Hi Rachel,

First of all, I think it is really an interesting research question which is worth investigating further. I recently read a review of Everaert et al. on cognitive bias modification (2012, http://dx.doi.org/10.1016/j.cpr.2012.04.003 ) and I think it is a promising modeling approach that opens up new possibilities for the treatment of depressive.

The question which method to use for analyzing your data clearly depends on the data you have. If it is all quantitative (what I assume) and you have reliable and well-designed experiments, SEM might be a good option to study the complex interplay of different biases on depressive symptoms. You could model the outcomes of the experimental tasks as latent variables (“biases”) and test for direct and indirect associations between them and cognitive outcomes (“depression”), adjusting for covariates (age, gender, …). However, it may quickly become computationally intensive if you are looking at interactions (“combined biases”) between the latent variables. Also, what you have to keep in mind is that with a single cross-sectional study/experiments you will not be able to rule out the possibility of reverse causality. E.g., it is possible that biased attention facilitates interpretation bias but it may be that wrong interpretations enhance attentional bias as well. So, given that you have cross-sectional data you will only be able to describe associations, not causality.

Other methods which are probably easier to apply (and to interpret) than SEM include correlation analyses or simple ANOVA/MANOVA – which would, however, not account for complex covariance structures between the different sources of bias. Furthermore, if you are able to obtain longitudinal data from your participants, i.e. to test them two or three times, you could also consider latent growth curve analysis in order to describe changes over time and interindividual differences using a combination of fixed and random effects. This would approximate causal relationships a bit better.

However, I would go for SEM as well, given that you have appropriate data. A good option here is to use MPlus software which can handle a large variety of latent variable models. Good introductions to SEM include Hox (2007), Lei/Wu (2007) and the various Mplus guides of Muthén and Muthén at their homepage (https://www.statmodel.com/ugexcerpts.shtml).

I hope, this helps you a bit to go on.

Good luck!

Best,

Alex

Article An Introduction to Structural Equation Modeling

Hi Rachel,

Our lab has been interested in this question for a while. My PhD student, Jonas Everaert has done some cool studies in this regard (see his googlescholar page: https://scholar.google.be/citations?user=j6InguYAAAAJ&hl=en&oi=ao).

Feel free to contact us if you have specific questions about methodology and analyses.

Best wishes

Ernst

Thank you both so much for your advice! It's been very helpful! This is an area that i've been interested in for a while but I'm a newbie to SEM. Thanks again!

Also Ernst, I am familiar with Everaert and the papers he has written are essentially what I am basing my dissertation on. Originally, I wanted to focus solely on interpretation bias but I felt that there was already enough literature surrounding the topic and I wanted to do something worth while. Then I stumbled across yours and Everaert's papers and decided to look further into the CCBH as it's such an interesting topic. I really hope that I'll be able to conduct a worth while study!