I'm sure there is a "proper" way to do this that respect the fact that Likert answers are not truly normally distributed because of their ordinal nature. But I find that simple solutions often work well for my purposes (and publishing in journals for that matter). If you want the true answer, hold out for someone with more experience. 

First thing I would do if perform a test to understand if your agreement/disagreement to the statements is normally distributed. Something like qnorm plots will help do this visually. (Again, quick and dirty). 

Then get the means and distributions of each ( descriptive statistics). 

From there you can perform t-tests to see if the means are the same between your variables of interest based on your dummy variables. The below example should get you started.

ttest [likertq1], by [nat.culture] 

I agree with Simon. A t-test is known to be quite robust. I am not familiar with STATA, but I guess the output of the 'ttest' includes test values for heterogeneity of variances etc.. If you want to be really sure, you could run a Kruskal-Wallis or a Mann-Whitney test (if your data are far from normally distributed and or with heterogeneous variances).

https://statistics.laerd.com/stata-tutorials/kruskal-wallis-h-test-using-stata.php

Please there are two ways of dealing with likert scale data:

1. You can use ordered probit analysis 

2. You can use chi-square 

Yes; yet, in humanities researches (as in this case, here), such scales are usually treated as ordinal or even interval variables. For the 2 cultures samples case here, for ordinal data, a Wilcoxon-Mann-Whitney test, a Robust rank-order test, a Komogorov-Smirnov two-sample test, or a Siegel-Tukey test could be used. For interval data, a t-test is the obvious choice.

For or against parametric tests for Likert type scales, see attached files

You can run an ordered logistic regression, I believe the command is 'ologit".

Hi guys, thanks a lot for your answers! I really appreciate. 

Isn't it a problem that my independent variable (culture) is binary with oprobit regression? My variables are omitted because of collinearity since if culture1= 0, culture2 = 1.

Do I need to encode dummy variables? Or do I need to encode the ordinal data of my Likert-scale questions?

Thanks in advance,

Marc

You do not need to encode dummy variables, neither for your independent variable (culture), nor for the dependent variables (Likert type questions). 'oprobit' is fine since it does not require normality assumptions. The Mann-Whitney U test is a good alternative and should yield the 'same' results. See:

http://www.reed.edu/psychology/stata/analyses/nonparametric/mannwhitneyu.html

For oprobit, see:

https://www.youtube.com/watch?v=c9kvqeLFF8U

Sorry, for the ordered logistic regression, ologit, see:

https://www.princeton.edu/~otorres/Logit.pdf

I'm getting the same coefficient for both cultures per item/ question just with - for culture 1 and + for culture2. I think something is wrong there... :(

Marc, if needed, my email address (remove blanks):

jos.feys @ telenet.be

You should define your independent variable (culture) as just 1 variable? You do not need to use dummy variables (culture1 and culture2) to assess the impact of culture on the dependent variables (Likert items), as measures of leadership preferences. You do not need to complicate things.

Just another follow-up question on this, as I have a similar issue:

How do you manage your survey results and use them for the logistic regression?

I want to identify which factors affect the probability to invest into social impact.

In my model, I have 5 independent variables (Pro-social attitude, Socially-responsible consumption, Perceived effectiveness, Perceived financial return, Perceived risk) and my dependent variable is the probability to invest into social impact.

For each independent variable, I have a set of questions in my questionnaire (with a 7 point Likert scale). I guess I will have to use the means of the items for each question, but how can I do that in specific?

Thanks a lot for your help!

For simple analytics Microsoft Excel can be a preferred option by adopting different referencing criteria like absolute and mixed. Transpose matrix criteria can well be used for graphical representation and data visualization.