In behavioral, health, and social sciences, any endeavor involving measurement is directed at accurate representation of the latent concept with the manifest observation. However, when sensitive topics, such as substance abuse, tax evasion, or felony, are inquired, substantial distortion of reported behaviors, attitudes and opinions might occur due to the self-representational issues. One major concern is the impact of the response distortion on the survey or test results. Reporting about socially undesirable or disapproved behaviors often involves systematic misreporting. For example, being strongly advised by a pulmonologist to cease smoking, a lung patient that is failing to quit will feel strong incentive to lie about his smoking behavior. Without validation measures, it is not possible to assess the amount of misreporting, and the resulting data can be exceedingly misleading when drawing inferences. In anticipation of response distortion, an alternative method of data collection, assuring confidentiality of individual responses, can lead to more accurate observations. When dealing with sensitive topics so-called randomized response techniques for data collection can provide the necessary degree of response protection. The models presented in this thesis are meant for multivariate randomized response data analysis. The models are useful for sensitive topic research, where a randomized response data collection method is used to neutralize systematic response bias. A distinction is made between models suited to small and large-scale surveys. First, for small data samples, Bayesian estimation procedures are developed for ordinal count data. Second, for mixed large-scale survey data, Bayesian randomized item response theory models are developed for measuring single and multiple latent respondent characteristics.
|Award date||6 Dec 2012|
|Place of Publication||Enschede|
|Publication status||Published - 6 Dec 2012|