Re-examining the effects of parental characteristics on educational attainment for a minor child.
One of the most difficult cases for a forensic economist is to predict lost earnings capacity for an injured or deceased child. A common method for calculating lost earnings capacity is to use nation mean or median earnings (from Current Population Survey census data, for example) by educational level with the appropriate adjustments for characteristics such as race or gender. Therefore, forensic economists are very interested in the ability to predict a child's educational attainment based on information available to the economist, such as the parents' educational attainments and family demographic information.
Four previous articles have addressed this issue. Hardy (1993) uses the correlation between parental education and children's education to predict educational attainment and subsequent earnings. Spizman and Kane (1992) provide estimates of the effect of family background on the educational attainment of a minor child. They find that children who are raised in an urban area and children whose parents have more education are more likely to attend college. Gill and Foley (1996) extend the analysis of Spizman and Kane (SK92) in three important ways. First, unlike SK92, Gill and Foley (GF) use a data set (National Longitudinal Survey of Youth (NLSY)) that contains information on high school dropouts. Second, GF utilize the extensive family controls for the occupation of the parent or adult, religion, and family structure. Third, they use a more recent data set than in SK92. GF find support for SK92's results and find that the additional controls matter when explaining the educational attainment of a child. Spizman and Kane (2001) extend GF by using more recent NLSY data. SK01 find that the additional data increase the number of survey participants who report at least a high school degree relative to GF.
At the same time, the human capital literature, starting with the Coleman (1966) report, has long emphasized that family background characteristics are important determinants of educational attainment. (1) More recently, Cameron and Heckman (2001) find that family background characteristics, particularly parental income, account for much of the disparity between blacks and whites in college attendance. (2) Similarly, Keane and Wolpin (2000) show that "endowments" at age 16 explain much of the difference between black and white college enrollment and labor market success. Another development in the literature is the research on private versus public high school attendance. Evans and Schwab (1995), Neal (1997), and others show that private school attendance has a large, positive effect on high school graduation and college attendance.
By utilizing the human capital literature, we further extend the forensic economics literature in four areas. First, we provide a more complex measure of educational attainment. Like SK01, we utilize data from the NLSY up to 1998 (GF use data through 1992). Unlike previous work, we restrict our sample in such a way as to minimize the possibility that a respondent obtains additional schooling after the NLSY survey. Second, we include respondents who report only one parent's education level. This modification increases the number of minority respondents, as well as the number of respondents from single-parent backgrounds. Third, following the human capital models, we include two additional family background characteristics, household income and private schooling. Throughout our analysis, we focus on variables that are easily verified by the forensic economist. Fourth, because the results generally support previous work in this area, this study provides additional validity for the use of this methodology by the forensic economist.
We follow the estimation strategy of SK92, GF, and SK01. The dependent variable is the highest degree received, a measure of educational attainment. As is common in the human capital literature, we model educational attainment as a function of student characteristics, family characteristics, and demographic information. Specifically, the equation for individual i can be written as follows:
(1) [Z.sub.i] = [X'.sub.i][beta] + [u.sub.i]
where [Z.sub.i] is educational attainment, [X.sub.i] is a set of student, family and demographic characteristics, and [u.sub.i] is the random error. In our study, as well as in previous work, we define educational attainment as a categorical variable measuring the highest degree received. Because this variable has a natural ordering, from high school dropout to Ph.D. or equivalent degree, we estimate an ordered probit model (as do SK92, GF, and SK01). According to the model, individual i obtains:
less than a high school diploma if [Z.sub.i] [less than or equal to] 0, a high school diploma or GED if 0 < [Z.sub.i] [less than or equal to] [[mu].sub.1], some college (but less than a BA/BS) if [[mu].sub.1] < [Z.sub.i] [less than or equal to] [[mu].sub.2], a bachelor's degree if [[mu].sub.2] < [Z.sub.i] [less than or equal to] [[mu].sub.3], a Master's degree if [[mu].sub.3] < [Z.sub.i] [less than or equal to] [[mu].sub.4], and a Ph.D. or equivalent if [Z.sub.i] > [[mu].sub.4].
As GF point out, the [mu]'s are threshold values that, like [beta], are unknown and need to be estimated.
