Heterogeneous Effects by Age and Education

Table 4 Linear Probability Model of Divorce Rate: Males Reference Ages 30 to 54 1 Medical Condition 2 Accident Back or Spine Other Accidental Not Accidental Prevented − 0.47 1.33 0.3 0.92 0.84 0.99 0.95 0.98 Year = −2,−1 2.31 − 0.3 1.33 0.17 1.3 1.2 1.25 1.2 Year = 0,1 3.89 0.4 2.28 1.3 1.4 1.3 1.4 1.3 Year = 2,3 − 1.04 0.15 0.31 − 0.8 0.85 1.4 1.4 1.2 Limited 1.48 1.18 1.59 1.02 1.0 0.81 0.97 0.85 Year = −2,−1 − 0.73 − 0.49 − 0.95 − 0.22 1.2 0.97 1.1 1.0 Year = 0,1 − 1.88 − 0.39 − 1.24 − 0.81 1.2 0.98 1.1 0.99 Year = 2,3 − 1.06 − 0.63 − 1.31 − 0.29 1.3 1.0 1.2 1.1 Observations 295,137 295,137 Data: Survey of Income and Program Participation, panels 1990 through 1996. Note: The unit of analysis is person and calendar year. The sample is conditioned on respondents who are married during the calendar year. The model includes controls for race, education, age, and marriage tenure and number. “Year” is relative to the reference year, which is the year disability onset for the disabled. Estimates are in percentage points. Standard errors, clustered by person, are in parentheses. , , and indicate significance at the 10, 5, and 1 percent level, respectively. The SIPP data also report whether the disabling condition was accidental. A simi- lar event-study model is estimated that differentiates between accidental and non- accidental disabilities and between work-preventing and work-limiting disabilities. The estimates are presented in Columns 2 of Table 4. The most notable estimates are for accidental disabilities that prevent work, reaching 2.28 percentage points in the year of and immediately following disability onset. Although the estimate is only significant at the 10 percent level, the result suggests that accidental disabilities have a greater effect on divorce.

D. Heterogeneous Effects by Age and Education

The next set of regressions estimate whether the effect of disability on divorce varies by age and education. According to the theoretical model above, the effect depends on the decline in expected marital value, which in turn depends on the incidence and severity of disability, and whether the decline is sufficient to trigger divorce. Table 5 Linear Probability Model of Divorce Rate: Males Reference Ages 30 to 54 Reference Age 30–44 30–44 45–54 45–54 Education High Low High Low Prevented − 1.2 − 0.23 3.88 − 0.24 1.1 1.5 2.0 1.2 Year = −2,−1 2.37 1.75 − 2.44 1.24 1.5 2.0 2.2 1.7 Year = 0,1 6.24 0.68 − 3.56 2.43 1.8 1.8 2.2 1.9 Year = 2,3 2.79 − 0.17 − 3.67 0.9 1.8 1.7 2.2 1.9 Limited 0.16 1.69 1.75 5.77 0.89 2.1 1.3 3.5 Year = −2,−1 1.00 − 1.78 − 1.43 − 6.96 1.2 2.4 1.5 3.5 Year = 0,1 0.41 − 1.13 − 1.91 − 5.89 1.1 2.4 1.4 3.7 Year = 2,3 0.57 0.7 − 1.04 − 6.91 1.2 2.8 1.6 3.5 Observations 295,137 Data: Survey of Income and Program Participation, panels 1990 through 1996. Note: The unit of analysis is person and calendar year. The sample is conditioned on respondents who are married during the calendar year. The model includes controls for race, education, age, marriage tenure and number, and nature of the work disability. “Year” is relative to the reference year, which is the year disability onset for the disabled. Estimates are in percentage points. Standard errors, clustered by person, are in parentheses. , , and indicate significance at the 10, 5, and 1 percent level, respectively. Given the same severity of disability, the effect of disability on divorce should be greater among younger males. This is because younger males have a lower incidence of disability, more work-years remaining, and presumably lower-quality marriages on average. Additionally, the effect should be greater among the more educated, who have greater expected earnings and a lower incidence of disability onset. To estimate heterogeneous effects by age and education, the index corresponds g to eight groups. The groups are differentiated by work-preventing and work-limiting disabilities, young and old ages, and low and high education. Young is defined as reference age 30 to 44; old is defined as 45 to 54. Low education is defined as no high school diploma; high education is defined as a high school diploma or more. To control for the types of disabling conditions across age and education groups, the index separately indicates back and spine conditions and accidental disabilities. g As discussed, these disability types are more likely to lead to divorce Table 4 and vary by and age education Table 2. Table 5 presents the estimates of the model. As shown, the association between disability onset and divorce is only significant among the young and educated who report work-preventing, rather than a work-limiting, disabilities. The increase in divorce is 6.24 percentage points and is statistically significant at the 1 percent level. These results are consistent with the theory presented above, as the young and educated have greater expected earnings, longer remaining work-years, and a lower incidence of disability onset. Table 5 presents other estimates that are statistically significant, most notably for low-educated, work-limited males ages 45 to 54. For this group, the divorce rate in year −3 is greater than the nondisabled, but the rate declines in subsequent years. Mentioned above, this downward trend may reflect that, as marriages dissolve, av- erage, unobserved marital quality increases, decreasing the divorce rate over time. Alternatively, the downward trend may simply reflect sampling noise, as the divorce rate is close to or at zero during most of the analysis period, except in year −3, when it is 7.1 percentage points.

E. Reverse Causation