Not sure about coefficient meanings

[Yifan Xiao]

Good day,

I am looking to find which variables result people vote Right parties, but the dataset of my professor give to me, the variable of 'general election vote' ranking from 0 to 12 represent people vote neutral to radical but not left to right of parties, so, I plan to do the second regression by using 'left-right position of party' variable, the higher of the value means more right of this party that voter thinks. Then, if one variable have both positive coefficient with them, can I say that this variable may the reason that cause people vote Right parties? Thank you.

Sincerely,
Yifan Xiao

[Clyde Schechter]

I wouldn't do it this way. Assuming your left-right position variable is coded 0 (left)/1 (right), and is not more fine grained, I would

gen party_position = -general_election_vote if left_right_position == 0

This will give you a party_position variable ranging from -12 to +12, with all the negative values corresponding to left parties and the positive values to right. The larger magnitudes will reflect how extreme the parties are in their particular direction. Then variables that are associated with right (or less extremely left) voting will have positive coefficients as predictors of this variable. If you aren't really interested in such a fine-grained outcome variable, you could also just use the left_right_position variable by itself as the outcome. This might work better for you, because with the -12 to +12 variable, a predictor variable that primarily is associated with a less extreme left vote, could be confused with looking like it is associated with a right vote.

Finally, unless this data comes from a randomized experiment, I would avoid making any claims about anything causing anything else. You may find associations between variables, but in observational data you cannot claim causality.