Proper coding for Disriminant Analysis
Message posted by Shaun (via 193.132.229.110) on January 7, 2002 at 12:50 PM (ET)

Hi again

I'll add more info on my problem

I have a problem coding agreement scores intended for discriminant analysis, with the aim to create Fishers Classification Function Coefficients for predictive purposes.
My questionnaire gave the following response options Agree, Disagree. This would be simply dummy coding (1,0). However a lot of respondents did not reply to all the questions. I can either specify missing values to equal the mean but it would be more appropriate to specify a mid point where the responses are not bias.

Is recoding to this scale a viable option?
Agree=1
Neither=0
Disagree=-1
where missing values are 0.

My reasoning is that this would have less effect on the fisher coefficients, making it more accurate for predictive purposes. Please tell if this is ok or should I replace missing values with the mean and keep the scores(0,1).

The statements are lifestyle statements such as 'I'm ambitious'.

Any advice on this matter is most welcome.

thanks.