A Plan for Analyzing Nonresponse Bias
Survey of Landlords, Baseline, Site I
Presents methods for identifying, evaluating, and correcting for nonresponse bias in the baseline survey of landlords in Site I of the Housing Assistance Supply Experiment. A Distinction is made between statistically significant and meaningful nonresponse bias. Differences between respondents and the population that are too large to have readily occurred by chance indicate significant bias caused by survey nonresponse. However, only bias large enough to impair our understanding of the population is meaningful. The percent increase in standard deviation caused by using the respondent mean instead of the population mean is used as a test for meaningful nonresponse bias. Where meaningful nonresponse bias is found, it can in principle be corrected by substratifying so that nonrespondents in each substratum form a random sample of that substratum, and by reweighting the data for respondents accordingly.