1998 Ohio Family Health Survey
County Level Analysis Guidelines

Center for Public Health Data and Statistics

Sampling Design: The 1998 Ohio Family Health Survey included interviews in each of the state's 88 counties. Counties were grouped into four categories: Appalachian, Rural Non-Appalachian, Metropolitan, and Suburban (see Figure 1). Suburban counties were defined as urban nonmetropolitan. The original sample design included over sampling of several counties in each of these four categories. Six counties also purchased additional surveys (see Figure 2). Reports currently in production for the state will include separate measures of the major survey content areas for each of the four county types.

Over Sampled Counties: A minimum of 380 households were surveyed in each of the 25 over sampled counties. Gallup statisticians considered this number the minimum size necessary to produce reliable county level estimates of uninsured rates. County specific estimates for the other major content areas of the survey are also possible. However, for most counties the sample size is too small to allow for reliable estimates of county subpopulations. It may be possible to develop reliable estimates of health insurance type, health status, health risk behaviors, health care utilization, and unmet health needs for the entire population in each over sampled county, but not for specific age groups, races, income classes, or other subpopulation groups within each county. Cross tabulations of multiple content areas, such as usual source of care by health or insurance status, also will not produce statistically reliable results for most over sampled counties.

Counties Not Over Sampled: Between 60 and 79 interviews were conducted in each of the 63 counties that were not oversampled. Estimates derived from such small samples are generally not reliable. However, The Gallup Organization is estimating uninsured rates for adults and children in each county using a composite estimation model. This model will include county-specific survey data, survey data from populations in other counties with similar characteristics, and data from other sources such as Medicaid enrollment, Medicare enrollment and the U.S. Census Bureau. Reliable estimates from analyses limited to survey data will require combining of several counties with similar characteristics.

Suggestions for Users Planning County Level Analysis: Due to potentially high sampling variability, the accuracy of county level and other subpopulation estimates from The Ohio Family Health Survey will vary widely. Using such estimates without taking into account their margin of error may result in misleading conclusions or unwarranted comparisons between population groups. The Ohio Department of Health recommends that subpopulation estimates be presented with indicators of variance such as standard errors or confidence intervals, and that differences between populations not be assumed without testing for statistical significance. Specialized survey analysis software (such as SUDAAN, WESVAR, or STATA) is needed to calculate standard errors and perform statistical tests because the survey design was not a simple random sample. The Ohio Family Health Survey design included stratification by counties and clustering within households. Many common versions of standard spreadsheet and statistical software packages (such as EXCEL and SAS) should not be used for analysis of variance because their formulas assume a simple random sample and do not allow users to input survey design statements.