Comparing the Selection of One Person per Household to the Selection of All Household Members: Can Less Be More?
Iannacchione, V. G., & Shook-Sa, B. E. (2011, May). Comparing the Selection of One Person per Household to the Selection of All Household Members: Can Less Be More?. Presented at AAPOR 2011, Phoenix, AZ.
The decision to select one or all eligible members of a sampled household (HH) is influenced by a number of factors including burden on the HH, cost, and the sampling variance of survey estimates. Design effects quantify the influence of a complex sampling design on the variance of survey estimates. Restricting a sample to one eligible person per sampled HH can have counteracting impacts on design effects. A one-person per HH sample requires subsampling in multi-person HHs which increases the design effects attributable to unequal weighting. Conversely, selecting one person from each sampled HH could reduce the design effects attributable to clustering because the potential intra-household correlation among respondents in the same HH is avoided. If this reduction is larger than the increase caused by subsampling, a one-person per HH sample can achieve the same sampling variance as a multi-person sample with less cost and burden.
We present the results of a simulation study that evaluated the design effects associated with the selection of one person per HH on personal victimization rates based on the 2008 National Crime Victimization Survey (NCVS) which currently selects all persons 12 and older for the survey. We selected replicate samples of one respondent from each HH from the 2008 NCVS public-use database and then compared the design effects associated with victimization rates for a one-person per HH sample to those of the existing multi-person sample. We found that the increase in design effects caused by unequal weighting associated with a one-person sample was significantly greater than the decrease caused by the elimination of intra-household correlation. We discuss the pros and cons of a one-person per HH design, and estimate the number of HHs that would be needed to equalize the sampling variances of the current multi-person sample design and a one-person design.