Is a z-transformation appropriate in this case?

[thegermanguy]

Hi everyone :-)

n=150
Everyone gets a test with 14 items and a variable "score" is calculated from the responses.
Trouble is, 50 of my tests have been falsely printed and omitted two of the 14 items, so 50 folks only answered 12 questions, instead of 14. Now what.

My initial idea was to calculate z-values of "score" for the group of 100 (14 items) and the group of 50 (12 items) separately and, well, go about my business with the z-score-values as if everyone had answered the same test. The more I think about it, the more unsure I get if this is correct.

What do you think? And if this doesn't work, any other idea? Would be a shame to kick 1/3 of my vpn out...

Thanks in advance :-)

[Dason]

What kind of response did the questions have? Were they yes/no? Categorical? Likert? Continuous scaled?

What do you want to do with the data? Are you looking to compare different groups?

[thegermanguy]

Each correct answer to an item gave 1 point. For "score" points were added up.
Yes, I will compare two groups A and B. (To be honest it will be more complicated, a 2 faktor multivariate ANOVA, and I'm a bit scared, but this is the first point I need to clarify, because otherwise I don't even have a good dependent variable. So I try to simplify here)
An initial Chi-square showed that the proportions of 14ers and 12ers are pretty equal in Groups A and B, so we're ok on that front (in case it matters).

[hlsmith]

In may be interesting to perform factor analysis or at least a cronbach alpha, to help guide your decision.

[thegermanguy]

Hmm... not sure.

We're talking about a performance test. People were trained in first-aid, two training methods, Peer vs. non-peer trained.

Taught were things like "call 911", "check for pulse" etc.

Then they had to perform and whether or not they called 911 or checked for pulse they got a point or not. So total score was "performance of first aid". As I said, fo some reason in 1/3 of the vpn sadly 2 items weren't checked. (and actually not by mistake, but because the guidelines changed during the period of the study). So here I am with 100 vpn with a possible max score of 14 and 50 with a possible max. score of 12 and would like to still combine them into one sample of 150 vpn with a variable "score" to then compare the two teaching methods.

Sorry for trying to simplify too much before, I thought, the simpler the better. Learned something :-)

[Dason]

You might want to look into logistic regression which allows for a response that is binomial in nature.