Based upon the literature review and previous work done by Weissert, the following variables were expected to influence the prevalence of dependency among the noninstitutionalized elderly population:
- Demographic characteristics of the aged individual--measured by age, gender, race, marital status and living arrangement;
- Socio-economic characteristics of the aged individual--measured by education and income;
- Contextual characteristics of the elderly individual's community--measured by the supply of physicians, hospital beds, and nursing home beds; Medicaid nursing home eligibility policies; area mortality rates; urbanity; and climate.
Of course, the choice of predictor variables was limited to variables available on the merged 1984 NHIS-SOA/ARF data set and for which population distributions could be obtained for states and counties. Coupling the constraints of the merged data set and Medicare data, the following variable definitions were available for use:
- Sex--male and female (coded 1 if female and 0 if male);
- Race--white and nonwhite (coded 1 if nonwhite and 0 if white);
- Age Group--age in 5 year intervals from 65 to 85 and over (coded as a zero-centered variable equal to the youngest age in the five year interval minus 75, divided by 5, i.e. -2, -1, 0, 1, or 2) ;
- Age-Squared--a quadratic of the "age group" variable (coded as the square of the "age group" variable, i.e. 4, 1, 0, 1 or 4); and
- Interactions--pairwise combinations of all of the above (coded as the product of the pair).
In addition to these variables a number of contextual variables were hypothesized to affect the rate of functional dependency among the noninstitutionalized elderly. For the functionally dependent, residency in the community versus residency in hospitals or nursing homes is determined in part by access to nursing home beds (Weissert and Cready in press), and perhaps also by the supply of hospital beds, which sometimes serve as a substitute for nursing home beds (Weissert and Cready 1988). Income also is believed to enhance access to nursing homes (Scanlon 1980a; Scanlon 1980b).
The supply of physicians and Medicaid eligibility policies, both of which may enhance an individual's access to nursing homes and hospitals, may further affect rates of institutionalization among the functionally dependent.
Mortality rates are reflective of the health status of the elderly population. Measures of urbanity also are reflective of health status in as much as dwellers of urban areas face different threats to mortality and morbidity than residents of rural areas. In addition, urbanity is also a proxy for available health care options--both acute and long-term care--as well as available social supports, and as such may affect rates of institutionalization. Contextual variables available for inclusion in our model after merging the ARF and the 1984 NHIS-SOA included:
- the number of nursing home beds per 1000 elderly;
- the number of unoccupied nursing home beds per 1000 elderly;
- the number of acute care hospital beds per 1000 elderly;
- the per capita income of the population;
- the percent of the elderly who reside in poverty;
- the number of primary care physicians per 1000 elderly;
- the percent of the poverty population that is covered by Medicaid;
- the age-adjusted mortality rate;
- the number of heating degree days;
- the population per square mile;
- the elderly population per square mile; and
- the percent of the population that resides in an urban area.
The contextual variables were entered into our models as both continuous and categorical variables. For the categorical analysis the variables were collapsed into three levels: high, medium and low. To collapse the community variables they first were arrayed in descending order by size. Then using the upper and lower quartiles as starting points, breaks were set at the point in the array where large differences between two consecutive values existed and where consistency with substantive meaning applied.