0
$\begingroup$

I estimated a model with three-stage least squares (3SLS) and have two main explanatory variables (the model also includes a set of control variables but they not important at the moment). I want to compare the size of the coefficients of these two variables in a bar chart. Something like this:

enter image description here

I know that these coefficients are not standardized Beta coefficients. I was just wondering whether such a comparison makes sense or statistically plausible in my case?

I am not intent on discussing their relative importance in relation to the dependent variable. Also, I am not discussing their marginal effects. I just want to say something along the lines: "the first coefficient is almost four times the second one"

Does this make sense or I am making a serious mistake?

Thank you

$\endgroup$
  • $\begingroup$ This type of comparison may or may not make sense, depending on the model. For e.g., if there are interactions, this may not be the most meaningful way to display your model output. $\endgroup$ – mkt - Reinstate Monica May 27 '19 at 12:41
  • $\begingroup$ (side note: 0.467 is about twice 0.2, not four times) $\endgroup$ – mkt - Reinstate Monica May 27 '19 at 12:41
  • $\begingroup$ @mkt Thank you for your explanation. There are not interaction terms: I have 2 main variables of interest and 5 control variables. Since I do not have a strong background in Statistics and Math, I am a bit afraid of making some stupid mistakes by comparing these coefficients. $\endgroup$ – gluark May 27 '19 at 12:44
  • $\begingroup$ @mkt Yeah I know, I put the wrong bar chart here, there are a few of them. $\endgroup$ – gluark May 27 '19 at 12:47
  • $\begingroup$ It depends on what the two main explanatory variables measure. If they have the same scale (or a roughly comparable scale) then this comparison is fine, otherwise the comparison is completely nonsense. $\endgroup$ – Maarten Buis May 27 '19 at 16:57
0
$\begingroup$

You cannot compare the unstandardized coefficients. If you standardize the raw scores (i.e. convert to z-score) before calculating the correlation coefficients, they will be comparable. This assumes that the raw scores are continuous or pseudo-continuous.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.