Weighting

Weighting involves counting a respondent as either more than one person or less than one person. In the 2008 ANES, African-Americans and Latinos were over-sampled. If we analyzed the unweighted data, the sample would have far too many African-Americans and Latinos to be representative of the US population. In order to restore representativeness to the sample, each African-American or Latino is counted as less than one person. For the most part, this weighting is transparent–you will not notice that it is happening at all. At times, however, some contingency table marginals or frequency distributions will have fractional respondents–this is because of weighting.