Medical consumption over the life-cycle

Abstract

We investigate the association between age and medical spending in the U.S. using data from the Medical Expenditure Panel Survey. We estimate a partially linear seminonparametric model and construct “pure” life-cycle profiles of health spending simultaneously controlling for time effects (i.e., institutional changes and business cycles effects) and cohort effects (i.e., generation specific conditions). We find that time and cohort effects together introduce a significant estimation bias into predictions of health expenditures per age group, especially for individuals older than 60 years. The estimation bias introduced by cohort effects increases monotonically with age while the bias due to time effects is not significant. The overall effect of Medicare on the cohort and time effects biases is negligible.

Appendix

See Tables 1, 2, 3 and Figs. 1, 2, 3, 4, 5, 6, 7, 8, and 9.

Table 1 Summary statistics of the pooled data: MEPS 1996–2007

Table 2 Summary statistics of the pseudo panel data: MEPS 1996–2007

Table 3 Frequencies per cohort and year: MEPS 1996–2007

Fig. 1

Stylized facts from cross section summary data. Source MEPS 1996–2007

Fig. 2

Cross section of health expenditure using a constructed pseudo panel. Source MEPS 1996–2007

Fig. 3

Health expenditure profiles controlling for time and cohort effects, including bootstrapped confidence intervals. Source MEPS 1996–2007

Fig. 4

Health expenditure profiles controlling for time and cohort effects. Cohort effect bias: $\Delta _{\text {cohort}} = ({\text {Age}}+{\text {cohort}}+{\text {time}}) - ({\text {Age}}+{\text {time}})$ from panel 1. Time effect bias: $\Delta _{\text {time}} = ({\text {Age}}+{\text {time}}) - ({\text {Age}})$ from panel 1. The dotted lines are $95$ % confidence intervals. We do not report confidence intervals for the time effects bias as this bias is insignificant over the entire age range. Source MEPS 1996–2007

Fig. 5

Out-of-pocket health expenditure profiles controlling for time and cohort effects. Cohort effect bias: $\Delta _{\text {cohort}} = ({\text {Age}}+{\text {cohort}}+{\text {time}}) - ({\text {Age}}+{\text {time}})$ from panel 1. Time effect bias: $\Delta _{{\text {time}}} = ({\text {Age}}+{\text {time}}) - ({\text {Age}})$ from panel 1. The dotted lines are $95$ % confidence intervals. We do not report confidence intervals for the time effects bias as this bias is insignificant over the entire age range. Source MEPS 1996–2007

Fig. 6

Out-of-pocket health expenditure profiles controlling for time and cohort effects by gender. Cohort effect bias: $\Delta _{\text {cohort}} = ({\text {Age}}+{\text {cohort}}+{\text {time}}) - ({\text {Age}}+{\text {time}})$ from panel 1 or 2. Time effect bias: $\Delta _{{\text {time}}} = ({\text {Age}}+{\text {time}}) - ({\text {Age}})$ from panel 1 or 2. The dotted lines are 95 % confidence intervals. We do not report confidence intervals for the time effects bias as this bias is insignificant over the entire age range. Source MEPS 1996–2007

Fig. 7

Out-of-pocket health expenditure profiles controlling for time and cohort effects by skill level. Low skilled individuals have up to $12$ years of education. High skilled individiuals have more than $12$ years of education. Cohort effect bias: $\Delta _{\text {cohort}} = ({\text {Age}}+{\text {cohort}}+{\text {time}}) - ({\text {Age}}+{\text {time}})$ from panel 1 or 2. Time effect bias: $\Delta _{{\text {time}}} = ({\text {Age}}+{\text {time}}) - ({\text {Age}})$ from panel 1 or 2. The dotted lines are $95$ % confidence intervals. We do not report confidence intervals for the time effects bias as this bias is insignificant over the entire age range. Source MEPS 1996–2007

Fig. 8

Health expenditure and out-of-pocket health expenditure profiles controlling for time and cohort effects by insurance status. We only distinguish between insured and uninsured individuals amongst the working population. Almost all individuals older than 65 are insured via Medicare. Source MEPS 1996–2007

Fig. 9

Health expenditure and out-of-pocket health expenditure estimation biases by insurance status. We only distinguish between insured and uninsured individuals when they are younger than 65, since almost all individuals older than 65 are insured via Medicare. Cohort effect bias: $\Delta _{\text {cohort}} = ({\text {Age}}+{\text {cohort}}+{\text {time}}) - ({\text {Age}}+{\text {time}})$. Time effect bias $\Delta _{{\text {time}}} = ({\text {Age}}+{\text {time}}) - ({\text {Age}})$. The dotted lines are $95$ % confidence intervals. We do not report confidence intervals for the time effects bias as this bias is insignificant over the entire age range. Source MEPS 1996–2007