Weighting variables in a questionnaire

#1
Hi all,

I am trying to develop a medical questionnaire to predict the presence or absence of pain. The questionnaire comprises of 4 different components, usually an overall score would be derived by adding each component together eg A + B + C + D. However, this approach doesn't consider the weighting of each component, eg A may be x10 more important than B therefore should be given a bigger weighting of the overall score.

I am trying to figure out how to work this out statistically. I have tried logistic regression, where the dependent variable would be the presence or absence of pain, in order to see which components predict pain by looking at their odds ratio. The odds ratio would therefore be used a coefficient. The problem I am finding is that not all of components have a P < 0.05. Are there any other analyses I could use?

Any help would be very appreciated! :) Thanks!
 
#2
About the weighting, there are different methods for estimating the weight of a variable. However, if you already have estimated the relative importance of the variables, you can simply calculated the weighted variable by crossing it in its weight.

The problem I am finding is that not all of components have a P < 0.05. Are there any other analyses I could use?
Firstly it doesn't sound like a "problem". Actually it is good to have some significant and some non-significant results together! However, if you expected that all of them should be significant and you are obtaining non-significant results for certain variables that are likely to be predictors according to the literature and theory, it is possible that your sample size is small or other obstacles have caused some non-significant artifacts.

You need to give us much more detail on your study before we can help you more
: How many variables are involved? Can you possibly name them or tell us if the predictors might be correlated to each other in some way? What is your sample size? How many groups are there? What is your design (explain in detail)?
 
#3
Thanks for your reply. I haven't already estimated the relative importance of the variables, that's what I am currently trying to achieve.

My sample size is 170, perhaps I am under-powered to see significant results? I am expecting all of the variables to be predictors. There are 4 variables involved, there are some correlations between them and I am just looking at one group. The questionnaire is designed to discriminate people with low pain from high pain.
Thanks :)