Results using consumption data

3. Results using consumption data

Consumption can be argued to be superior to earnings as a proxy for permanent status. 6 In this sample, consumption will be a better measure of status than earnings if sons who will eventually have high earnings receive intergenerational transfers that augment their earnings or are able to borrow against their future Ž . earnings in their early years i.e., via mortgages . Children whose parents have higher incomes and more wealth are more likely to receive transfers and these transfers are compensatory, that is, the less well-off children in a family receive Ž . them Altonji et al., 1992; McGarry, 1997; McGarry and Schoeni, 1996, 1997 . The PSID does not contain a complete measure of household or individual consumption, but does provide data on family spending on food eaten at home, food eaten out, rent, and the value of owned housing. These components of consumption are used to construct several proxies of status. The proxies are the sum of food expenditures, weighted sums of the available components of con- sumption, and a predicted value of earnings. 7 I treat family size and structure in Ž . Ž . three ways: 1 no adjustment is made, 2 the proxies are divided by household Ž . size, and 3 the proxies are divided by adult equivalents using a scale where each adult equals 1 and each child equals 0.4. 8 Ž . The weights for the weighted sums are from Skinner 1987 . Skinner estimates the weights by regressing total consumption on the four categories of spending Ž . using data from the Consumer Expenditure Survey CEX . He estimates two sets of weights, one from the 1972–1973 CEX and one from the 1983 CEX. I use three combinations of the weights to estimate the degree of intergenerational mobility: Ž . 1 the 1972–1973 weights are used to construct the proxies for fathers and sons w x Ž . FS7273 , 2 the 1983 weights are used to build the proxies for fathers and sons w x Ž . FS83 , and 3 the 1972–1973 weights are used to build the fathers’ proxy and the w x 1983 weights are used to build the sons’ proxy F7273S83 . Additionally, a predicted value of permanent status is constructed using the available data on consumption. To construct the predicted value, earnings are regressed on the four components of consumption. 9 The predicted values contain Ž . the permanent portion of earnings. Dearden et al. 1997 show that when predicted values are used to estimate the intergenerational coefficient, the results may either over- or under-estimate the true value depending on whether the unobserved 6 Ž . For instance, Attanasio and Browning 1995 find that after controlling for family composition, the age profile for consumption is Aremarkably flat.B 7 Ž . The sum of food expenditures is used as a proxy for consumption in Altonji et al. 1992 . The Ž . Ž . weighted sum is used to proxy for consumption by Becker and Mulligan 1997 and Mulligan 1997 . Ž . Dearden et al. 1997 use predicted values of earnings to proxy for permanent economic status. 8 Ž . This adult equivalent scale comes from the work of Lazear and Michael 1988 . 9 The predicted earnings regressions are estimated separately for each year of data. The results are available upon request. portion of permanent status is more or less correlated across generations than the portion that is captured by the predicted value. Ž . Mulligan 1997 was the first to estimate the relationship between father’s and son’s consumption levels. Although the criteria used to draw his samples are less restrictive than those used here and he does not account for family size, he uses the PSID and one of the measures of consumption that he uses is one of the Ž . weighted sums that is used here F7273S85 . Mulligan’s OLS estimates of the Ž correlation between fathers’ and sons’ consumption levels range from 0.53 SE s . Ž . 0.04 to 0.56 SE s 0.04 . Table 4 presents the OLS estimates of the intergenerational coefficient when the various consumption measures are used to proxy for the sons’ and fathers’ statuses and when different methods are used to adjust for household size and structure. Columns 1–3 use 1984 measures for the sons’ outcomes. The first column presents the average estimates where the status of fathers is measured Ž . using a single year of data 1967–1971 and the unbalanced sample is used. The second column reports the average correlation estimate when fathers’ status is measured in a single year and the balanced sample is used. The third column of Table 4 reports the estimates when the father’s status is proxied by a 5-year average of the measure of his status. Columns 4–6 are analogous to columns 1–3 but use the 1989 measures of the sons’ outcomes. In Table 4, the top five rows present the results when the proxies used are divided by household size. The second five rows divide the proxies by a household equivalent scale to adjust for household size and structure. The bottom five rows do not account for household size or structure. The estimates of intergenerational correlation in columns 1–3 of Table 4 point to greater persistence in status and well-being than was implied by the earnings data. When the food sum is used to proxy for socioeconomic status, the average correlation estimates using a single year