Measurement error: Estimation, correction, and analysis of implications

Given the potentially strong, adverse effects of measurement error and the possibility of minimizing these using Hidden Markov models (HMMs), the aim of this thesis is twofold: first to understand in more detail the problem of measurement error and second to investigate whether extended HMMs that are applied to linked data can be used for error, and to what extent this method can be feasibly implemented.

While the findings presented in this dissertation suggest that HMMs are a promising tool to correct for measurement error in categorical, longitudinal data, several additional aspects need to be considered before this approach can be applied in practice. Namely, the performance and feasibility of the method should be tested in a different context that goes beyond the topic of labor mobility and on data from different countries than the Netherlands. Also, if possible, additional sources apart from surveys and administrative registers should be considered. Furthermore, a thorough examination of model robustness and the sensitivity of parameter estimates to varying model specifications containing different assumptions ought to be carried out. Finally, it is also important to consider how researchers can use error-corrected microdata in their analyses, while accounting for the uncertainty of the “true state” membership.

Pankowska, P. K. (2020). Measurement error: Estimation, correction, and analysis of implications. Dissertation, VU Amsterdam, handle:1871.1/092c64b6-14ef-4816-a1ff-feb9fed73931.