaged whites were oversampled. The sampled individuals were reinterviewed annually. This report includes data through 1990. Thus, this sample is now old enough (aged 25 to 32 in 1990) for us to examine nearly completed school-to-work transitions for a cohort of youth that entered the U.S. labor market in the 1980s.

For each year, the interview collected complete retrospective calendars of employment. Beginning in 1981, monthly schooling attendance records also were collected.[1] Using these data, we constructed monthly records of school attendance and work for each person in the sample for the period January 1, 1978, to the last completed interview date, usually mid-1990.

All the individual education and employment histories are censored (i.e., we do not know what happens at later ages), at least as of the 1990 interview, when the young people were aged 25 to 32. Furthermore, there is some sample attrition, so some of the data may be censored even earlier. To use all the collected data when computing time to events or percentage of people experiencing an event (by age or by time since an event), we computed monthly hazard rates.[2] We then transformed the hazard rates back into the percentage of people experiencing (or not experiencing) the event as of a given age or time since an earlier event (where the probability of the event not happening through age A is the product of the hazards of the event not happening at each age in months up to age A).[3]

Defining School-Leaving Groups

School-to-work transition patterns vary widely by schooling attainment when the individual leaves school. Not only does the age at school leaving vary with attained schooling, but (as we show below) the pace of settling into stable employment also varies. Following this empirical observation and the rest of the school-to-work transition literature, we stratified our analyses by schooling attainment at school leaving. However, the heterogeneity and complexity of transitions between school, work, and leisure make operationalizing the concept of school leaving difficult and make the results sensitive to the definition chosen.[4]

We assigned each sample member to a school-leaving group (SLG). Conceptually, a sample member has left school when his or her primary activity is no longer school. However, summer vacation should not be considered school leaving. In practice, we used the following algorithm. School is considered the primary activity for an individual only if he/she is attending school and is not working full time (FT; defined as 35 or more hours per week).[5] Given this definition of "full-time school attendance," we then filled in any gaps in school attendance that were probably due to school breaks (including the transition from high school to college). If the gap began in May, we filled in up to five months (i.e., May through September); if in June, we filled in four months. Gaps beginning in any other month were allowed to last up to three months without being considered "school leaving."

Once we determined that a gap in schooling constitutes school leaving, we set the date of school leaving at the first month of the gap and, for our analysis, assigned a permanent SLG to the sample member, based on school attendance and degree receipt through that date.[6] Even if the sample member later returned to school and/or attained a degree, the ascribed SLG is not changed.

Below we examine the importance of return to school and thus the difference between ascribed SLG and schooling attainment for an individual at a point in time. Given this algorithm for defining school leaving, we defined the hierarchy of five SLGs shown in Table 2.1, from the lowest SLG, high school dropout, to the highest SLG, those with some post-college education.

Sample Restrictions

Because the NLS-Y oversampled blacks, Hispanics, and poor whites, the results that follow are all weighted by the 1979 interview weight. That weight corrects for the oversampling and for differential nonresponse to the first interview. We made no further correction for subsequent interview nonresponse or permanent panel attrition.[7] All the results presented in this report stratify our samples by gender. The complexity introduced into female work histories by childbirth (whether as teenagers before they enter the labor market or as adults possibly interrupting careers) implies that the two-activity (school and work) analyses presented below fail to capture a crucial element of young women's work histories. For this reason, we looked at the patterns for women separately from those for men, and in less detail.

Since our analyses require complete work and schooling histories up to a given date, we imposed several important sample-selection restrictions. First, to ensure that we observed the beginning of the transition from school to work (and to avoid analyzing left-censored histories), we required that the individual still be in school as of the period covered (retrospectively) by the first interview (January 1, 1978). Thus, our high school graduate subsample should have graduated high school with the classes of 1978 through 1983. Our some college group and college graduates may include members of the high school classes of 1976 through 1983. Finally, we included only high school dropouts from the classes of 1981 and later.

Second, we included an individual's experiences in the estimation only until he/she had missing data. Even if the respondent missed an interview but was interviewed at a subsequent interview, that information is included in our calculations only if we could fill in his/her experiences over the gap caused by the missing interview.

Combined with the sampling scheme for the NLS-Y (i.e., it is a stratified sample from several cohort-year groups), these sample-selection conditions make our sample extremely unbalanced. We oversampled men and women who were younger at the first interview and those who received more education. Table A.1 shows the weighted distribution of SLGs by age at the first interview, separately for men and women. For men, the percentage of high school dropouts decreases by 1.4 percentage points (from 35.32 to 33.95) between those aged 15 and those aged 16 at the first interview. The decrease is fully 9.3 percentage points between ages 16 and 17. The difference is primarily due to the increase in the fraction of men who left school before the retrospective period covered by the first interview, which began on January 1, 1978 (shown in the "<78" column of Table A.1). The proportion of men in this category increases steadily with the age of first interview, representing nearly 30 percent of the original NLS-Y sample. Similar patterns hold for women. Those individuals who were school leavers before the first interview are the main group excluded from our sample. Problems with missing data led us to delete another 7.2 percent of the original sample of men and 7.0 percent of the sample of women.

