5. Results
5.2 DID Approach 1: Maternal Age
section of the paper uses a difference-in-differences approach estimate the differential effects of NHI implementation on women of different ages and in different treatment groups.
5.2 DID Approach 1: Maternal Age
Results of the difference-in-differences estimators for regression using maternal age to create a cutting point are reported in Table 5. For the latest 2016 WMFES survey wave, respondents born in 1971 would have reached age 45 and likely completed childbearing, therefore we have limited the sample to women born between 1948 and 1971, aged 24 to 47 years in 1995 during NHI implementation. The government group is comprised of all women where either the respondent or her husband (or both) are a government employee, and conferring the increased childbirth and child healthcare benefits of state sponsored insurance to the
respondent prior to NHI implementation. The non-government treatment group includes all other women in the sample. The pre-treatment group is born at or before the cutoff age, and the post-treatment group born afterwards. We first test 35 years old as the maternal cutoff age, under the logic that women 35 or older were less likely to have future fertility decisions affected by NHI implementation, since 35 years is the minimum age for advanced age maternal benefits under
Table 5: Difference in Differences Estimation
35 year old cutoff 40 year old cutoff 45 year old cutoff DID estimator (𝛿𝛿) -0.79***
Notes: All figures are significant at the 1% level, p < .01, coefficients from Eqn. 2.
‧
NHI. We further use 40 years and 45 years as additional cutoff points to test the robustness of regression results.
Prior to NHI intervention, we expect to see higher fertility in the group covered by government insurance, or if fertility is higher in the non-government group initially, to see a larger positive percentage change in fertility after NHI. Regression results in Table 6 for the 35-year-old cutoff group show that prior to NHI intervention, we see 17.01% higher fertility in the non-government group than the government group. After NHI intervention, however, expected fertility of the non-government group is 37.30% lower than that of the government group. This finding contradicts our hypothesis that NHI intervention would have stimulated childbearing in the newly insured non-government respondents, and actually shows the opposite effect since after NHI the state-insured group had higher fertility than the non-government group.
Table 6: DID Results for 3 Cutoff Ages (35, 40, and 45 years old)
Pre-treatment Post-treatment Cutoff age: 35 years old
Fertility 1.94 2.27 17.01% 1.26 0.79 -37.30%
Sample size 17,139 105,947 10,051 117,741
Cutoff age: 40 years old
Fertility 2.24 2.68 19.64% 1.40 1.09 -22.14%
Sample size 9,492 56,500 17,698 167,188
Cutoff age: 45 years old
Fertility 2.55 2.99 17.25% 1.61 1.39 -13.66%
Sample size 2,311 13,648 24,879 210,040
Notes: All figures are significant at the 1% level, p < .01
After NHI implementation, regressions for all three cutoff ages (35, 40, and 45) reveal lower fertility in the non-government respondents in comparison to the government group
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(22.14% lower for the 40 year old cutoff, and 13.66% lower for the 45 year old cutoff), though the magnitude of the percentage change is slightly different. These results contradict the hypothesis that NHI implementation may been seen as a successful pronatal policy that stimulated fertility by lowering cost barriers to childbearing. That the percentage change
decreases as the cutoff age is raised, on the other hand, indicates the wider the range of potential childbearing years, the lesser the difference between the government and non-government groups. Between the 35 and 45 years cutoff age regressions, the percentage change between government and non-government changes decreases by 23.64%. However, the R-squared value also drops from 0.23 in the 35 years old cutoff regression to 0.07 in the 45 years old regression, exhibiting negligible explanatory power when the broadest estimation parameters are used.
DID results using the 35-year old cutoff point, and controlling for income and education (and without limiting respondent birth years), are presented in Table 7. As in the OLS regression, education was stratified into three groups, (1) no formal education to elementary school, (2) some middle to high school, and (3) a post-secondary or post-graduate education level. Similarly, income was bracketed into three groups. The lowest monthly income was 0 - 39,999 NT, the middle income group was 40,000 – 69,999 NT, and the highest group had a monthly income of more than 70,000 NT. The highest magnitude DID estimators were in the low education group models, in the low monthly income (-0.94) and high monthly income (-1.32) cells. Both of these cells evidenced statistically significant, negative relationship across the time and treatment groups, yet the explanatory power of the regressions were fairly low, at only 7% and 14%, respectively. The lowest magnitude estimator was in the low education, middle income cell; this was also the only model that showed a nonnegative DID coefficient in regression, at 0.03. This
‧
result was not found to be statistically significant at the 10% level, however, possible due to a low sample size of only 3,077 observations.
Generally speaking, the DID estimators for those with the highest level of educational attainment were less negative/smaller in magnitude than those for other education groups. This result may indicate that both pre- and post-treatment, both non-government treatment and government control groups who were highly educated were more likely to restrict their fertility, so that the new NHI had less of an effect on childbearing decisions. The low income, high education cell had the strongest explanatory power of all groups (R2 = 0.45), yet the DID coefficient did not differ significantly from zero. In sum, the results of DID regression stratified by income-education cell support the DID results from Table 5. Although the DID coefficients
Table 7: Difference in Differences Results by Income-education Cell Low monthly
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
were smaller in magnitude than those found in the collapsed income/education groups for the 35, 40, and 45 year cutoff regressions, the direction of relationship seems consistently negative robust to controlling for income and education.