In a survey sent to business school students I want to measure predictors of ethical behavioral intentions. I have several latent independent variables such as attitudes, norms, self-efficacy to predict several DVs describing behavioral intentions such as inaction, avoidance (change job), confrontation, anonymous reporting, and whistle-blowing.

I am using four scenarios with each secnario differing in the severity of its illegality from blatant illegal to a gray area issue.

For each scenario respondents have different options ranging from "do nothing", via "change the department", "report the issues anonymously", to "becoming a whistleblower". For each option I want to give respondents the chance to select from a five point Likert scale ranging from strongly disagree to strongly agree. In other words the options are NOT mutually exclusive.

I am now looking at initial survey results which I want to use for validating my measures. My IVs factor very nicely and I get good internal and external validity. But the DVs are all over the place. My hope was that I would get factors around certain types of behavior (do nothing, quit, report anonymously, whistleblow). But that does not work out. Instead I get complete and utter chaotic factors for my DVs. :-( I understand why (the four scenarios differ in terms of illegality). But is there a suggestion as what I can do to get my wished for results namely several scenarios, each with several options, each with 5 point likert scale options?