This application seeks support for an investigation of the relationship between age and the reliability of survey measures. It proposes to conduct research on four sets of issues concerned to this relationship. First, to investigate the relation between measurement reliability and age for a large number of survey measures. Second, to examine the relation between age and reliability of measurement by topic of question, in order to assess interactions between age and question content in affecting reliability. Third, to examine the relationship between question characteristics and measurement reliability by age. And fourth, to ascertain the extent to which conclusions about these relationships are affected by estimation strategy, specifically comparing LISREL maximum-likelihood estimation to alternative least-squares estimation strategies, namely using the EQS estimates based on arbitrary-distribution theory, and recently-developed techniques for structural equation models involving categoric variables, via LISCOMP. In order to accomplish these aims, support is requested for the secondary analysis of sex existing survey data sets, involving three-wave panel designs, which permit unbiased estimation of components of error variance in survey measures. The data sets to be analyzed in the proposed research are: (a) the 1956-58-60 National Election Panel Study (n = 1,132), (b) the 1972-74-76 National Election Panel Study (n = 1,320), (c) the 1980 National Election Panel Study (n = 769), (d) the 1973 reinterview subsample of the General Social Survey (n = 195), (e) the 1974 reinterview subsample of the General Social Survey (n = 195), and (f) the 1984 reinterview subsample of the German National Social Survey (the ALLBUS) (n = 154). Using these data sets the proposed research will investigate variation in reliability of measurement by age", by topic of the questions, and formal characteristics of the question. Reliability estimates will be obtained in these three-wave studies using several different approaches (LISREL, EQS, and LISCOMP) to estimating the parameters of structural equation models.