2011 Housing Planning Study - Technical Report

Home

Data Processing The CATI system is programmed to conduct certain types of data editing as the interview is being conducted. Out-of-range codes are not allowed and contingencies are enforced. Following the fielding process, data files are reviewed and edited for internal consistency and other possible errors that may have passed the automatic editing routines. Edited data are then coded by professional staff who assign numeric codes to open-ended items, and sort and check verbatim responses. Weighting and Balancing of Demand Survey Data An analysis was conducted to identify any serious non-response bias in the demand survey data and the check for mode effects between the Iandline and cell phone surveys. It was determined that there was no need to statistically adjust for mode effects. Disproportionate coverage for several demographic variables was noted, especially in the cell phone surveys. Following the procedures developed by The Centers for Disease Control for the Behavioral Risk Factors Surveillance System, with some adaptations based on system applied at Pew Research, SMS has developed a weighting system for dual frame sample surveys in Hawaii. The weighting has three components as shown below. 1. Sample Weights: The disproportionate sample design assured equal precision by district, but resulted in an unbalanced sample by district. Sample weights are designed to statistically adjust survey results for a disproportionate design by weighting survey results to the distribution of the populations form which the sample was drawn. Weights were constructed by dividing the population estimates by the sample counts on a cell-by- cell basis. This procedure is the same as the weighing procedure used in previous Housing Policy Study Demand Surveys. 2. Sample Raking: The weighting scheme for the housing demand survey in 2011 must also account for dual frame sampling (a difference in telephone service available to each household) and the heavier non-sampling error associated with two-frame sample surveys involving cell phones. Since the exact number of households by type of phone services, household size, home ownership, age and gender of respondents, etc., is unknown, the standard methods of poststratification (statistical adjustment for non- sample error) will not work. The solution is to use one of several methods of sample balancing, or raking as it is better known these days. The method begins with sample weighs applied as noted above, and then balances the sample for type of phone service (Iandline only, Iandline mostly, wireless mostly, wireless only, and no phone service). In the same procedure survey data are simultaneously balanced for disproportionality in other raking variables including: age of respondent, household size and type, homeownership, marital status, and households with and without children under the age of 18. 3. Replicated Weights: Replication-based weights have been developed to adjust for variance distortion resulting from to complex sample designs. They are required to adjust sample variances used for statistical tests and certain forms of multivariate analysis. Using the replicated weights, users can estimate standard errors for simple estimators like totals or complicated ones like logistic regression parameter estimates. Hawaii Housing Planning Study,2011: Technical Report Page 9 0 SMS, Inc. November,2011