Basically I want to find out public attitudes towards punishment of sexual offenders. I have two scales to measure general attitudes, 6 vignettes which takes into account how the type of offence, age of the offender and gender of the public impacts the level of severity of punishment. I know this sounds very complicated but is it best for me to perform a multiple regression with my 4 IVs or Factorial ANOVAs and a multiple regression for the originl scales? Any help appreciated!!
I would use the SPSS generalized linear models implementation. It has a rather clean UI (the legacy dialogues are often horrible), gives you all flexibility to include any mix of multiple categorial and metric predictors and tinker with the distribution family distribution. I'd definitely also look at interaction effects. Perhaps, use AIC for some model pruning.
Use a (generalized) linear model, as Martin suggested. It allows using categorical as well as numerical predictors and further allwos you to investigate interactions between predictors, as Martin also mentioned.
What distribution family is adequate and whether or not you should transform your response depends on the statistical properties of your response and on the sample size. If the sample size is rather large and/or the assumption of a unimodal symmetric distribution is resonable, the normal distribution is usually ok.
I don't like data-driven model selection procedures (select "significant" predictors by p-value, R², AIC or some other statistic). The model should be set up on the basis of subject knowledge. Only when there is no subject-knowledge and a lot of possibly relevant predictors exist, one may try to identify a reasonably simple model based on the data. But then the same data cannot be used to estimate the significances. In such cases you may perform the model selection on a subset of the data and estimate the significances in the final model on another subset of the data. This requires having enough data, though.
You have 4 variables and 6 vignettes, so you are not looking at each combination of these 4 variables in the vignettes, or am I missing something. And echoing Jochen, your analyses should be guided by your models. How do you believe those 4 variables (and any interactions) are associated with each of the two scales? I assume with this complex a design a lot of thought went into the models that may underlie the data.
I share Daniel's confusion about how the 4 IVs relate to the 6 vignettes--please provide more info. Are there any repeated measures factors in your model?
Also, it seems you have two measures of attitude (the DV). Therefore, you need to choose between two univariate models vs one multivariate model (where univariate and multivariate describe the number of DVs in the model). I think that Huberty & Morris (1989) is still very relevant on that issue--see the link below. HTH.
https://pdfs.semanticscholar.org/ee6c/77c99c8e4530d0cccaedf85ed525fb22a02d.pdf
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA