Can I use mediation with my data?

Thanks for taking the time to read this in advance it is quite detailed, but I'm currently working on my psychology honours dissertation, and due to some feedback from my proposal, I want to change the statistical analysis I am going to conduct. Unfortunately I'm looking at mediation and my supervisor is not familiar with this concept, so I need to seek help elsewhere.

I am conducting a study with 2 IVs being 'Risk Factors' (high and low) and 'Protective Factors' (High and Low) with DVs being 'Affect' (continuous rating of 1 to 5), 'Emotion Words' (participants are asked to pick three words which have different ratings between 1-5 based on intensity and valence) and 'Risk Rating' (continuous rating of 1 to 5).

There will be two separate parts of the questionnaire to complete:
Part 1) All participants will be asked to rate two lists of factors Risk and Protective and provide ratings of 1 (no risk/protection) to 5 (extreme risk/protection) for the various factors
Part 2) All participants will be presented with 4 vignettes of hypothetical people with different combinations of the Risk and Protective factors (please see below)
- Risk Factors LOW, Protective Factors HIGH
- Risk Factors HIGH, Protective Factors LOW
- Risk Factors LOW, Protective Factors LOW
- Risk Factors HIGH, Protective Factors HIGH
(I have previously obtained ratings for the two lists, risk and protective factors, through a pilot study I conducted to combine these factors to construct the vignettes, but am including the lists again for rating in this study)

After people read each vignette they will be asked to rate their affect (overall evaluation/feelings) towards the person on a scale of 1 (very positive/pleasant) to 5 (very negative/unpleasant).

Then they are asked to choose three emotion words from categories of positive, neutral and negative words (they can choose words from multiple categories if they choose) that best describe their feelings towards the person. These words prior to the experiment have been given ratings by myself, based on prior research, of 1-5 (very positive to very negative). A participant's choice of words will be averaged into a single measure of emotion.

Finally they will be asked to give a risk rating for the person on a scale of 1 (no risk) to 5 (extreme risk) for each vignette.

It is repeated measures so all subjects will receive the exact same measures.

My first hypothesis is that 'affect' or a person's overall evaluation/first impression will precede their emotions felt towards the person portrayed in the vignette, and these will guide (mediate) their risk rating for that person (risk factors + protective factors => affect => emotion => risk rating). Ideally, I believe, the data should be analysed through a hierarchical regression to prove mediation, but my two IVs (risk factors and protective factors), have constant values (the mean combination of ratings obtained for the risk factors and protective factors as obtained from the pilot study). For example, for the 'Risk Factors LOW, Protective Factors HIGH' scenario/condition the combination of the risk factors value is 4.27 and the combination of the protective factors value is 2.45, and these are the same for every participant as they are receiving the same scenarios. Also, there are two mediating variables 'affect' and 'emotion', instead of one.

I also want to investigate whether measures of 'affect' and 'emotion words' are measuring the same thing, and are just as effective in predicting risk rating. Or if it is a mediating relationship in which an overall impression comes prior to the feeling of different emotion states.

Finally for my second hypothesis I want to look at what is more important to people when rating the level of risk a person is (risk factors or protective factors). For this I was going to conduct a 2x2 repeated measures factorial ANOVA.

I guess what I want to know is if, for this study, I can use mediation, and if not what is the best way to analyse this data?

I hope this has all made sense, if not, please let me know so I can explain in a clearer manner.

Thanks again!



Super Moderator
Hi Kirsty, always nice to see other psych students on the boards :) I don't quite have time to look over everything in your post (getting ready for data collection tomorrow!), but a few thoughts:

-If you're looking at trying to demonstrate a mediating effect, get your hands on a copy of the classic Baron and Kenny (1986) article discussing the mediator/moderator distinction, which should help you get a really good conceptual grasp of what you're trying to show
-BUT, while B&K's criteria were developed before SEM usage became widespread in psychology, structural equation modeling is really the best way to test the fit models including mediating variables. Are you familar with SEM at all?
Yes I have read Baron & Kenny. The problem is my IV's are categorical (High vs Low) and these cannot be used in a hierarchical regression for that reason. I do have mean values for the different combination of Risk and Protective factors, but I don't know if these can be used in the analysis as they will be constant values, and what if you have two mediating variables?

Unfortunately I am unfamiliar with SEM. I just searched online and read through a brief explanation. As we haven't been taught this method I'm a little worried of going outside the box, especially as this is for an honours dissertation. I'll have to look some more into it. Thanks!