Results using earnings Directory UMM :Data Elmu:jurnal:L:Labour Economics:Vol7.Issue6.Nov2000:

mobility in status across generations than was suggested by earlier work. 1 Recent studies have found both greater and less mobility than that estimated by Solon or Zimmerman. 2 After attempting to replicate Solon’s results for earnings, I consider how the relative youthfulness of the sons in Solon’s sample affects the results by estimating the intergenerational correlations later in the sons’ lives and by looking at different measures of status for fathers and sons. First, I compare the estimates when earnings are used to proxy for status, and sons’ earnings from 5 years later are used in the estimation. Second, I examine the effects of using data on consumption as opposed to earnings to measure economic status. Using data on sons from 5 years later produces estimates of the correlation in earnings that are about 0.4. In contrast, the results using consumption data point to greater transmission of status than do those using earnings data. The consumption results that use later data are smaller and more similar to the results that use earnings than the consumption results that use the earlier data, particularly when family size is accounted for. As a whole, the results indicate that consumption is transmitted to children to a greater extent than are earnings. This is feasible if saving behavior is transmitted or if transfers which make the consumption patterns of the two generations more similar are passed between parents and children.

2. Results using earnings

I draw two samples from the PSID using Solon’s selection criteria. Sons in the samples were born between 1951 and 1959, have positive earnings in 1984, and are from original PSID families that were headed by men. Sons from the original Ž . Survey of Economic Opportunity SEO component of the PSID, which over-sam- ples low-income households, are omitted. The sample of all father–son pairs that meet these criteria has 428 observations while Solon’s sample contained 433. Of the pairs in the all-son sample, the eldest sons are retained in the eldest-son sample. My eldest-son sample contains 322 pairs while there were 330 father–son pairs in Solon’s. Although I have not succeeded in extracting Solon’s exact samples, the samples used in this study are quite similar to his. Table 1 reports descriptive statistics for Solon’s all-son sample and my all-son and eldest son samples. The descriptive statistics for the two all-son samples are nearly identical. As in Solon’s work, all monetary variables are transformed into 1984 dollars. Solon points out that averaging the father’s log earnings reduces the measure- ment error associated with using a snapshot of father’s earnings to proxy for 1 Ž . See Becker and Tomes 1986 for an extensive review of earlier work. 2 Ž . Ž . Mulligan 1997 and Lillard and Kilburn 1997 are examples of recent work that estimate less Ž . Ž . intergenerational mobility, while Couch and Lillard 1998 and Couch and Dunn 1997 estimate greater mobility across generations than do Solon or Zimmerman. Table 1 Descriptive statistics Variable Solon’s all-son sample All-son sample Eldest-son sample Mean Std. dev. Mean Std. dev. Mean Std. dev. Son’s age in 1984 29.6 2.4 29.5 2.5 29.6 2.5 Son’s earnings in 1984 22,479 15,019 21,624 14,028 21,860 12,996 Son’s log earnings 9.75 0.94 9.73 0.89 9.75 0.89 in 1984 Son’s consumption – – 8522.18 6090.48 8608.99 6291.09 in 1984 Son’s log cons in 1984 – – 8.86 0.64 8.86 0.66 Father’s age in 1967 42 7.7 42.6 7.1 42.7 7.4 Father’s earnings in 1967 29,304 20,015 29,429 20,926 29,342 20,787 Father’s log earn in 1967 10.1 0.69 10.1 0.68 10.09 0.7 Father’s consumption – – 6524.32 2909.65 6675.57 3005.97 in 1967 Father’s log cons in 1967 – – 8.69 0.46 8.71 0.46 Number of observations 433 428 322 permanent earnings, and thus reduces the downward bias of OLS estimates. Consequently, he uses father’s earnings from 1967 to 1971 and multi-year averages of earnings to proxy for father’s status to estimate the degree of intergenerational mobility. Table 2 compares Solon’s results from this procedure with my results. Panels a and b of Table 2 present Solon’s results from the unbalanced and balanced samples. 