Like GF and SK01, we use the NLSY data set, a nationally representative sample of 12,686 young adults, ages 14-21 in 1979. We restrict the sample to individuals who have no missing data and who answered at least one survey between 1988 and 1998. By restricting the sample in this way, we maximize the opportunity that respondents have completed their schooling before responding to their last survey. Respondents are ages 24 to 31 in 1988 and ages 34 to 41 in 1998, and the average age at last survey is 35. Our final sample contains 3,638 men and 3,582 women.
As in previous work, our dependent variable is the highest degree received. It is based on the most recent values of three survey questions: the highest degree received (on every survey between 1988 and 1998), the highest grade completed (on every survey), and high school completion/GED (on every survey between 1980 and 1998). The categories for the survey question on highest degree received include high school, bachelor's, master's, and doctorate/equivalent degrees. We, along with previous work, add categories for "some college" and "less than high school." In GF and SK01, the "some college" category consists of two groups: recipients with associate's degrees, and recipients with high school diplomas (or GEDs) and 13-15 years of schooling. We also include high school graduates (and GED recipients) with 16 or more years of schooling in our "some college" category.
GF and SK01 define "less than high school" as all respondents with missing values for highest degree received, since that question does not contain a high school dropout category. Our treatment of students with missing values of highest degree received is much different from GF and SK01, as we believe that some of these missing values are due to a failure to report the highest degree received, rather than the absence of a degree. For respondents with missing values of highest degree received, we use the most recent survey value of highest grade completed to identify bachelor's degrees (16 years of more of schooling) and some college (13 to 15 years). For respondents with 12 years of schooling or less (and missing highest degree received), we use the most recent value of a question identifying whether the respondent has received a high school diploma or a GED. This expanded definition, along with our sample restriction to people who answered at least one survey when aged 24, minimizes the possibility of misidentifying high school dropouts.
Appendix Table 1 provides unweighted descriptive statistics (separately for males and females) for educational attainment and student characteristics. Nearly 15% of the men in our sample are high school dropouts, compared with 11% of the women. Similarly, 43% of the men completed high school or a GED but did not attend college, compared with 40% for women. Roughly 22% of both sexes have either an undergraduate or graduate degree, while over 20% have some college attendance (but no degree). Because our definition of some college is more inclusive than either GF or SK01, both studies have a much smaller percentage of respondents in that category. However, comparisons between our sample and the two mentioned previously are complicated by the fact that they report weighted means, and we report unweighted means. For example, while the dropout rate in SK01 is much lower than our unweighted dropout rate, our weighted dropout rate (not reported), using the 1998 sample weighted employed by SK01, is similar to theirs. Weighted means for our sample are similar to those in SK01. (3) Furthermore, the issue of weights is not important for the probit results, which control for demographic characteristics. In fact, SK01 report that their weighted and unweighted probit results are quite similar.
Appendix Table 1 also contains the controls for family background and demographics. These variables are from the 1979 survey. As stated previously, we control for the highest education level of either parent, rather than for each parent (as is done in SK92, GF, and SK01). This definition allows us to provide estimates for single-parent families and other families that contain educational information for only one parent. On the other hand, we implicitly assume that the effect of parental education is the same for each parent.
In addition to the family background characteristics used by GF (and SK01), we also include controls for family income and private school attendance. The income variable is the natural log of the household income in 1979, as reported by the respondent in the 1979 survey. (4) We also include a dummy variable for whether or not the student attended private secondary schooling. As stated previously, both variables are important determinants of educational attainment. Furthermore, the inclusion of private schooling allows us to separate the effect of religion from that of private schooling, as religion is an important predictor of private school attendance as well as educational attainment (Evans and Schwab, 1995; Neal, 1997).
IV. Estimation Strategy
Like GF and SK01, we estimate several models, the first two of which are equivalent to the first two models in those papers (except for the different definition of parental education). The first model is also analogous to the model estimated in SK92 and contains dummy variables for black, Hispanic, living in an urban area at age 14, and highest education level of either parent (high school, one to three years of college, and four or more years of college). The second model contains additional controls for household characteristics. We measure socio-economic status with two dummy variables that identify whether the adult in the household at age 14 had a professional/managerial occupation or a sales/clerical occupation. We also include dummy variables for being an only child, having both parents in the household, and for the religion raised (Baptist, other Protestant, Catholic, Jewish, other, or no religion). The third model specification includes variables for household income in 1979 and private school attendance for most recent school in 1979, in order to isolate better the controls in the second model as well as to provide more complete controls for vital family background information. Finally, the fourth model adds three dummy variables used by GF and SK01 to measure household literacy: regular newspaper, magazine subscription, and library card.
