Page 102 - GHES-1-1

Global Health Econ Sustain Effects of community-based activities on LTC needs

drinking, among this group was lower than that of people among older people living with family members; in
living with family members. The ratio of refraining from contrast, it was significantly positive among older adults
going out among older adults living alone was 0.319, living alone. We also noted a distinct association between
which was almost three times that of the comparison the prevalence of poor mental health and the logged LTCI
group. The survey of LTC insurance needs was conducted premium. A negative relationship was found between
prior to the spread of the COVID-19 virus in Japan, and poor mental health and the LTCI premium at the 10%
refraining from going out was not associated with the fear level among older people with other family members. In
of contagion. contrast, we found no systematic relationship between
About 23% of older adults who lived alone required the two variables among older adults living alone. These
care, double that of the control group (0.10). The ratio of results indicate that older adults living alone with a higher
those living in public housing among older adults living income may tend to have an unhealthy lifestyle.
alone was 0.07, which was much higher than that of the Poor mental health and being male were associated
comparison group. The results of the two-sample t-test with SAPH in both groups. Age squared had a negative
with unequal variances showed that the proportion of non- impact on SAPH among the two groups, and the effect of
taxed households among the two groups classified by living age on SAPH seemed to decrease beyond 80 years among
in public housing was statistically different at the 1% level. older adults living alone.
The proportion of non-taxed among older adults living Although the likelihood ratio test for over-dispersion
alone in public housing was 0.833, which was higher than is recommended (Mahmood, 2020) before the Vuong

that of the control group (Appendix, Section B). test (Vuong, 1989), for checking the zero-inflation in the

For people cohabiting with family members, the data, the estimate of the size parameter becomes too small
results of the two-sample t-test with unequal variances when the negative binomial regression model is fitted to
also showed that the proportion of requiring care among data with excess zeros (Minami & Lennert-Cody, 2013).

the two groups classified by LTC taxation categories Therefore, in this study, we compared the Poisson model
(non-taxed versus taxed) were statistically different at the with the ZIP model when selecting an appropriate model.
1% level. Despite cohabitation, this suggests that older The Vuong statistic follows asymptotically normal
adults in the lower-income group tend to need formal or distribution under the null hypothesis that the standard
informal care. count model is preferred when V > 1.96 at the 5%
In contrast, the mean SAH of older adults living alone significant level. The null hypothesis of the Vuong test is
(2.92) was slightly lower than that of people cohabiting that the standard model and zero-inflated model fits the
with other family members. The respondents’ mental data equally well (Vuong, 1989). Therefore, both models

health was measured using two questions: (i) “Have you are assumed equivalent when |V| < 1.96.
been depressed during the past month?” (ii) “Have you The Akaike information criterion (AIC), the Schwarz
not been happy for some time during the past month?” Bayesian criterion (SBC), and the corrected Vuong test
Respondents were classified as having poor mental health statistic were used when selecting regression models
if their answer to both questions was “yes.” The prevalence (Appendix, Section C). For older adults living alone or
of poor mental health among older adults living alone was cohabiting, the AIC and SBC values of the ZIP models
0.16, which was slightly higher than that among people were lower than those of the Poisson models. The results
cohabiting with other family members. of the AIC-corrected Vuong test imply that the ZIP model
Table 3 shows the results of the IV probit estimation. can be accepted (Figure 2).
The results of the weak instrument robustness test Table 4 presents the estimation results of the ZIP
(AR test) for IV probit models indicated that the null models. Both generalized residuals and the interaction
hypothesis of zero correlation coefficient between the term of the generalized residuals and LTC insurance
error terms of SAPH and LTC insurance premiums was premium categories in the two equations are not
not rejected. All the estimated models indicated a zero- statistically significant at the 5% level. Accordingly,
correlation coefficient between the error terms at the 5% the precise impact of SAPH on requiring care can be
level (see “ath rho” in Table 3). However, the exogeneity measured. Among older adults living alone, the impact of
condition was not satisfied when a dummy variable for SAPH on requiring care was five times larger than in the
past history of CVD was added.
control group. It can be concluded that maintaining good
We noted the sign of former smokers using an subjective health is most important for older adults living
instrumental variable. The sign was significantly negative alone to prevent requiring care.

Volume 1 Issue 1 (2023) 5 https://doi.org/10.36922/ghes.0891