-2
$\begingroup$

I have to analyze the results of a survey. The questions regard different motivators that the interviewees encounter ("I do it to help my colleagues" as a motivator for instance); all the answers are ordinal (1=strongly disagree -> 5=strongly agree) and the aim of this study is to see which factor (or set of factors) (nationality, age, site they work in, gender, role in the organization, center they belong) is influencing the answer.

I have looked in a lot of websites and lectures (ordered logit/probit model, discrete dependent variables, Likert scale, correlation, regressions, and so on) but it seems that my study could match a lot of ways to analyze these data.

So far I only have 15 interview results, which is not a lot: we expect around 100 answers at the end. I would like to have causal inference: their current behavior (motivations) should help us to draw a conclusion (causal effect, does their nationality influence their motivation factors?) regarding the factors of motivations. Could someone tell me which is the best way to treat these results?

$\endgroup$
  • 6
    $\begingroup$ Probably not without a lot more details about your survey. Are the answers intended to indicate the same construct, or different ones? How much data do you have? Are the "etc." factors continuous/counts/ordinal/nominal? Did you have any kind of experimental control, or would you be content to predict even without being able to make causal inferences? $\endgroup$ – Nick Stauner Jul 8 '14 at 7:29
  • $\begingroup$ (1)so each question is regarding one motivator ("I do it to help my colleagues" for instance) which the interviwee answers according to the scale. (2) so far I only have 15 interview results, which is not a lot, we expect around 100 answers at the end. (3)Then the other factors are the center they belong and their role in the organization. (4) I would like to have indeed causal inference. their current behavior should help us to draw a conclusion regarding the factors of motivations. $\endgroup$ – sol.mer Jul 8 '14 at 9:33
  • 3
    $\begingroup$ Welcome to the list. Could you move your 2nd comment into the body of the question? That will make it easier to read. $\endgroup$ – Peter Flom Jul 8 '14 at 9:49
  • $\begingroup$ in addition, so far my data are not normal (qqnorm and shapiro tests) so an ordered logit model seems to be the best way $\endgroup$ – sol.mer Jul 8 '14 at 10:06
2
$\begingroup$

This question is very broad--to the point of not being very answerable. You should probably take some statistics classes or read some textbooks, and get very clear on the basics. That said, here are a couple of very broad points:

  • An ordinal regression model will be appropriate for your data. There are many questions loosely similar to yours already on the site, it may help you to read through the threads categorized under the tag.
  • You might as well go with ordered logit, because that is the cannonical link, but ordered probit is unlikely to differ importantly. It may help you to read my answer here: Difference between logit and probit models.
  • Regarding causal inference, the short answer is you cannot infer causality in your situation, you need a true experiment for that.
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.