Page 109 - GHES-1-1

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

econometric modeling. Journal of Health Economics, [Nihon Koshu Eisei Zasshi] Japanese Journal of Public Health,
27(3):531-543. 69(3):215-224. [Article in Japanese].
https://doi.org/10.1016/j.jhealeco.2007.09.009 https://doi.org/10.11236/jph.21-060
Vuong, Q.H. (1989). Likelihood ratio tests for model selection Wood, A.M., Kaptoge, S., Butterworth, A.S., Willeit, P.,
and non-nested hypotheses. Econometrica, 57(2):307-333. Warnakula, S., Bolton, T., et al. (2018). Risk thresholds for
https://doi.org/10.2307/1912557 alcohol consumption: Combined analysis of individual-
participant data for 599 912 current drinkers in 83
Watanabe, S., Murayama, H., Takase, M., Sugiura, K., &
Fujiwara, Y. (2022). Longitudinal association between work prospective studies. Lancet, 391(10129):1513-1523.
and self-rated health in older adults: A systematic review. https://doi.org/10.1016/S0140-6736(18)30134-X

Appendix

Section (A) Zero-inflated Poisson model
When overdispersion is generated by excess zeros, we must use zero-inflated count models, as they allow for excess zeros in
the data by modeling the counts as a mixture of two distributions: a spike at zero and a distribution of positive outcomes.
These two separate types of zeros are known as zero-inflated count models(Lambert, 1992).

ZIP models assume that the population consists of two groups of people with varying probabilities Φi and (1−Φi). This
model can be interpreted as a finite mixture model that includes a degenerate distribution with a mass point at zero. In the
ZIP regression of requiring care,

y = 0 with probability iΦ + (1−Φ i )e −λ i ,
i

e −λ i λ k
y = k with probability (1−Φ ) i ! k i , k = 1,2
i
Where λ is the intensity parameter that represents the expected number of occurrences in a fixed period.
i

Volume 1 Issue 1 (2023) 12 https://doi.org/10.36922/ghes.0891