Table 1 reports the results of the ordered probit in Model I. As GF emphasize, these results are not directly comparable to the SK92 results, which do not contain data on high school dropouts. Instead, we compare our results to those in GF and SK01, although our sample and explanatory variables are slightly different. Even with these caveats, our results for Model I are similar to those in the previous studies: parental education has a strong, positive effect on student achievement, while the effects of race and urbanicity are not as strong. The size of the coefficient on parental education increases with the amount of parental education. In other words, the largest parental education coefficient is for college graduate, the next largest is some college, and the smallest is high school graduate. (5) Like GF and SK01, the coefficients of race for black and Hispanic are negative for both sexes. While our coefficients for race are always insignificant, GF find a significant coefficient for black males, while SK01 find a significant coefficient for Hispanic males. We find insignificant effects of urbanicity, but GF find positive effects of urbanicity for males and SK01 find negative effects for females.
To understand better the role of background characteristics such as race on educational attainment, Model II provides additional controls for student background. The results from this model are reported in Table 2. The overall pattern of results is quite similar to the results in GF and SK01, as well as to the earlier model. Again, parental education has a highly significant, positive effect on educational attainment. Similarly, being raised in a two-parent household or having an adult in a professional, sales, or clerical occupation is associated with higher educational attainment. Like GF, we also find a positive and significant coefficient for being black in the model for females. Being raised in a religion generally has positive effects on educational attainment, although the size of the effect depends on the religion. Being an only child is associated with higher educational attainment for female respondents, but has no effect for male respondents.
While the overall pattern of results for Model II is quite similar between our results and those in GF (their Table 4) and SK01 (also their Table 4), some interesting differences occur. The differences between our studies and theirs with regard to urbanicity and race remain, although all of us find positive effects for black females. The only significant religious effect in SK01 is for Jewish males. Our results and those in GF find significant effects for every religious category for males. For females, GF find a marginally significant, negative effect of being Baptist, while we find no effect. They also find no effect of being Catholic, while we find a significant positive effect. The results from Model II demonstrate the important predictive power of family background characteristics such as occupation, family structure, and religion.
In Table 3 we report the results from Model III. This model contains two additional family background variables: household income and private schooling. We find that income has a positive and significant impact on educational attainment for women, but has no effect for men once we control for adult occupation. (6) For both sexes, we find a highly significant, positive effect of private schooling. The inclusion of these variables does not affect the significance of any of the independent variables from Model II. Thus, these additional variables provide more information about educational attainment, but the their exclusion (as in GF and SK01) does not appear to bias the coefficients of the included variables.
Appendix Table 2 contains the results from Model IV, which includes three measures of household literacy from GF Model III: regular newspaper, magazine subscription, and library card. Consistent with GF, we find that each of these is associated with higher educational attainment and that the inclusion of these variables does not affect the significance of the coefficients of the variables from Model III. SK01 find an insignificant effect of newspapers in the female model, but find positive effects for all other measures of household literacy.
While the results in the tables provide estimates of the specific impacts of background characteristics on educational attainment, they do not contain predictions for education levels. Table 4 contains predicted probabilities for a hypothesized individual who has similar values as in Table 7 of SK01. For this "person," all the variables are equal to zero except for parental education, professional occupation, and household income. Thus, this person is a white male in a rural area, with parental education of some college, at least one parent has a professional occupation, has the mean parental income of approximately $12,400, no religion raised, at least one sibling, in a single-family home, and public schooling. As suggested by GF, such an individual is probably not very common, as most individuals have both parents and report a religion. We include the estimates for Models I through III, which contain characteristics of use to forensic economists because of the ease of verifying such information. A comparison across models illustrates the impact of additional explanatory variables on the predicted educational attainment. The comparison of our results to those of GF and SK01 also allows us to investigate the impact of differences across samples.
