All our work begins with thoughtful sample design to assure our research will provide data from the population and subpopulations of interest.
Data from our studies is often used for important population estimates which require a known level of accuracy. This necessitates special methods for sampling and sophisticated weighting of data before analysis.
We use SPSS and SUDAAN statistical software to properly analyze data from our complex survey projects. For these studies, we prepare variance estimates that fully account for sample design effects due to intra-cluster correlation, unequal weighting, stratification, and without-replacement sampling.
Subpopulations within an overall population vary by size and the use of completely random sampling would yield responses from subpopulations in proportion to the total population. However, with this approach, responses from a subpopulation may not be sufficient to provide the necessary precision. By stratifying the sample and sampling each subpopulation independently, the number of responses within each stratum can be set to meet the required level of precision.
This sampling technique is used to address the variation in the sizes of subpopulations by increasing the sample in one specific subpopulation. It is often used when it is important to have sufficient precision to allow analysis within an important group. For example, when gathering data on income dependent factors such as health insurance coverage, it may be necessary to oversample lower income households in order to have sufficient precision to make accurate statements about the characteristics of this group.
Cell Phone Samples
The increase in the number of cell phone-only households has made it necessary to include cell phones in all random digit dial (RDD) sampling. A sample including only landlines will skew the data collected. Sophisticated software allows us to sample cell phones in specific local areas. We also account for the differences in demographic characteristics of landline and cell phone households by weighting data from each differently.
Address-Based Samples (ABS)
The widespread use of cell phones and unlisted landline numbers has made it more difficult to survey respondents in compact geographic areas. Address-based sampling allows surveys to be mailed and/or conducted by telephone to target small geographic areas. Sample can be stratified based on geography with a high level of precision.