3 Panels c and d present the analogous results from the samples used in this paper. The differences between the two sets of results are small, and I believe that they arise from differences in the samples. Ž . Instrumental variables IV can also be used to reduce the error-in-variables problem associated with the measurement of father’s status. In Solon’s use of this technique, the number of years of schooling that the father has completed by 1968 serves as an instrument for father’s status. The PSID’s 1968 information on education is in interval form. Solon assigns the father’s years of schooling to be the midpoint of the reported interval except for fathers in the highest category, who are assigned 18 years of schooling. My IV estimate of the correlation between Ž . the two generations’ outcomes is 0.552 SE s 0.132 while Solon’s IV estimate is Ž . 0.526 SE s 0.135 . Solon points out that father’s education may not be a valid 3 The unbalanced and balanced samples are different in that the unbalanced sample contains all father–eldest son pairs while the balanced samples includes only those pairs for which father’s earning are available in all 5 years. The balanced sample is constant; as a result, the effect of changes in the measure of father’s status is isolated. In Solon’s work, the balanced sample contains 290 observations, while there are 280 in the balanced sample used here. Table 2 Comparison of Solon’s OLS results and my OLS results Year of father’s Measures of father’s log earnings log earnings 1-Year 2-Year 3-Year 4-Year 5-Year measures measures measures measures measures a Solon’s results from unbalanced sample 1967 0.386 Ž . 0.079 w x 322 0.425 Ž . 1968 0.271 0.090 0.408 Ž . w x Ž . 0.074 313 0.087 w x w x 326 0.365 309 0.413 Ž . Ž . 1969 0.285 0.081 0.369 0.088 0.413 Ž . w x Ž . w x Ž . 0.073 317 0.083 301 0.093 w x w x w x 320 0.342 309 0.357 290 Ž . Ž . 1970 0.285 0.078 0.336 0.088 Ž . w x Ž . w x 0.073 312 0.084 298 w x w x 318 0.290 301 Ž . 1971 0.247 0.082 Ž . w x 0.073 303 w x 307 b Solon’s results from balanced sample 1967 0.369 Ž . 0.094 0.409 Ž . 1968 0.396 0.093 0.431 Ž . Ž . 0.087 0.422 0.093 0.420 Ž . Ž . 1969 0.406 0.088 0.405 0.094 0.413 Ž . Ž . Ž . 0.085 0.382 0.090 0.397 0.093 Ž . Ž . 1970 0.309 0.089 0.374 0.090 Ž . Ž . 0.087 0.324 0.088 Ž . 1971 0.285 0.086 Ž . 0.078 c OLS results from unbalanced sample 1967 0.398 Ž . 0.076 w x 330 0.440 Ž . 1968 0.361 0.087 0.516 Ž . w x Ž . 0.078 324 0.097 w x w x 324 0.411 292 0.537 Ž . Ž . 1969 0.282 0.085 0.506 0.098 0.447 Ž . w x Ž . w x Ž . 0.068 292 0.093 286 0.098 w x w x w x 294 0.464 286 0.420 279 Ž . Ž . 1970 0.315 0.086 0.371 0.093 Ž . w x Ž . w x 0.070 288 0.088 279 w x w x 321 0.282 281 Ž . 1971 0.169 0.076 Ž . w x 0.062 313 w x 313 Ž . Table 2 continued Year of father’s Measures of father’s log earnings log earnings 1-Year 2-Year 3-Year 4-Year 5-Year measures measures measures measures measures d Solon’s results from unbalanced sample 1967 0.389 Ž . 0.094 0.440 Ž . 1968 0.359 0.096 0.450 Ž . Ž . 0.084 0.363 0.094 0.457 Ž . Ž . 1969 0.274 0.083 0.431 0.095 0.447 Ž . Ž . Ž . 0.070 0.413 0.090 0.420 0.098 Ž . Ž . 1970 0.336 0.087 0.394 0.093 Ž . Ž . 0.084 0.308 0.090 Ž . 1971 0.176 0.084 Ž . 