The results in Table 4 illustrate that the predicted probabilities vary both by model and by sample. The variation in predictions for Model II between the three samples is striking. SK01 find a probability of not completing high school of only 7.2%, compared to 14.1% for our sample and 18.7% for GF. Adding the 1993 to 1998 NLSY data creates a tremendous increase in the predicted educational attainment, illustrated by the difference in predicted probabilities between GF and SK01. The changes in samples between our data and SK01, who also use 1993 to 1998 NLSY data, produce smaller but substantial decreases in the predicted educational attainment. In particular, these differences result from the way in which we define parental education and the necessity that respondents in our sample contain data on family income. The overall result is that our predicted probabilities for Models I and II are more similar to those from GF than to those from SK01.
Despite the fluctuations between samples for Model II, several noticeable changes exist between models regardless of the sample used (GF, SK01, or ours). The inclusion of parental occupation, religion, and family structure in Model II has a large impact on the predicted probability. Model I has a much lower predicted probability of dropping out of high school (29 to 33%), compared to Model II (7 to 19%). Model III has the lowest predicted dropout rate at only 4%. For high school graduation, Model II has a considerably higher probability than either Model I or III. Correspondingly, the predicted probability for each subsequent educational level is lower in Model II than in Models I and III. In other words, Model III provides the highest probability of at least some college education.
This paper blends human capital models with forensic economic models to best predict the educational attainment of a child. We extend the analysis in GF and SK01 in three ways. First, we define parental education as the highest educational level of either parent. This change allows us to include respondents who have information on only one parent's education level. Second, we define the dependent variable, educational attainment, to include only individuals surveyed from 1988 to 1998, and to include more individuals in the "some college" category. This definition maximizes the likelihood that respondents in our sample have completed their schooling. Consequently, our sample contains more respondents who have at least some post-high school education. Third, we include household income and private schooling in the model, since the human capital literature finds that these variables are strong predictors of educational attainment. As expected, we find strong positive effects of private schooling on educational attainment. Income has a positive and significant effect for women but not for men.
Our results support the findings from GF and SK01 that family background characteristics are important predictors of educational attainment. In particular, we also find that adult occupation, religion, and family structure (two-parent family and only child status) generally have significant, positive effects on educational attainment. However, the models that omit variables for family income and private schooling produce lower predicted educational attainment than does our model, as illustrated by the predicted probabilities in Table 4.
These results can assist the forensic economist in several ways. One contribution is that our analysis, in conjunction with the previous studies, identifies a set of characteristics that the forensic economist can use to predict the educational attainment of a minor child. For example, the results reported in Table 4 illustrate these predicted probabilities for a child with certain demographic characteristics.
