Interaction or moderation effect - suitable method for analysis?

#1
I have run a two-way between groups ANOVA to analyse my data. However, I have one more variable, which is not one of the factors in the ANOVA. I want to see how this variable interacts with one of the factors. I predicted that it will be a moderator.

This variable is measured based on scores given to 3 statements (1-9 scale). First of all, should I compute a new variable that combines these 3 statements in one variable, so I can analyse it? This new variable will have, for example 3 levels - 1-3(low),6-9(high), and everything else (neutral).

I was thinking of making this variable into a factor and include it in the ANOVA. However, I don't think I will have equal sample sizes for each of the levels of the variable, which will be problematic in the ANOVA.

Other way is to run a regression and add this variable as a predictor, and the product of the interaction with the other IV.

Could you give me some advice what will be better or maybe some other method that will fit better the situation?

Any help is highly appreciated. Thanks.
 

Karabiner

TS Contributor
#2
I have run a two-way between groups ANOVA to analyse my data. However, I have one more variable, which is not one of the factors in the ANOVA. I want to see how this variable interacts with one of the factors. I predicted that it will be a moderator.
You could add the presumed moderator as covariate in ANCOVA,
and also the moderator*factor interaction.
Other way is to run a regression and add this variable as a predictor, and the product of the interaction with the other IV.
Yes, that is very common.

With kind regards

K.
 
#3
Thank you for your answer K.

I understood that there is some controversy about putting a likert scale in ANOVA. Is it appropriate to use it? Also, I dont think I will have enough people to fall into each category (high, neutral, low). Can I use just the number instead of making categories so I can see how it affects the other variable with increasing/decreasing for example?

Is using the mean of the 3 statements a good way to it? I read some opinions that it is not a good practice. Even though, in this case the 3 statemets measure the same latent construct.
 

Karabiner

TS Contributor
#4
I understood that there is some controversy about putting a likert scale in ANOVA. Is it appropriate to use it?
Do you consider my person as the very authority
which can decide this controversy one and for all?
Unfortunately, you have to decide yourself.
Is using the mean of the 3 statements a good way to it? I read some opinions that it is not a good practice.
Just add them up (mind missing values, though).

With kind regards

K.