img

FIGURE 2-1 Plot of R-index (top line) and response rate (bottom line) for Current Population Survey cohorts by months in sample.
NOTE: Top line includes 95 percent confidence interval error bars around month-in-sample (mis) values for the R-index.
SOURCE: John Dixon presentation to the panel (Dixon, 2011).

assumptions, citing the work of Andridge and Little (2009) as an exception. This method also calls for the filling in of missing data. The results are valid within the context of the survey, but they are not necessarily comparable across surveys and, as a result, this method is not commonly used. Examples of these indicators are the correlations between post-survey weights and survey variables, variation of means across the deciles of survey weights, comparisons of early and late responders, and the fraction of missing information (FMI). The FMI is computed within a multiple imputation framework (Rubin, 1977) in which the model relates the complete frame data and paradata to the incomplete survey data. The FMI is the ratio of the imputation variance to total variance.

Balance indicators (B-indicators) were introduced by Särndal (2011) in a presentation at the annual Morris Hansen Lecture. They are an alternative indicator of bias—a measure of the lack of balance between the set of respondents and the population. The degree of balance is defined as the degree of fit between the respondent and population characteristics on a (presumably) rich set of frame variables. Särndal introduced a concept of a balanced response set, stating, “If means for measurable auxiliary



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement