I'm studying if the nature of misconduct varies by education level, within a given profession. It's essentially a two level IV (associate degree; baccalaureate degree) and a thirteen level DV (see attached misconduct scheme). One case can be charged on a single type of misconduct, or all thirteen types simultaneously (although that would be heinous).
Each of the thirteen misconduct categories are independent, and coded as a binary (0,1). Can I treat these as multiple response data? I reviewed 5-years worth of misconduct reports, so this may qualify as a content analysis instead? Should I just use chi-squ for independence with each misconduct category's frequency against the two education levels? I'm primarily using SPSS.
Any insight would be great!
Hi Mark - thanks for responding.
The ultimate question is whether or not the State should impose a higher education requirement on this profession (rather than let market forces determine the workforce). There are a number of considerations that we are testing. However, for this particular study, the theory is that more career-specific education results in an increased preparedness or "dedication" to proper conduct. The hypothesis is that individuals with more education will be less likely to commit professional misconduct.
However, I was going to run this as two-tailed. I considered ANOVA/ANCOVA but I think the binary data would violate assumptions? I'm not entirely familiar with a logistic regression, although that sounds interesting... I believe these samples are independent.
Attached are the descriptive stats for each education level. I've run Fisher's exact test for AD vs. BA on each of the misconduct categories, which had non-significant results. There are a few controls I'd like to add to future larger samples, but I need to know if this is an appropriate methodology.
Thanks for any more consideration!
It seems like you have 13 separate kinds of misconduct, so you should simple sum them up to determine how many each person did.
Multi-Response in SPSS is something entirely different.
I would probably be inclined to run a regression in a structural equation modeling program where I could use tetrachoric correlation coefficients on those binary independent variables to model an underlying assumed continuous latent variable ('propensity to conduct mischief,' so to speak.) That's bringing a lot of statistical horsepower to your question, but it does enable you to do what you seem to be thinking about. See http://john-uebersax.com/stat/tetra.htm for a good explanation of what I'm suggesting. If you want to email me directly (my email is at the university web site) I'll be happy to try some stuff on your data set.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS