Browsing by Author "Talkkari, Anna"

Objectives This study examined depression risk by defining its latent distribution in the general population. Within the fields of behavioral genetics and epidemiology, such distributions of liability are presumed normally distributed. However, this assumption has never been confirmed for depression risk, and in recent research permitting non-normality has produced different results than assuming normality. Defining the shape of the latent distribution of liability for depression would add to our knowledge of depression as a construct, aid future researchers in choosing suitable research methods and pave the way for predictions on depression onset. Methods The data was the 2015–2018 National Health and Nutrition Examination Survey (NHANES). Respondents’ liability for depression was measured with the 9-part Patient Health Questionnaire (PHQ-9). The respondents were 18-80+ years old, 5214 of them were women and 4985 were men. Davidian Curve Item Response Theory (DC-IRT) was fitted for the entire data (n = 10 199) and separately for the data collected between 2017–2018 (n = 5065). The DC-IRT allows for the estimation of the latent trait distribution simultaneously with the item parameters. The parameters were estimated using the response patterns from the questionnaire and Davidians semi-nonparametric distribution estimate. The methods ability to detect a latent normal distribution was also tested with simulated data. Results For the entire data, the estimated shape of the latent distribution of liability was bimodal, with the right peak more pronounced than the left. For the 2017–2018 data, the shape was unimodal, although left-skewed. With the simulated data the DC-IRT method failed to detect the latent normal distribution when the sample size was large (10 000) and the ordinal data was highly skewed. Conclusions The bimodality of the latent distribution of liability for depression would challenge the current consensus of depression as a continuous trait, but these results should be taken with a grain of salt. According to the simulation study the DC-IRT method is not fool proof when it comes to large sample sizes and highly skewed data. Nevertheless, there was no indication of normality for the latent distribution of liability for depression. In the light of this study the latent risk for depression cannot be assumed normally distributed, which ought to be accounted for in the methodology of future studies. In addition, the possible bias introduced by the DC-IRT under certain conditions is valuable information for future applications of the method.