Type I or Type III regression model

#1
Hello,
I am analyzing questionnaire data and I am not sure how to proceed. Hopefully, you can give me some input.
So, the questionnaire was set up in a way that different groups of people received slightly different questions. The first questions were something like

Group 1:
- how acceptable do you think the treatment is to achieve 'purpose 1'
- how appropriate do you think the treatment is to achieve 'purpose 1'
- how good do you think the treatment is to achieve 'purpose 1'

Group 2:
- how acceptable do you think the treatment is to achieve 'purpose 2'
- how appropriate do you think the treatment is to achieve 'purpose 2'
- how good do you think the treatment is to achieve 'purpose 2'

--> These were measured on Likert scales and we combined them into a construct that we want to use in linear regression as the outcome.

Later in the survey, participants were asked about the perceived effectiveness of the treatments. They all received the same questions.

Both Groups:
- How likely do you think it is that the treatment will help achieve "purpose 1"
- How likely do you think it is that the treatment will help achieve "purpose 2"

The thing is that people who were in Group1 gave a higher rating on the perceived effectiveness for purpose 1 and vice versa, probably because we influenced them with the first questions in the questionnaire where we somewhat implied that the treatment is effective for that particular purpose.

My question is: how can I account for that in my regression model(s)? We are mostly interested in the effect of purpose on the attitude construct but perceived effectiveness appears to be highly predictive as well. There are interaction effects between purpose and effectiveness.
Should I use Type I (or Type III) regression models or should I assess the predictor variables in completely separate models?
If I use Type I models, would the order of covariates be: purpose before effectiveness and then all the other variables (e.g. demographics)?

Thank you!
 
Last edited: