Impact of Reducing Assessment Length on Statistical Mediation Conclusions

5^TH PAPER WITHIN ORGANIZED POSTER FORUM

ABSTRACT BODY: Statistical mediation analysis is a common approach used in prevention research to evaluate interventions. A critical assumption of the mediation model is that the scores from the mediator measure are reliable and can be attributed with valid interpretations of the construct (MacKinnon, 2008). Furthermore, an important aim in social and health sciences is to reduce participant burden in questionnaire research to increase response rates and response quality (Galesic & Bosniak, 2009). However, reducing the length of a measure might lead to scale scores that are less reliable and might not represent the construct accurately. In other words, reducing measure length might violate an assumption of the statistical mediation model. Also, reducing measure length may lead to Type I or Type II errors when the scores from the reduced measure are used as either control or predictor variables, respectively (Credé, Harms, Niehorster, & Gaye-Valentine, 2012). Therefore, using scores from a reduced measure of the mediator could compromise the statistical mediation conclusions. Although the influence of reliability and validity in the mediator have been previously studied (Gonzalez & MacKinnon, 2016), the consequences of reducing measure length on mediated effect estimation have not been extensively investigated. The purpose of this presentation is to investigate how reducing the length of the measure of the mediating impacts the estimation of the mediated effect and inference based on those estimates.

Monte Carlo simulation methods were used to investigate the properties of mediated effect estimation in the single mediator, where the mediator was an item response theory (IRT) model. The Monte Carlo study conditions included varying the effect size of the mediated effect, sample size, number of items in the latent variable, and the number of categories of the item responses (dichotomous or polytomous).

Based on previous research, preliminary results suggest that the estimation of the mediated effect would be attenuated, and that there might be Type II errors in the mediated effect estimation because there would be less variance in the mediator. Implications on power and Type I errors of the mediated effect are also discussed.

Results suggest the importance of evaluating how reducing the number of items in measures can compromise finding evidence in favor of statistical mediation. This paper also raises awareness of the importance of psychometric and measurement theory in the statistical mediation model.