Added value of multivariate regression model (mvreg)

[Murielle Fuchs]

Dear forum members,

I am using a multivariate model (mvreg) to analyse the effect of many control variables (age, education levels…) on the adoption of 6 categories of agricultural practices. I use this model since I have many independent and many dependent variables, with a high correlation between the dependent variables.

I understand from the Stata manual that coefficients and standard errors produced by mvreg are identical to those that are produced by multiple OLS regressions done separately. The main difference between mvreg and reg seem to be that I can conduct tests of the coefficients across equations.

But if I do not conduct any such tests, what is the relevance of mvreg? Since the coefficients are the same as an OLS regression, does this mean that multivariate and simple regressions are in fact the same thing (when one does not conduct further tests)?

I am somewhat confused and hope someone could shed light on the added value of mvreg.

Many thanks!

[Clyde Schechter]

But if I do not conduct any such tests, what is the relevance of mvreg? Since the coefficients are the same as an OLS regression, does this mean that multivariate and simple regressions are in fact the same thing (when one does not conduct further tests)?

It's very simple. If you do not do any cross-equation tests (or cross-equation -lincom- calculations or cross-equation -margins- calculations) then the only differences using -mvreg- instead of several -regress- commands are:

1. Less typing because the predictor variables and any options only have to be typed once.
2. The results of all the regressions are in a single table (which may be desirable or undesirable depending on your workflow).

[Joro Kolev]

All that Clyde said, and
1. -mvreg- cannot produce robust standard errors, for this you need to use -suest- if desired.
2. You might want to consider -sureg- if you want a model when some regressors enter in some equations but not in other equations.

[Murielle Fuchs]

Thanks to you both for the clarifications, that makes much more sense now!