In retrospective surveys, classification errors are usually of a systematic nature, since respondents tend to be consistent in their answers and to forget about past changes in their labor market status.
In this paper, I use descriptive and model-based approaches to analyze errors in the retrospective reports of employment careers. First, I use algorithm-based techniques to quantify dissimilarity between event sequences reported at different temporal distances from the events. I then use a model-based approach to determine measurement error in life-history data. While the Markov assumption allows us to take into account transition probabilities between each pair of states of the outcome variable, the latent component of the model accounts for measurement error. In most classical methods of correcting for measurement error, the independent classification error assumption (ICE) is made, that is, it is assumed that measurement errors are independent of one another. This means that errors referring to two different occasions are independent of each other conditional on the true (labor market) states, and that errors depend only on the present true state. Unfortunately, such assumptions appears to be unrealistic in regard to retrospective longitudinal data, where a respondent could forget or misplace episodes, which would lead to the same mismatch between true and observed state at consecutive months. As a result, errors at consecutive time points would not be independent of one another, violating the ICE assumption. Having more than one indicator for our latent variable at each time point, we are able to estimate a latent Markov model relaxing the ICE assumption.