Behavioral Health Risk Factors: the Interaction of Personal and Country Effects

Abstract

Purpose

This study investigated the relationship between the individual’s self-assessed health status (SAHS) and health-risk factors (smoking, alcohol consumption and obesity), in 16 European countries. The associations were studied for the individual and for the country measures—and in particular, for the unexplored aspect of interaction between individual and country levels of the three risk factors.

Method

Data for 47,114 adults, who participated in the Survey of Health Aging and Retirement Europe (SHARE), were analyzed using Multilevel Regression Analysis. The individual data were complemented by OECD data that provided country-specific risk measures: percentage of daily smokers, annual per-capita consumption of alcohol (liters), and percentage of obese individuals.

Results

We found that the individual’s SAHS is negatively associated with smoking and with weight-risk factors and is positively associated with her/his alcohol consumption. The most pronounced associations relate to the weight variables, albeit they are attenuated in countries with higher percentages of obese individuals. Significant differences across countries were evidenced in the association between SAHS and smoking and between SAHS and alcohol consumption.

Conclusion

Individual health levels are associated with individual risk factors and also with the behaviors in the country. Significant interactions might indicate that psychological factors are at work.

Appendices

Appendix 1. Formal presentation of the econometric models

A two-level structure is considered, where individuals (i, first level) are nested into countries (c, second level).

As an illustration, and for simplicity of presentation, we will include in the formal model (presented below) only one of the individual risk factors: “obesity of grade 3.” In a similar manner, we can add the other risk factors.

Let SAHS _ic denote the self-assessed health status of individual i in country c. O _ic is a dummy variable that equals one if individual i in a country c has obesity of grade 3 (and zero otherwise), and the vector X _ic contains the socio-economic characteristics of the individual.

When only variables of the first level are considered and random effects are included only in the intercept, we arrive at model 1 specification:

$$ {SAHS}_{ic}=\left({\gamma}_0+{U}_{0c}\right)+{\gamma}_1{O}_{ic}+{\gamma}_2{X}_{ic}+{R}_{ic} $$

(1)

where U_0c represents the random intercept at the country level, and R _ic corresponds to the random-error at the individual level. The inclusion of a random intercept is based on the assumption that the countries differ with respect to the average value of the SAHS variable (γ₀).

We now extend model 1 to answer the next interesting question: Are the associations between subjective health and “grade 3 obesity” different by country?

We use model 2 specification to answer this question, allowing U_1c as the random-slope (coefficient) of the dummy variable that denotes individual obesity of grade 3:

$$ {SAHS}_{ic}=\left({\gamma}_0+{U}_{0c}\right)+\left({\gamma}_1+{U}_{1c}\right){O}_{ic}+{\gamma}_2{X}_{ic}+{R}_{ic} $$

(2)

The inclusion of this term in the equation means that the regression coefficient of O _ic is country-dependent. The random effect U_1c is not a statistical parameter and thus it is not estimated as part of the estimation routine. However, it can be predicted. Using the estimate of γ₁ and the prediction of the vector U_1c, the country-specific coefficients of the grade 3 obesity variable could then be calculated.

Sticking to the illustration that relates to the one risk factor of obesity, the country-specific obesity level will be added, i.e., the percentage of the population (aged 15 years and over) who are obese (centered). Let PO _c denote this percentage in country c and $\overline{PO}$ the overall average for all countries included in our analysis. Model 3 specification will now be

$$ {SAHS}_{ic}=\left({\gamma}_0+{U}_{0c}\right)+\left({\gamma}_1+{U}_{1c}\right){O}_{ic}+{\gamma}_2{X}_{ic}+{\gamma}_3\left({PO}_c-\overline{PO}\right)+{R}_{ic} $$

(3)

A significant coefficient γ₃ indicates a significant association between the prevalence of obesity in the country and the subjective health of residents.

While the study of the relationship between country-level measures and health is not new and can be found in the epidemiological literature, the more innovative and interesting question is

Are individual-level and country-level risk factors independent or inter-related?

To answer this question, model 4 specification includes an interaction term between the individual risk factor, O _ic , and the country-specific risk prevalence, PO _c (its deviation from the overall average):

$$ {SAHS}_{ic}=\left({\gamma}_0+{U}_{0c}\right)+\left({\gamma}_1+{U}_{1c}\right){O}_{ic}+{\gamma}_2{X}_{ic}+{\gamma}_3\left({PO}_c-\overline{PO}\right)+{\gamma}_4{O}_{ic}\left({PO}_c-\overline{PO}\right)+{R}_{ic} $$

(4)

For a non-grade 3 obese person, the slope of the country-level prevalence of obesity is γ₃. For a grade 3 obese individual, the slope of the country prevalence of obesity becomes γ₃ + γ₄. It follows that γ₄ measures the difference in the association between the country prevalence of obesity and subjective health for obese people (of grade 3) versus people without this type of obesity.

Appendix 2

Fig. 3

Coefficients of alcohol consumption in SAHS regressions, with a fixed effect (vertical line) plus country-specific random-slopes (horizontal bars)

Table 3 Country-level measures