0.066 Standard errors are in parentheses. Sample sizes are in brackets for the unbalanced samples, and are 290 and 280 for Solon’s and my balanced samples, respectively. instrument but may belong in the model for son’s status as a regressor. If that is Ž the case and if father’s education is positively related to his son’s earnings as is . likely , the IV estimates will be upwardly inconsistent. My estimates present the same picture as those in the work of Solon. Both sets of estimates imply that the correlation between son’s and father’s earnings lies in the neighborhood of 0.4. If the earnings of the sons do not reflect their socioeconomic status because of their youth, then examining them at older ages may provide more accurate estimates of intergenerational income mobility. I explore this possibility by using Ž . 4 sons’ outcomes 5 years later in 1989 to measure status. Table 3 reports the results using the 1989 son outcomes. When 1989 earnings data are used for the sons and a 5-year average of earnings for the fathers, the Ž . estimate of the intergenerational coefficient is 0.466 SE s 0.087 . Using the 1984 Ž . earnings data produced an estimate of 0.447 SE s 0.098 . The correlation esti- mates using the 1989 earnings are similar, but slightly greater than the correlations in father’s and son’s earnings using 1984 measures for sons. 5 When father’s 4 Between 1985 and 1990, 32 sons were lost from the sample because of attrition. Thus, the sample used here is not identical to that used to produce the estimates in Table 2. The sons who attrited tend to be those who had low 1984 earnings andror whose fathers reported low earnings in 1968–1972. For comparison, Appendix A presents the estimates using 1984 earnings and the sample still in the survey Ž . in 1990. Similar to the results in Fitzgerald et al. 1998 , the estimates of the intergenerational coefficient are slightly higher for the non-attriting sample, but the differences are not different from zero. As Fitzgerald et al. point out, some attrition will have occurred prior to 1984 and thus the estimates may already be biased. 5 Ž . This is consistent with Reville 1995 , which shows that the coefficients on fathers’ earnings increase with the ages of the sons. Table 3 Results using son’s 1989 earnings Ž . a Unbalanced sample Year of father’s Measures of father’s log earnings log earnings 1-Year 2-Year 3-Year 4-Year 5-Year measures measures measures measures measures 1967 0.397 Ž . 0.069 w x 298 0.460 Ž . 1968 0.364 0.074 0.498 Ž . w x Ž . 0.065 294 0.085 w x w x 294 0.376 262 0.512 Ž . Ž . 1969 0.248 0.074 0.469 0.086 0.466 Ž . w x Ž . w x Ž . 0.059 262 0.081 259 0.087 w x w x w x 264 0.450 259 0.430 254 Ž . Ž . 1970 0.380 0.074 0.408 0.082 Ž . w x Ž . w x 0.057 261 0.077 254 w x w x 294 0.354 256 Ž . 1971 0.231 0.064 Ž . w x 0.052 288 w x 288 Ž . b Balanced sample Year of father’s Measures of father’s log consumption log consumption 1-Year 2-Year 3-Year 4-Year 5-Year measures measures measures measures measures 1967 0.413 Ž . 0.088 0.459 Ž . 1968 0.349 0.089 0.461 Ž . Ž . 0.076 0.350 0.085 0.464 Ž . Ž . 1969 0.228 0.073 0.426 0.086 0.466 Ž . Ž . Ž . 0.062 0.398 0.081 0.430 0.087 Ž . Ž . 1970 0.349 0.078 0.401 0.082 Ž . Ž . 0.075 0.326 0.080 Ž . 1971 0.197 0.074 Ž . 0.059 Standard errors are in parentheses. Sample sizes are in brackets for the unbalanced sample. The sample size for the balanced sample is 254. earnings are instrumented for, the estimate of the intergenerational coefficient is Ž . 0.661 SE s 0.121 . Again, this estimate is upwardly biased if the instrument— father’s education—is positively correlated with son’s earnings. Using more recent data on the sons does not change the inference about the degree of mobility in the US. The correlation between fathers’ and sons’ earnings lies in the neighborhood of 0.4.

3. Results using consumption data