Another contribution is that our results can bolster the credibility of a range of earnings capacity estimates provided by a forensic economist. For example, suppose a forensic economist is asked to calculate the lost earnings capacity for a child (white, female) who was injured at age 10. The forensic economist presents a range of pre-injury earnings capacity calculations based on the average earnings of a white female for three different levels of educational attainment: a high school degree, some college, and a college degree. If the forensic economist knows that the injured child's mother has a college degree, that the family had above-average income (pre-injury), and that the child attended a Catholic elementary school (pre-injury), our results suggest that the upper end of the range based on the average earnings of a white female with a college degree represents a reasonable earnings capacity prediction (assuming that all other factors are appropriately addressed). (7)
Appendix Table 1 Descriptive Statistics Males Variable Mean Std Dev Highest Degree Less than HS/GED 0.144 0.351 High School/GED 0.429 0.495 Some College 0.210 0.407 BA/BS 0.162 0.369 Master's 0.039 0.193 PhD or Equivalent 0.016 0.126 Hispanic 0.158 0.365 Black 0.230 0.421 White / Other 0.613 0.487 Urban 0.789 0.408 Highest Education Level of Either Parent Less than HS/GED 0.290 0.454 High School/GED 0.415 0.493 Some College 0.127 0.333 4+ Years of College 0.168 0.374 Adult Occupation (Either Adult) Professional 0.253 0.435 Sales/Clerical 0.221 0.415 Log Household Income 9.425 1.015 Religion Raised Baptist 0.270 0.444 Protestant 0.239 0.427 Catholic 0.342 0.475 Jewish 0.009 0.095 Other 0.097 0.296 None 0.042 0.199 Only Child 0.032 0.175 Both Parents 0.748 0.434 Private Schooling 0.057 0.231 Newspapers 0.794 0.404 Magazines 0.618 0.486 Library Card 0.711 0.453 Observations 3638 Females Variable Mean Std Dev Highest Degree Less than HS/GED 0.108 0.311 High School/GED 0.396 0.489 Some College 0.276 0.447 BA/BS 0.169 0.375 Master's 0.044 0.206 PhD or Equivalent 0.006 0.078 Hispanic 0.166 0.372 Black 0.225 0.417 White / Other 0.609 0.488 Urban 0.787 0.409 Highest Education Level of Either Parent Less than HS/GED 0.318 0.466 High School/GED 0.400 0.490 Some College 0.126 0.332 4+ Years of College 0.156 0.363 Adult Occupation (Either Adult) Professional 0.251 0.433 Sales/Clerical 0.217 0.412 Log Household Income 9.332 1.164 Religion Raised Baptist 0.272 0.445 Protestant 0.226 0.418 Catholic 0.351 0.477 Jewish 0.008 0.090 Other 0.112 0.315 None 0.031 0.174 Only Child 0.026 0.159 Both Parents 0.742 0.437 Private Schooling 0.063 0.242 Newspapers 0.784 0.412 Magazines 0.595 0.491 Library Card 0.738 0.440 Observations 3582 Appendix Table 2 Ordered Probits, Model IV Males Variable Coeff Std Err Intercept -0.4770 0.1954 ** Hispanic 0.0419 0.0599 Black 0.1362 0.0506 *** Urban at Age 14 0.0288 0.0461 Parent is High School Graduate 0.3205 0.0484 *** Parent Has Some College 0.5349 0.0672 *** Parent Has 4+ Years of College 1.0141 0.0699 *** Adult is Professional 0.4047 0.0502 *** Adult is Clerical or Sales 0.2085 0.0452 *** Log Household Income 0.0128 0.0185 Raised in Baptist Religion 0.3950 0.0980 *** Raised in Protestant Religion 0.5300 0.0982 *** Raised in Catholic Religion 0.4874 0.0989 *** Raised in Jewish Religion 1.1335 0.2066 *** Raised in Other Religion 0.5608 0.1071 *** Only Child 0.1742 0.1008 * Two Parent Family 0.1495 0.0431 *** Attended Private School 0.3644 0.0775 *** Newspaper at Age 14 0.0997 0.0500 ** Magazines at Age 14 0.3618 0.0420 *** Library Card at Age 14 0.2080 0.0431 *** [[mu].sub.1] 1.4816 0.0320 *** [[mu].sub.2] 2.2247 0.0374 *** [[mu].sub.3] 3.2315 0.0501 *** [[mu].sub.4] 3.8689 0.0678 *** Observations 3638 Log likelihood -4697 Females Variable Coeff Std Err Intercept -0.3264 0.1834 * Hispanic 0.0755 0.0599 Black 0.3300 0.0501 *** Urban at Age 14 0.0824 0.0461 * Parent is High School Graduate 0.3195 0.0476 *** Parent Has Some College 0.6632 0.0660 *** Parent Has 4+ Years of College 1.0596 0.0721 *** Adult is Professional 0.3237 0.0509 *** Adult is Clerical or Sales 0.1093 0.0456 ** Log Household Income 0.0679 0.0165 *** Raised in Baptist Religion 0.0110 0.1089 Raised in Protestant Religion 0.2281 0.1092 ** Raised in Catholic Religion 0.1681 0.1094 Raised in Jewish Religion 0.8387 0.2220 *** Raised in Other Religion 0.0574 0.1154 Only Child 0.2476 0.1115 ** Two Parent Family 0.1807 0.0428 *** Attended Private School 0.2807 0.0754 *** Newspaper at Age 14 0.1146 0.0496 ** Magazines at Age 14 0.2522 0.0419 *** Library Card at Age 14 0.1588 0.0449 *** [[mu].sub.1] 1.4488 0.0331 *** [[mu].sub.2] 2.3756 0.0393 *** [[mu].sub.3] 3.4102 0.0517 *** [[mu].sub.4] 4.3672 0.0895 *** Observations 3582 Log likelihood -4596 Notes: *, **, and *** indicate statistically significant coefficients at ten, five, and one percent, respectively. Table 1 Ordered Probits Model I Males Variable Coeff Std Err Intercept 0.5685 0.0517 *** Hispanic -0.0532 0.0527 Black -0.0797 0.0445 * Urban at Age 14 0.0653 0.0443 Parent is High School Graduate 0.5509 0.0454 *** Parent Has Some College 0.9367 0.0620 *** Parent Has 4+ Years of College 1.5815 0.0594 *** Adult is Professional Adult is Clerical or Sales Log Household Income Raised in Baptist Religion Raised in Protestant Religion Raised in Catholic Religion Raised in Jewish Religion Raised in Other Religion Only Child Two Parent Family Attended Private School Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.3983 0.0300 *** [[mu].sub.2] 2.0985 0.0351 *** [[mu].sub.3] 3.0533 0.0471 *** [[mu].sub.4] 3.6662 0.0646 *** Observations 3638 Log likelihood -4877 Females Variable Coeff Std Err Intercept 0.8033 0.0523 *** Hispanic 0.0248 0.0522 Black 0.1012 0.0448 ** Urban at Age 14 0.0228 0.0441 Parent is High School Graduate 0.5215 0.0447 *** Parent Has Some College 0.9932 0.0614 *** Parent Has 4+ Years of College 1.5810 0.0604 *** Adult is Professional Adult is Clerical or Sales Log Household Income Raised in Baptist Religion Raised in Protestant Religion Raised in Catholic Religion Raised in Jewish Religion Raised in Other Religion Only Child Two Parent Family Attended Private School Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.3886 0.0317 *** [[mu].sub.2] 2.2752 0.0375 *** [[mu].sub.3] 3.2679 0.0493 *** [[mu].sub.4] 4.2172 0.0883 *** Observations 3582 Log likelihood -4730 Notes: *, **, and *** indicate statistically significant coefficients at ten, five, and one percent, respectively. Table 2 Ordered Probits Model II Males Variable Coeff Std Err Intercept -0.0816 0.1045 Hispanic -0.0491 0.0587 Black 0.0950 0.0501 * Urban at Age 14 0.0520 0.0449 Parent is High School Graduate 0.4269 0.0470 *** Parent Has Some College 0.6888 0.0656 *** Parent Has 4+ Years of College 1.1991 0.0677 *** Adult is Professional 0.4695 0.0497 *** Adult is Clerical or Sales 0.2738 0.0447 *** Log Household Income Raised in Baptist Religion 0.3788 0.0974 *** Raised in Protestant Religion 0.5591 0.0977 *** Raised in Catholic Religion 0.5312 0.0979 *** Raised in Jewish Religion 1.1715 0.2064 *** Raised in Other Religion 0.5387 0.1066 *** Only Child 0.1865 0.1004 * Two Parent Family 0.1865 0.0424 *** Attended Private School Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.4406 0.0310 *** [[mu].sub.2] 2.1673 0.0364 *** [[mu].sub.3] 3.1550 0.0488 *** [[mu].sub.4] 3.7836 0.0665 *** Observations 3638 Log likelihood -4775 Females Variable Coeff Std Err Intercept 0.4440 0.1153 *** Hispanic -0.0285 0.0584 Black 0.2812 0.0496 *** Urban at Age 14 0.0344 0.0449 Parent is High School Graduate 0.4256 0.0461 *** Parent Has Some College 0.7961 0.0645 *** Parent