of data on fathers are approximately 0.5 and 0.6 for the unbalanced and balanced samples across the different methods of treatment for household size and structure. Using the weighted sum of the four components of consumption produces estimates of the correlation between fathers’ and sons’ statuses that are considerably higher. The single-year estimates using the weighted sums as proxies for permanent status range from 0.689 to 0.934 for the unbalanced sample and 0.691 and 0.891 for the balanced sample. The correlation estimates using the predicted value to proxy for status are lower, with the correlation estimates ranging from 0.423 to 0.595 for the unbalanced sample and from 0.506 to 0.565 for the balanced sample. Using the average outcomes of the fathers decreases the measurement error associated with the consumption measures and hence increases the estimates of the correlation between the status of fathers and sons. When the 5-year averages are used to proxy for father’s status, the correlation estimates are in the neighborhood of 0.7–0.8 for food sum, near 1.0 for the weighted sums, and around 0.6 for the predicted value. Table 4 OLS estimates of intergenerational correlation using consumption proxies for permanent status Proxy for 1984 1989 permanent Average Average 5-Year Average Average 5-Year status of 1-year of 1-year estimate of 1-year of 1-year estimate estimate, estimate, estimate, estimate, unbalanced balanced unbalanced balanced sample sample sample sample DiÕided by household size Food sum 0.511 0.526 0.669 0.348 0.310 0.403 Ž . Ž . Ž . Ž . Ž . Ž . 0.122 0.132 0.140 0.105 0.107 0.113 FS7273 0.689 0.691 0.779 0.580 0.588 0.674 Ž . Ž . Ž . Ž . Ž . Ž . 0.099 0.107 0.107 0.091 0.099 0.099 FS85 0.821 0.794 0.952 0.562 0.569 0.657 Ž . Ž . Ž . Ž . Ž . Ž . 0.149 0.163 0.161 0.092 0.101 0.100 F7273S83 0.843 0.816 0.977 0.587 0.593 0.684 Ž . Ž . Ž . Ž . Ž . Ž . 0.149 0.163 0.161 0.094 0.103 0.102 Predicted value 0.595 0.565 0.605 0.466 0.468 0.506 Ž . Ž . Ž . Ž . Ž . Ž . 0.131 0.147 0.117 0.085 0.101 0.079 DiÕided by household equiÕalent scale Food sum 0.519 0.576 0.784 0.352 0.328 0.473 Ž . Ž . Ž . Ž . Ž . Ž . 0.126 0.145 0.159 0.109 0.107 0.116 FS7273 0.863 0.874 1.052 0.645 0.658 0.781 Ž . Ž . Ž . Ž . Ž . Ž . 0.124 0.136 0.141 0.082 0.093 0.097 FS85 0.855 0.864 1.054 0.624 0.637 0.766 Ž . Ž . Ž . Ž . Ž . Ž . 0.127 0.141 0.146 0.083 0.095 0.098 F7273S83 0.882 0.891 1.085 0.654 0.666 0.799 Ž . Ž . Ž . Ž . Ž . Ž . 0.127 0.141 0.146 0.085 0.097 0.101 Predicted value 0.547 0.557 0.586 0.484 0.487 0.537 Ž . Ž . Ž . Ž . Ž . Ž . 0.113 0.129 0.106 0.078 0.095 0.076 No adjustment for household size or structure Food sum 0.451 0.568 0.811 0.327 0.360 0.585 Ž . Ž . Ž . Ž . Ž . Ž . 0.140 0.159 0.182 0.119 0.110 1.124 FS7273 0.934 0.823 1.010 0.716 0.754 0.956 Ž . Ž . Ž . Ž . Ž . Ž . 0.110 0.127 0.131 0.087 0.100 0.102 FS85 0.720 0.812 1.012 0.689 0.726 0.942 Ž . Ž . Ž . Ž . Ž . Ž . 0.137 0.131 0.137 0.087 0.097 0.103 F7273S83 0.747 0.839 1.047 0.720 0.757 0.979 Ž . Ž . Ž . Ž . Ž . Ž . 0.112 0.132 0.137 0.089 0.104 0.106 Predicted value 0.423 0.506 0.526 0.504 0.526 0.617 Ž . Ž . Ž . Ž . Ž . Ž . 0.121 0.141 0.114 0.086 0.102 0.080 Ž . Standard errors SE are in parentheses. Sample sizes vary in the unbalanced sample. The sample size for the balanced and the 5-year samples is 254. As was done for earnings, the correlation in the consumption-levels is estimated when son’s 1989 consumption is used to construct the proxies for his status. These results are presented in columns 4–6 of Table 4. The correlation estimates from consumption are substantially lower when the 1989 measures vs. the 1984 measures are used to proxy for sons’ status. When the food sum is used to measure status, the estimates of the correlation between sons’ outcomes and their fathers’ average outcomes over the years 1967–1971 lie near 0.4. When the weighted sums are used to measure status, the estimates of the intergenerational coefficient range from about 0.7–1.0. When the predicted values are used to measure status, the estimates of the correlation are in the neighborhood of 0.5. When either the sons’ 1984 or 1989 measures are used, the estimates where consumption is divided by family size are the smallest. The estimates where no account of family size is taken are greater than those that use an adult equivalent scale. When the sons’ 1989 measures are used and when either the household equivalent scale is used or the measures are divided by family size, the estimated correlations between fathers’ and sons’ consumption proxies are smaller and more similar to the estimates from the earnings data. This is not the case when 1984 consumption measures are used. Additional research is required to explain why the estimates using consumption data lie above the estimates that use earnings to measure economic status. The higher transmission of consumption measures may be a result of the transmission of saving behavior or intra-familial transfers. Alternatively, the consumption measures used here may overstate the transmission of status if the intergenera- tional correlation in one of the components of the consumption proxies exceeds the degree of transmission of consumption overall. This paper does not examine these issues.

4. Conclusion