Appendix: Detailed Methodology
Current Population Survey data from IPUMS provide information on earnings, education, and age that allow us to estimate a Mincer equation, which is a regression model that explains an individual’s earnings using demographic and economic variables.
Ln(Yi) = B1 NonHSi + B2 SomeCollegeYearsi + B3 Associatei + B4 Bachelori + B5 Advancedi + A Xi + ei
We use dummy variables to define educational attainment for those with less than high school (NonHS), associate’s degrees (Associate), bachelor’s degree (Bachelor), and advanced degrees (Advanced). For those with some college but no degree, we include a measure of the number years of school completed (SomeCollegeYears). We also include dummy variables for each single-year age group in the vector Xi that controls for differences in earnings due to age and experience.
A second model is estimated that includes controls for detailed occupations. This helps to reduce the bias in the estimates that come from capturing underlying ability differences versus changes in human capital by focusing on differences in earnings for people within the same occupation. If the differences in earnings between occupations are driven by unmeasured ability rather than by the differences in human capital attained through education, than this estimate will be less biased by focusing on differences within occupation only. However, one of the important ways that education increases productivity is by allowing entry into higher-skill occupations. To balance the risks of over- versus under-controlling, we average the results from the models with and without occupation controls (see Appendix Table).
Based on coefficients from those models, we can estimate the higher earnings that result from going from some college to completing a bachelor’s degree. On average, those with some college have completed 2.6 years of college. By comparing the coefficients B2 and B4, we can estimate the change in earnings that can be expected from going from 2.6 years of college to earning a degree. Specifically, we can estimate that it is B4 - (2.6 x B2).
Table: Earnings and Employment Models
| Variable (first row: coefficent; second row: p-value) |
Earnings model 1 | Earnings model 2 | Avg earnings model | Employment model 1 | Employment model 2 | Avg employment model |
| Less than high school | -0.231 | -0.205 | -0.218 | -0.103 | -0.011 | -0.057 |
| 0.000 | 0.000 | 0.000 | 0.000 | |||
| Yrs of college for those with some college | 0.036 | 0.024 | 0.030 | 0.016 | 0.000 | 0.008 |
| 0.000 | 0.000 | 0.000 | 0.000 | |||
| Associate’s degree | 0.136 | 0.083 | 0.109 | 0.082 | 0.006 | 0.044 |
| 0.000 | 0.000 | 0.000 | 0.000 | |||
| Bachelor’s degree | 0.475 | 0.309 | 0.392 | 0.116 | 0.013 | 0.065 |
| 0.000 | 0.000 | 0.000 | 0.000 | |||
| Advanced degree | 0.654 | 0.431 | 0.542 | 0.144 | 0.007 | 0.076 |
| 0.000 | 0.000 | 0.000 | 0.000 | |||
| Constant | 5.479 | 6.040 | 0.329 | -0.000 | ||
| 0.000 | 0.000 | 0.000 | 0.001 | |||
| Adjusted R-squared | 0.629 | 0.680 | 0.803 | 0.929 | ||
| Sample (2014–2016) | 503,363 | 503,363 | 852,267 | 852,267 | ||
| Dependent variable | ln(earnings) | ln(earnings) | ln(earnings) | Employed = 1 | Employed = 1 | Employed = 1 |
| Age fixed effects | X | X | X | X | ||
| Occupation fixed effects | X | X | ||||
Source: Moody’s Analytics