R-3: Relation of Sociodemographics and Local Characteristics to Community Participation and Community Living

Sample

This project relies on restricted-use data from pooled years of the ACS.  The ACS sample is a two-stage stratified sample design to yield a representative sample of the U.S. population. In the 2008 ACS public-use file, there were 1,827,509 respondents ages 18-64, of which 194,980 reported a disability. The projected sample of a pooled 2008-2010 data file would result in approximately 5,483,000 working-age respondents with 585,000 reporting a disability. 

One concern that arises in large secondary data analysis is that the sample sizes are too large - variables with very small effect sizes will be statistically significant (i.e., measure with high precision) but not relevant programmatically. However, statistical significance should not be confused with relevance. The magnitude of an estimate must still be placed in context. If this issue arises, we will consider, as a robustness check, reporting alternative test statistics based on the square root of the sample size.  Nevertheless, the extremely large sample size and sampling design of the ACS allows for precise local-level estimates.  The restricted-access data, available via RDCs, provides zip code level identifiers (upon which local characteristics from other data sources will be merged) and the type of institution for those not living in the community, both of which are not available in the public-use microdata (PUMS) files.

Data Collection and Measurement 

This project utilizes data from the ACS, a nationally representative survey with a broad scope, focused on collecting primarily comprehensive demographic and household data alongside a range of data on key social and economic issues. This is consistent with its role to replace the decennial census long-form. The ACS identifies the disability population using a six-question sequence:



(1) Is this person deaf or does he/she have serious difficulty hearing? 

(2) Is this person blind or does he/she have serious difficulty seeing even when wearing glasses? 

(3) Because of a physical, mental, or emotional condition, does this person have serious difficulty concentrating, remembering, or making decisions? 

(4) Does this person have serious difficulty walking or climbing stairs? 

(5) Does this person have difficulty dressing or bathing? 

(6) Because of a physical, mental, or emotional condition, does this person have difficulty doing errands alone such as visiting a doctor’s office or shopping?

Although the ACS does not provide direct measures of community participation it asks, “Because of a physical, mental, or emotional condition, does this person have difficulty doing errands alone such as visiting a doctor’s office or shopping?” Because this question asks about the perceived difficulty in going outside the home alone, this study will examine factors associated with perceiving that doing errands is difficult, rather than the extent to which people actually do so.  We will define community or institutional living in the following categories: resides in the community, a nursing facility/skilled-nursing facility, mental (psychiatric) hospital and psychiatric units in other hospitals, or residential schools for people with disabilities.

Independent variables will include person-level characteristics (i.e., disability type, age, ethnicity, income, education) and location-specific characteristics (e.g., urban/rural location, availability of public transportation, policies and programs, topography, and climate).  For independent living difficulty, the models will also include household composition, traits of the housing physical structure, including age of the structure, number of stories, structure type (e.g., single family home, mobile home, apartment building). In all models, disability type will be interacted with the other independent variables to see if the relationship between the dependent variable and independent variables differ by disability type.

Data Analysis

We will estimate the effect of individual and location characteristics on independent living difficulty, using a multi-level mixed effects logit model, akin to hierarchical linear modeling (HLM). We will take into account the nonlinearity needed to estimate a model with a discrete (zero-one) dependent variable. In essence, the first step is to estimate a logistic regression with independent living difficulty as a function of demographic/household characteristics and location-specific effects using logistic regression. The second step is to estimate these location-specific effects as a function of location characteristics. The same approach will be used for community living. 



To test the hypothesis that individual and environmental characteristics influence participation, we will use the Wald chi-squared tests to evaluate the overall significance of the model and groups of variables and use the Wald z-test for specific coefficients. Wald tests are similar to traditional F-tests and Student t-tests, but adjust for any potential heteroskedasticity.  A 0.05 level of significance will be used for all tests.  All estimates and tests will be adjusted to compensate for the survey design factors, using sample weights, replicate (jack knife) sample weights provided by the Census Bureau, and survey procedures in Stata.