Has 4+ Years of College 1.2213 0.0703 *** Adult is Professional 0.3913 0.0503 *** Adult is Clerical or Sales 0.1603 0.0452 *** Log Household Income Raised in Baptist Religion 0.0042 0.1083 Raised in Protestant Religion 0.2747 0.1085 ** Raised in Catholic Religion 0.2506 0.1085 ** Raised in Jewish Religion 0.8601 0.2213 *** Raised in Other Religion 0.1083 0.1148 Only Child 0.2607 0.1112 ** Two Parent Family 0.2419 0.0421 *** Attended Private School Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.4182 0.0324 *** [[mu].sub.2] 2.3315 0.0385 *** [[mu].sub.3] 3.3553 0.0508 *** [[mu].sub.4] 4.3104 0.0890 *** Observations 3582 Log likelihood -4652 Notes: *, **, and *** indicate statistically significant coefficients at ten, five, and one percent, respectively. Table 3 Ordered Probits Model III Males Variable Coeff Std Err Intercept -0.3203 0.1940 * Hispanic -0.0338 0.0589 Black 0.0898 0.0502 * Urban at Age 14 0.0477 0.0450 Parent is High School Graduate 0.4140 0.0474 *** Parent Has Some College 0.6737 0.0658 *** Parent Has 4+ Years of College 1.1649 0.0683 *** Adult is Professional 0.4586 0.0498 *** Adult is Clerical or Sales 0.2602 0.0448 *** Log Household Income 0.0291 0.0183 Raised in Baptist Religion 0.3721 0.0974 *** Raised in Protestant Religion 0.5480 0.0977 *** Raised in Catholic Religion 0.4875 0.0983 *** Raised in Jewish Religion 1.1658 0.2064 *** Raised in Other Religion 0.5417 0.1066 *** Only Child 0.1767 0.1005 * Two Parent Family 0.1705 0.0429 *** Attended Private School 0.3951 0.0773 *** Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.4462 0.0311 *** [[mu].sub.2] 2.1767 0.0365 *** [[mu].sub.3] 3.1693 0.0490 *** [[mu].sub.4] 3.8012 0.0669 *** Observations 3638 Log likelihood -4760 Females Variable Coeff Std Err Intercept -0.2707 0.1828 Hispanic -0.0031 0.0586 Black 0.2937 0.0498 *** Urban at Age 14 0.0486 0.0450 Parent is High School Graduate 0.4022 0.0464 *** Parent Has Some College 0.7650 0.0648 *** Parent Has 4+ Years of College 1.1690 0.0709 *** Adult is Professional 0.3726 0.0505 *** Adult is Clerical or Sales 0.1468 0.0453 *** Log Household Income 0.0829 0.0164 *** Raised in Baptist Religion 0.0096 0.1085 Raised in Protestant Religion 0.2805 0.1086 *** Raised in Catholic Religion 0.2119 0.1088 * Raised in Jewish Religion 0.8793 0.2215 *** Raised in Other Religion 0.0975 0.1149 Only Child 0.2432 0.1113 ** Two Parent Family 0.2121 0.0426 *** Attended Private School 0.2852 0.0752 *** Newspaper at Age 14 Magazines at Age 14 Library Card at Age 14 [[mu].sub.1] 1.4284 0.0326 *** [[mu].sub.2] 2.3462 0.0387 *** [[mu].sub.3] 3.3751 0.0512 *** [[mu].sub.4] 4.3323 0.0894 *** Observations 3582 Log likelihood -4631 Notes: *, **, and *** indicate statistically significant coefficients at ten, five, and one percent, respectively. Table 4 Predicted Probabilities of Alternative Levels of Educational Attainment Model I Model II Outcome Gill and Present Gill and Present SK Foley Paper Foley Paper 2001 Less than HS 5.2% 6.6% 18.7% 14.1% 7.2% High School/GED 35.3% 39.1% 51.3% 50.1% 64.4% Some College 26.3% 26.6% 17.8% 22.0% 8.4% BA/BS 25.8% 21.6% 10.9% 11.9% 16.6% Master's 5.0% 4.5% 0.98% 1.5% 2.7% PhD 2.5% 1.5% 0.27% 0.34% 0.67% Model III Outcome Present Paper Less than HS 4.2% High School/GED 34.7% Some College 28.4% BA/BS 25.2% Master's 5.6% PhD 1.9% Individual has the following characteristics: male, white, rural, parental education of some college, at least one parent with professional occupation, average log family income (9.4), no religion raised, not an only child, not with both parents, and public schooling. Literacy measures (newspaper, magazine, and library card) are not used in this table. Note that this individual is similar to the one used in SK01, Table 7.
(1) For a comprehensive review of this literature, see Haveman and Wolfe (1995).
