Can I use ANOVA & regression with likert data?

Question

I am carrying out a research project investigating cheetah reintroduction as a conservation tool.

One component of the study is to determine the tourism demand for cheetahs (essentially, do people want to see cheetahs). I intend to do this with questionnaires, listing a number of animal species including cheetahs. The respondents scale the species from 0 -10 in terms of viewing preference.

Could I compare the means of all the species using one way ANOVA to see if there is any significant difference in species viewing preference? If there is a difference would a Bonferroni test by useful to see which species are different and possible that I could place them into categories.

I would also be collecting data on respondents' characteristics. I intend to use this to define the segments interested in cheetah viewing (e.g., country of origin, age group, etc.) and hopefully build a multivariate regression model that could be duplicated at other sites to see if there is a cheetah demand based on the sites visitor profiles.

Is it possible to use independent variables such as country of origin ( not continuous data) to develop a regression analyses?

Answer 1

Since the same people are going to be answering about multiple animals, the errors are not likely to be independent. This violates one of the assumptions of ANOVA and regression.

You need to account for that somehow. Two common ways are multilevel models and generalized estimating equations.

However, if all you are interested in is demand for looking at cheetahs, then you may not need the data on the other animals. You could use the 1 to 10 scale for cheetahs as the dependent variable and whatever you like as independent variables. If you do this, you probably want ordinal logistic regression, although the usual regression may be OK (technically, OLS regression requires a continuous dependent variable, but 10 categories may be enough to be close).