How to understand these (discrepant?) findings?

Hi everyone,

I'm doing a project where I'm developing/validating a new questionnaire about body image (BodyImageNEW).

I'm comparing it to an already-existing questionnaire of body image (BodyImageESTABLISHED).

I have two samples:
Undergraduate sample (control)
Eating disorder sample (clinical)

I wanted to see which questionnaire better predicted/accounted for variance in eating disorder psychopathology (in the clinical sample only). I ran a hierarchical multiple regression, entering first age and BMI, second BodyImageNEW, and third BodyImageESTABLISHED. It showed that BodyImageESTABLISHED was the only significant predictor of eating disorder psychopathology, accounting for more variance. This wasn't great news for BodyImageNEW.

So next I wanted to see how good each questionnaire was at discriminating between clinical and control. I ran a discriminant function analysis, entering both questionnaires and seeing how well they classified individuals into clinical or control.

I found that both were significant predictors, however BodyImageNEW was more significant and had a better correct classification rate.

If the established questionnaire better accounts for scores on a measure of eating disorders, why would the new questionnaire better differentiate between clinical and control?

So far my only guess is that maybe the established questionnaire and the eating disorder questionnaire are similar in some way (e.g. asking the same types of questions...?). Or that the old questionnaire captures body image concerns that are more ubiquitous, whereas the new one captures more pathological (hence why it differentiates better?)

Trying to make sense of this finding but stumbling!

Thanks :)


TS Contributor
Not much can be said about such things if sample sizes and p-values are not reported, or what you mean by better classification. And "significant" can mean anything between 0.049 and < 0.00000001.

With kind regards