Interpretation of results

[Doris Schmidt]

Hi all,

I want to look if more educated people are more altruistic. Can I do this with this code?
In a survey, I asked participants which education level they have and I tried to generate the education variable in this way.

gen edu=1 if education=="Still in school"
replace edu=2 if education=="Secondary School certificate"
replace edu=3 if education=="High School dilpoma"
replace edu=4 if education=="Bachelor degree"
replace edu=5 if education=="Master degree"
replace edu=6 if education=="P.h.D"


regress altruism_mean edu

Depending on the coefficient, can I now say that the more educated the person, the more/or less altruistic the person is (if the coefficient is significant)?

Many thanks in advance.

[Felix Bittmann]

As long as the original variable education is a string variable, this should work. However, since educational degrees are not a metric measurement, I would suggest using the i.prefix like

regress altruism_mean i.edu

For this question, a Spearman correlation might also be interesting.

spearman altruism_mean edu

[Carlo Lazzaro]

Doris:
as an aside to Felix's helpful advice, I would also check whether some latent variables (say, soft skills) can influence both the regressor and the regressand. If that were the case, your model would be misspecified for latent variable-led endogeneity.
In addition:
1) I find really^2 hard to believe that you can address a complicated issue like altruism with one regressor only (unless you're intrerested in correlation only, as Felix proposes in his reply; but reading your post once more you seem to be interested in linear regression);
2) clustering standard errors on some relevant predictor is probably useful here;
3) if you're dealing with survey data, see -svy- suite.