Sampling Design & Analysis
The foundation of any scientifically conducted survey must be a sound sampling strategy. ISR devotes considerable attention to sampling methods, balancing statistical considerations with practical limitations dictated by field costs.
The Institute has also developed sampling design strategies for telephone surveys and for studying special populations. In addition, ISR has experience in computing sampling errors and in devising weighting and imputation schemes to facilitate analysis of data. ISR is working with consultant Dr. Mansour Fahimi from Marketing Systems Group (MSG) on sampling design, weighting for complex designs, and analysis.
ISR recommends that survey data be weighted to reflect variations in probabilities of selection as well as differential nonresponse and other factors which cause the sample and population distributions to differ. Although the national household sample and most telephone surveys are designed to produce self-weighting samples of households, variations on this self-weighting design are inevitable. The first source of variation in sampling probabilities arises from the special needs of the survey where certain subpopulations are oversampled, the oversampling criterion usually being a set of demographic variables like age, sex, race, and residential location. The second source of variation is the usual requirement that only one person in a sample household be interviewed, although several eligible respondents may live there. Because the number of eligible respondents in a household varies, the actual probability of selection of an individual is inversely proportional to the number of eligible respondents in the household. Because all ISR samples are selected by probability methods, the probability of selection can be computed for every household and for every individual. Thus, sampling weights, which are inversely proportional to the overall probability of selection, can be assigned to each survey respondent.