Creating table with results from regressions

[Ursa Ferjancic]

Hello! I'm a student working on my master thesis and not very familiar with STATA's advanced syntax. I have to run more than 300 regressions and would like to avoid that. What I want to have in the end is the following:
I have 13 dependent variables, around 26 independent variables and need to run a regression for each independent variable on each dependent variable. As an output I would like to create a table where I would have dependent variable in the column and independent in rows. I want to save b, t and p-value (as *= p < 0.1, ** = p < 0.05 and *** = p < 0.01) for each regression. Can anybody help me create this table, please?
Thank you in advance.

[Enrico Zorzi]

Wellcome to statalist

you can use

h estout

[Ursa Ferjancic]

Thank you for your answer. Yes, I tried this option, but it doesn't give me exactly what I want. It gives me those results (b,t,p-value) for each regression (each column is a new regression), but I would like to have them in a way that one column (one dependent variable) has all independent variables below, like a correlation matrix.

Below is a screenshot of what I get if I use estout. I would like to have only one "bart" column and results from regressions on "bart" in this column, without constant below, like a correlation matrix.

Thank you for your help!

[Phil Bromiley]

You will increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

There is a good reason estout doesn't want to mix the parameters from regression (1) with (2) - putting them in the same column suggests they are from a single estimation. Note, that you'd also have to figure out what to do about having two estimates of the constant. Also, note that the -3.671 should probably be starred - if the standard error is .9 then the parameter is highly significant.

That said, there are few occasions where estimating 300 regressions can give you meaningful results (and having only 36 observations makes it even worse). When you report results in regression (1) you demonstrate that your results in regression (2) suffer from omitted variables bias and so are biased and inconsistent. That is, reg 1 shows dgender matters, so you can't then estimate the effect of age without controlling for dgender. [The only exception would be if all the rhs variables are orthogonal as they might be in an experimental manipulation.]

Actually, if all your regressions have only a single rhs variable, you can use correlations just as well and that is a standard way to present results. This doesn't solve the omitted variables problem.