|
Abstract:
|
To form weights for the sample respondents to its quarterly surveys, Arbitron uses a raking (or, sample balancing) procedure. For years, there have been three marginal variables in this raking procedure. Recently, a fourth variable has been added to the mix. This paper examines issues related to adding the new marginal variable. In particular, we examine different stratification configurations of the new variable and their effect on the statistical properties (e.g., mean square error) of radio listening estimators. Additionally, the population controls for this new variable are based on a survey with relatively small sample sizes. We examine the effect of these stochastic population controls on the precision of the estimators.
|