Table 2.2 contains the final sample sizes by SLG. The second column contains the raw sample sizes. For men and women, sample sizes for the first three school-leaving groups are well over 700, and for college graduates the numbers are over 300. The sample for the BA+ SLG is under 150 for both men and women--too small for analysis; consequently, we do not report results for them. The last three columns show the unweighted, weighted, and reweighted percentage distributions of the sample, respectively. The weighted column applies the 1979 NLS-Y interview weights. Those weights correct for nonresponse to the first interview. The final column presents our best estimate of the true distribution of membership in SLGs in the population. This reweighted distribution is computed by aggregating across the weighted distribution of those aged 14 and 15 at the first interview (from Table A.1). We use these reweighted weights for the analyses of the distribution of individuals across SLGs.

Thus, according to our definition of school-leaving groups, in the early 1980s we estimate that the male youth population consisted of about one-third high school dropouts (36.9 percent) and another one-third high school graduates (35.3 percent). About one in five men proceeded directly to post-secondary education but did not receive a BA[8] before leaving school (18.6 percent); fewer than one in ten left school with a college degree (6.7 percent). Less than 3 percent proceeded directly to college and then directly from college to post-college education (2.6 percent). Compared with men, the fraction of women leaving school without a high school diploma is smaller (28.9 percent). The difference is accounted for by the larger percentage of women who leave school with either a high school degree (39.6 percent), some college education (20.0 percent), or a college degree (8.9 percent). A similar fraction of women leaves school with additional schooling beyond their college degree (2.5 percent). For both men and women, the high school dropout percentage reported here is considerably higher than that reported in most other sources, and the college graduate percentage is considerably lower (Frase, 1989; Haggstrom et al., 1991).

Before discussing our main results, we first reconcile the difference between the distribution of sample members by SLGs and the distribution by completed schooling. As we show below, this discrepancy is due to the classification as of when school leaving occurs and to subsequent return to school after school leaving.

Return to School After School Leaving

We assigned NLS-Y respondents to SLGs on the basis of their degree attainment as of the first time they were not in school (as their primary activity) for longer than the typical school break. Thus, by our definition, school leaving occurs when full-time work (with or without simultaneous school attendance) or an activity other than school attendance takes place for more than three to five months.

The assigned SLG does not, however, indicate the final degree attained. To the extent that individuals return to school, either by combining full-time or part-time work with schooling or by attending school only after a break in their education, the SLG and attained schooling will differ. Thus, for example, under our definition, some high school students may be working 35 or more hours per week and attending school. In that case, they would be classified as high school dropouts, even though they attain a high school degree in the usual time frame or some amount of post-secondary schooling.

Table 2.3 addresses the timing of school attendance and the extent of return to school, separately by gender and SLG.[9]

This table presents the only results in the report for "completed schooling." All other results are for school-leaving group, regardless of "completed schooling." In Table 2.3 we see a considerable amount of return to school, especially for those with incomplete degree attainment. For example, over 80 percent of men in the some college SLG (which includes men with associate's degrees) ever return to school; and about 60 percent return to school on a full-time basis. Almost 70 percent of male high school dropouts eventually return to school; over half return to school on a full-time schedule. Rates of return are almost as high, about 60 percent, for men who first leave school immediately after having completed high school or bachelor's degrees, although full-time attendance is much less likely for these groups. The patterns are broadly similar for women, although fewer women with a high school degree ever return to school. Compared with men, more women with post-college education are likely to return to school, and they are more likely to do so on a full-time basis.

While a large fraction of youth returns to school, completion rates are much lower. The figures for male high school dropouts help to explain why our dropout rates are higher than those reported elsewhere in the literature: Our definition corresponds to the general image of dropouts as those who leave school without attaining a regular high school diploma. Figure 2.1 plots the timing of school attendance and diploma receipt for male high school dropouts by years since school leaving. Figure 2.2 plots the same information for female high school dropouts. One-third of the young men in this cohort eventually received regular high school diplomas (i.e., excluding high school equivalency certificates [GEDs]), and nearly another third received GEDs. Not surprisingly, 95 percent of the high school diplomas (excluding the GEDs), and four out of five of the GEDs were obtained within the first three years of school leaving.

Thus, at school leaving (when the SLG is set), the high school dropout SLG represents about one-third of our male sample (36.9 percent). However, in the adult population--after accounting for additional schooling attained after school leaving--the high school dropout SLG is only two-thirds of that figure (24.8 percent): If we include GED recipients among the high school graduates rather than dropouts, the high school dropout group is only one-third the size of the dropout SLG.[10]^,[11]

Across all SLGs, the pattern of return to school shown in Table 2.3 implies relatively standard schooling attainment rates. Eventually, 75.2 percent of the men and 80.4 percent of the women in this cohort received conventional high school degrees. Another 10.5 percent of men and 8.6 percent of women received GEDs. The remaining high school dropouts are only 14.3 percent of the population of men and 11.0 percent of the population of women. Another 21.1 percent of men eventually received college degrees, and 4.1 percent received at least a master's degree. A similar fraction of women achieved these two levels of higher education.

Thus, the distribution of the sample by SLGs differs from the distribution by completed schooling, because a substantial fraction of men and women reached their final degree status with gaps in their school attendance. Those leaving school without high school degrees are nearly evenly divided between those who eventually received high school diplomas, those who received GEDs, and those who received neither diplomas nor GEDs.

Figure 2.1--Return to School and Diploma Receipt for Male High School Dropouts

Figure 2.2--Return to School and Diploma Receipt for Female High School Dropouts

Similar patterns exist at higher levels of education. Less than half of those who eventually received BAs remained in school continuously until they received their degrees.

Alternative Approaches to Defining School-Leaving Groups

As we noted above, operationalizing the concept of leaving school and entering the labor market is difficult, given the complexities of the school-to-work transition. We proceed in the body of this report to use the definition of SLGs outlined above, but we also examine the sensitivity of our findings to two variations. Specifically, above we define school leaving to have occurred when a young person works full-time (35 or more hours per week) for 4 to 6 consecutive months, even if he/she is also in school. Figures 2.1 and 2.2 suggest that this definition is problematic for high school dropouts: Nearly half of them are in school the month after we consider them school leavers, i.e., this is the first of 4 or more months (up to 6 months over the summer) of full-time work. Consistent with other analyses (e.g., Michael and Tuma, 1984), these results can be interpreted as showing that many high school students work enough to be classified as "full time."

To address the sensitivity of our results to this issue in the definition of school leaving, we replicate in Appendix B the set of results we present in this section and in Sections 3 and 4 for men, using a modified definition of SLGs that does not consider school leaving to have occurred until either the young person is not in school at all or has graduated from high school (at which time the 4 to 6 months of full-time work rule is applied). Using the modified definition causes a smaller (reweighted) fraction of the NLS-Y cohort to be assigned to the dropout category (29.0 percent versus 36.9 percent), and a higher fraction to the remaining SLGs, particularly high school graduates (39.4 percent versus 35.3 percent). With the exception of high school dropouts, the results are very similar to those presented in Sections 3 and 4 for each of the schooling groups. The alternative definition gives bleaker perspectives of the early labor market experience of high school dropouts than the results in the following sections.

Appendix C extends the sensitivity analysis by using the alternative SLG definition and stratifying the analysis by those who never returned to school versus those who ever returned to school. With the exception of those in the some college group, there is very little difference in the timing of the transition to stable employment. In general, those who eventually return to school take more time to attain longer-tenure jobs, as would be expected because they are more likely to have interrupted their labor market career. The magnitude of the differences by return-to-school status is modest, however.

[1] Prior to 1981, the NLS-Y data collected more limited information on school attendance. During that period, most (but far from all) of the sample were in school. Individuals who were in school and who appeared to be at grade level (given their age and previous answers to school enrollment questions) were assumed to have always been in school. Additional details of our procedure for filling in the missing monthly schooling information are available on request.

[2] A hazard rate defines, for all people whom we observed at age Aand again at age A + 1 (where age is measured in months), the probability of a given event occurring (e.g., being in a job for M months) given that the event had not already occurred before age A.

[3] Although the raw data (on percentage of people experiencing an event) show some nonmonotonicity due to sampling error (and, perhaps, time nonstationarity, which is ignored in this report), this procedure forces the plots to be monotonic (i.e., the percentage of people who have received a high school diploma never drops).

[4] Other studies using the NLS-Y data define schooling groups according to schooling attainment at the end of the panel, which is not consistent with our perception of the way "high school dropout" or "high school graduate" is used colloquially or in the policy literature (see, for example, Veum and Weiss, 1993). In addition, such ex post classifications are difficult to use in our prospective hazard-based analysis strategy.

[5] We adopted this definition because the NLS-Y does not have a full-time school indicator. Before age 16, school attendance alone is used to define school leaving.

[6]People are not included in the calculations until they leave school and we can assign an SLG. This means that SLGs grow as people leave school (e.g., some of those in the "some college" SLG enter the sample at age 19, but many do not enter the sample until age 20 or later).

[7]Our methods implicitly assume that panel attrition is random, conditional on the stratifying variables--gender, age, education, and, in some analyses, race. This assumption is consistent with most detailed studies of attrition bias (Klerman, 1992; Becketti, Gould, Lillard, and Welch, 1988).

[8]For simplicity, we use "BA" to denote all bachelor's degrees.

[9]The incidence of school return and final degree status are based on the information as of the last available interview, which varies across individuals because of attrition and varying age at initial interview.

[10]These high school completion rates for the population are computed by multiplying the share of individuals in the high school dropout SLG by the percentage of high school dropouts that ever gets a high school diploma or a GED. Since almost all such schooling attainment (through about age 31 at least) takes place within the first three years (see Figure 2.1), these numbers are a fair approximation for a recent cohort of the general population aged 21 and over.

[11]See Cameron and Heckman (1993), who argue that a GED recipient should be treated as a dropout, not as a high school graduate.