Research master thesis | Psychology (research) (MSc)
open access
Across clinical assessment tasks, a statistical model trained on the assessments of one person (a person’s model) has been shown to be more accurate than the person on which the model is based, the...Show moreAcross clinical assessment tasks, a statistical model trained on the assessments of one person (a person’s model) has been shown to be more accurate than the person on which the model is based, the Model-over-Person effect. Because the language that people use to express their state of mind is clinically meaningful, the objective of this study was to examine whether the Model-over-Person effect extends to language assessments as well as to identify conditions in which the effect occurs. The accuracy of the assessments of a person versus a person’s model was measured as their agreement with a reference standard (the mean assessment of multiple assessors) in two conditions: 1) the assessment of single words and 2) the assessment of texts. Artificial Intelligence based language assessments were employed to create the person’s model. No Model-over-Person effect occurred in the assessment of single words or all texts (N = 500 words/texts). A small Model-over-Person effect took place for all three assessors in the assessment of the longer texts (⩾ 50 words; dz = .39-.42; n = 23 texts). This effect be explained by the finding that a high amount of input data can make an assessment more prone to human error. Additionally, the relation between the accuracy and different assessment and language characteristics indicated that a person’s model could be more accurate in case of a low agreement among assessors and that the accuracy is not related to the confidence of the assessor in the assessment. The results show how computational language assessments can complement a person in accuracy and may support the use of computational language models as decision-support in clinical decision-making.Show less
Obsessive-compulsive disorder (OCD) commonly co-occurs with other psychiatric conditions. In recent years, network analysis has been employed to investigate the relationship between OCD and some of...Show moreObsessive-compulsive disorder (OCD) commonly co-occurs with other psychiatric conditions. In recent years, network analysis has been employed to investigate the relationship between OCD and some of its comorbidities. The objective of the current thesis was to explore the comorbidity network of OCD in relation to seven other psychiatric conditions, some of whose interactions with OCD have not been investigated by the former network analysis papers. The thesis made use of an open-source data which was collected from patients who registered to the Behavioural Health Partial (BHP) hospital program at McLean Hospital, Massachusetts between 30 November 2018–16 October 2019. The final sample consisted of 532 people. Their responses to four measures, namely Obsessive-Compulsive and Related Disorders Dimensional Scales, Patient Health Questionnaire, Generalized Anxiety Disorder Scale, Drug Alcohol Craving Scale were analysed using network analysis. In total, we estimated two regularized partial correlation networks. The first network consisted of eight nodes representing eight psychiatric conditions. Differing from the former network in one way, the second one represented OCD and depression at a symptom level. Network 1 showed that OCD was connected to all other obsessive-compulsive and related disorder (OCRD)’s and generalized anxiety disorder (GAD) but not to depression and drug alcohol craving. In this network, OCD was only linked to depression through other OCRD’s and GAD. Further, our results highlighted the importance of fatigue as it was one of the most central nodes in Network 2.Show less