I am doing a simple linear regression to test the relationship between voters' self-reported policy stands and their approval ratings for two hypothetical presidential candidates in a simulated election campaign.
Participants first report their own position on a range of issues (e.g. 'The government should increase taxes to fund better public education') on a 0-100 scale where 0='strongly disagree' and 100='strongly agree'. Later in the experiment, they also rate the two candidates (Candidate A & Candidate B) on similar issues, using the same scale where 0='strongly opposed' and 100='strongly in favour'.
My question is: Is there a 'standard'/'common' way to control for the way respondents use the scales in general (i.e. across all questions in the experiment). Some respondents are much more cautious in their ratings, not rating anything outside of 40-60, while others are more extreme, entering values towards the ends of the 0-100 scale. Although in some cases this might reflect respondents' 'true position', in many cases it seems it is more a reflection of the way they understand the scale. Similarly, some respondents rate almost everything low, while others tend to rate most things high.
One suggested possibility might be to use the rating of Candidate B on issue X as a control in the model for Candidate A's rating on issue X (so Candidate B's rating on issue X is one of the independent variables along with age, income, respondents' position on issue X, Y, Z etc with the dependent variable Candidate A's rating on issue X). Would this be a reasonable solution?
Or perhaps a Multivariate linear model? With the two candidates' positions as the 2 dependant vars? Would this be a better way of doing this sort of analysis in any case?
Many thanks in advance