Using z-scores and ANOVA?

[Seon]

Hello helpful people,

I am wondering whether I can use z-scores to analyse variance and compare means - as though they were raw scores.

The project: Participants are divided into 4 different intervention conditions. I have pre- and post-intervention anxiety and depression scores, however participants did not use the same anxiety assessments or depression assessments for the duration of the data collection. Eg. with depression, half the participants used the PHQ9 to measure depression, half used the BDI.

Can I analyse the variance and compare the means using z-scores eg. a depression pre-treatment z-score which comprises PHQ and BDI pre-treatment scores (and the same for anxiety)? If not, is there another way I can compare the effect of the different conditions on outcome?

I have been trying to figure this out for days and have done lots of reading and online searching. So far all I've figured out is that I probably can't analyse variance like this (something to do with main effects being lost?). I don't know if this is correct, or what analysis I should do, if I can't use combined z-scores for the analyisis.
Thank you in advance, any help with this is appreciated,

Seon

 

[hlsmith]

I do not have an answer for your primary question, but look forward to seeing what suggestions are posted. However, back to the issue of comparing different instruments - you may or may not have been alluding to this, but I believe there is literature out there comparing different depression/anxiety scales. These data compare the instruments to gold standards (psychiatric diagnoses) and provide agreement metrics (e.g., accuracy, sensitivity, specificity, etc.). I would review these data and metrics to see how comparable your different scales may be. It may be that the PHQ and BID have the same accuracy, but one has a high SEN and low SPEC while the other has a low SEN and high SPEC making direct comparisons inappropriate.

 

[Seon]

Thank you hlsmith for the suggestion. I've just checked and the PHQ9 and BDI are similar in terms of sensitivity (88% and 87% respectively) and not too different with regards to specificity 88% and 79% resp.) I hadn't known to compare these so thank you for the tip. I'll check the same for the anxiety measures.

I still haven't figured out whether I can pool and analyse z-scores for these measures. The ranges are very different (0-63 for the BDI) and (0-27 for the PHQ9).
I am still hoping to find out whether it's correct use of anova and means comparisons to use z-scores. Any advice with this is greatly appreciated. I'll post if I find anything else out in the meantime.

 

[hlsmith]

Have you come across any examples of "use of anova and means comparisons to use z-scores" in the literature for comparision of data.

 

[Seon]

Hello,
I have since had a meeting with a statistician at work who very helpfully expalined that the scores first need to be restructuredinto a single 'index' (using SPSS). These can then be standardised/converted to z-scores, and these z-scores can be used to do anova.
Thank you for your help and ideas!
S