ZVI ECKSTEIN and OSNAT LIFSHITZ
Data and Estimation (section 3) – PSID Data
The data is based on the PSID survey (Panel Study of Income Dynamics) during the years 1983-1993. The yearly source files were downloaded from http://psidonline.isr.umich.edu. From each file we kept and/or calculate the model variables, using the following do file (create85.do).
We choose to use quarterly data which is available only from 1983 (the data is available on a monthly frequency, and we created the quarterly data using the do file: mon_to_q.do). We restrict the model for the first ten years since marriage. In order to give similar initial conditions to all individuals, we restrict the data, as in the model, to start at the date of marriage and we consider all married couples during the years 1983-1984. The file (PSID.dta) contains details on 863 couples and follows them until 1993 or until they are separated. 36.3% of the couples are divorced or separate during the sample period; 14.5% leave the sample from other reaons, such that after 10 years 49.2% of the couples remain in the sample (Descriptive statistics and reduced form estimation of the model’s equation, including tables 1 and 2 in the paper).
The data contains individual and household demographic information and labor supply, such as, wage, working hours, unemployment and non-participating in the labor force. The participation rate of women in the sample is 72% in 1984, and climbs up to 79% after 10 years of marriage. Note that the participation rate of married women, aged 25-55, in the year 1984 was 66% according to CPS data. The unemployment rate falls from 6% to 3.1% during those years. The participation rate of men in the sample is 93% and constant in time. Note that using the CPS data this rate among men aged 25-55 in the year 1984 was 93.9%. The unemployment rate decreased from 9% in 1984 to 3.6% in 1993. In addition, the income data and working hours in the sample is comparable to that in the CPS.
The correlation between the labor force participation of a wife to her husband is negative and significant for families with the following demographics: husbands have less than 12 years of schooling; black husbands; protestant husband; families in rural areas. This is consistent with the model prediction for classical families, which is what the above demographics indicate. On the other hand, the correlation between the labor force participation of a wife to her husband is positive but not significant for families with the following demographics: husband has more than 12 years of schooling; white husband; Catholic husband; live in large city.
Estimation Results (section 4)
| · Estimated parameters | (including Tables 3, 4, 5, 6 and 7 in the paper) | |
| · Model fit | (including Figures 1 and 2 in the paper) | |
| · Classical, Modern and Cooperative prediction | (including Figure 3A and 3B in the paper) | |
| · Chi-square tests | ||
| · Excluded Tables· Fertility rate – Actual and Simulated· Divorce rate – Actual and Simulated |
Counterfactuals (section 5)
| · Counterfactuals | (including Figures 4, 5, 6, 7 and 8 in the paper) |
CPS Data, variables and sample restriction for Table 1
Data was taken from the Annual Demographic Surveys (March CPS supplement) conducted by the Bureau of Labor Statistics and the Bureau of the Census. This survey is the primary source for detailed information on income and work experience in the United States. A detailed description of the survey can be found at www.bls.gov/cps. Our data, for the years 1962-2007, was extracted using the Unicon CPS utilities.
The sample is restricted to civilian adults, ignoring armed forces and children. We divided the sample into five education groups: high school dropouts (HSD), high school graduates (HSG), individuals with some college (SC), college graduates (CG) and post-college degree holders (PC). In order to construct the education variable, until 1991 we used the years of schooling completed and added 0.5 years if the individual did not complete the highest grade attended and from 1992 we used years of schooling as is.
Weekly wages are constructed by taking the previous year’s wage and salary income and dividing it by the number of weeks worked in the previous year. Hourly wages are defined as the weekly wage divided by the number of hours worked in the previous week in all jobs, while annual (annualized) wages are defined as the weekly wage multiplied by 52. Wages are multiplied by 1.75 for top-coded observations until 1995. Nominal wages are deflated using the Personal Consumption Expenditure (PCE) index from NIPA table 2.3.4 (http://www.bea.gov/iTable/iTable.cfm?ReqID=9&step=1). Since wages refer to the previous year, we use PCE for year X-1 for observations in year X and therefore all wages are expressed in constant 2006 dollars.
Information on number of children under six for the period 1968 – 1975, which is missing from the survey data, is completed where possible using the distribution of this variable in 1967 and 1976 for each gender, marital status and cohort separately. The completed information can be used to construct an aggregate trend, but not to identify the number of children for a specific individual.
In order to construct couples, we kept only heads of households and spouses (i.e. no secondary families were used) and dropped households with more than one male or more than one female. We then merged women and men based on year and household id and dropped problematic couples (with two heads or two spouses, with more than one family or with inconsistent marital status or number of children).
· code for constructing the file used in the paper
