Analyzing experiments with stimulus variables

[Jakob Johansen]

Hi!

I have a dataset that contains a number of questions that are typical survey experiments where some words or parts of the question vary between groups of respondents.


I have the h01_trust variable that contains the responses of all respondents, no matter what part of the question they have answered. In addition, I have the variables h01_country and h01_treatment that determine respectively which country to mention in the question (varies between two) and if the respondent gets stimulus or not (also varies between two).

My question is how I can extract for example descriptive data from how respondents responded with or without stimuli and with which country when the information is encoded in different variables like now.

I use Stata 15.1.

Hope this is understandable, thanks in advance!

Jakob

[Clyde Schechter]

table h01_trust h01_country h01_treatment

will give you counts of all response categories under all combinations of country and treatment. If you also want percentages of each response:

by h01_country h01_treatment, sort: tab h01_trust

[Jakob Johansen]

Thank you, that worked well!

How do I treat these variables if I want to make a regression analysis where I want to how other variables impact h01_trust with and without the stimulus? Do I have to generate new variables?

[Clyde Schechter]

So, let's say X is a continuous variable, and you want to examine how h01_country and h01_treatment modify the way X is associated with the h01_trust response. That would be something like this:

mlogit h01_trust c.x##i.(h01_country h01_treatment) // AND PERHAPS OTHER VARIABLES

If X is discrete, use i.X instead of c.X. Do read -help fvvarlist- to understand what this notation all means and so that you can adapt it to your own situation.

Note that this is a fairly complicated model because you are building in two interactions, and you have a polychotomous outcome. If the response categories of h01_trust can be viewed as ordered, and if the assumptions can be justified, using -ologit- instead of -mlogit- would simplify things considerably. Either way, you will want to use the -margins- command to help you understand your data. For a very clear introduction to -margins-, I recommend the excellent Richard Williams' https://www3.nd.edu/~rwilliam/stats/Margins01.pdf. Once you have absorbed that, you can learn about more advanced features in the Stata PDF documentation.