(2) Neal and Johnson (2001) report similar findings when they look at black-white wage differences.
(3) Sample weights are available for each year of the survey. As our sample combines individuals from the 1988 through 1998 samples, appropriate weights for our sample are not available. Because SK01 also combine people from across sample years, weights are not appropriate for their study either. They only include weighted means in order to facilitate comparison between their sample and GF. They report unweighted probit results.
(4) Like the Current Population Survey, the income variable in the NLSY is topcoded at 75,001 dollars.
(5) High school dropout is the omitted category.
(6) One likely explanation for the insignificant result for males is measurement error in the household income variable, as suggested by Andrew Gill. Household income for students aged 18 to 22 in 1979 is noticeably lower than income for students aged 14 to 17. This pattern is consistent with the fact that older students (ages 18 to 22 in 1979) were more likely to live away from their parents in 1979, in which case the household income variable would not include parental income. When we allow the effect of income to vary with age, we find positive and significant effects of income for respondents ages 14 to 17 but not for respondents ages 18 to 22. This finding holds for both males and females.
(7) The authors make no assertions as to the appropriateness of calculating a lost earnings capacity for an injured minor child using the methodology contained in this example. We merely provide this example as illustrative of a potential application to one method.
Cameron, Stephen V. and James J. Heckman, "The Dynamics of Educational Attainment for Black, Hispanic, and White Males," Journal of Political Economy, 2001, 109 (3), 455-499.
Coleman, James S. and others, Equality of Educational Opportunity, 1966, Washington, D.C.: Government Printing Office.
Evans, William N. and Robert M. Schwab, "Finishing High School and Starting College: Do Catholic Schools Make a Difference?," Quarterly Journal of Economics, 1995, 110(4), 941-974.
Gill, Andrew M. and Jack Foley, "Predicting Educational Attainment for a Minor Child: Some Further Evidence," Journal of Forensic Economics, 1996, 9(2), 101-112.
Hardy, William E., "Use of National Opinion Research Center Data When Estimating Lost Earnings Capacity of an Injured Child," Journal of Legal Economics, 1992, 3(3), 73-86.
Haveman, Robert and Barbara Wolfe, "The Determinants of Children's Attainments: A Review of Methods and Findings," Journal of Economic Literature, 1995, 33(4), 1829-1878.
Keane, Michael P. and Kenneth I. Wolpin, "Eliminating Race Differences in School Attainment and Labor Market Success," Journal of Labor Economics, 2000, 18(4), 614-652.
Neal, Derek, "The Effects of Catholic Secondary Schooling on Educational Achievement," Journal of Labor Economics, 1997, 15(1), 98-123.
Neal, Derek A. and William R. Johnson, "The Role of Premarket Factors in Black-White Wage Differences," Journal of Political Economy, 1996, 104(3), 869-895.
Spizman, Lawrence M. and John Kane, "Loss of Future Income in the Case of Personal Injury of a Child: Parental Influence on a Child's Future Earnings," Journal of Forensic Economics, 1992, 5(2), 159-168.
--, and --, "An Update on the Educational Attainment Model for a Minor Child," Journal of Forensic Economics, 2001, 14(2) (this issue), 155-166.
Christopher A. Jepsen and Lisa K. Jepsen *
* Respectively, Research Fellow, Public Policy Institute of California, San Francisco, CA; and Assistant Professor, Department of Economics, the University of Northern Iowa, Cedar Falls, IA. The authors would like to thank Fred Abraham, Kenneth Brown, Andrew Gill, Elizabeth Gunderson, David Hakes, Thomas Ireland, Kurt Krueger, Jay Zagorsky, participants at the 2001 Western Economic Association annual meetings, and two anonymous referees for useful comments. All remaining errors are our responsibility. All opinions are ours and are not necessarily those of the Public Policy Institute of California or the University of Northern Iowa.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||calculating earnings capacity of an injured or deceased child|
|Author:||Jepsen, Christopher A.; Jepsen, Lisa K.|
|Publication:||Journal of Forensic Economics|
|Date:||Mar 22, 2001|
|Previous Article:||An essay on the New Worklife Expectancy Tables and the continuum of disability concept.|
|Next Article:||An update of the educational attainment model for a minor child.|