[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Mfx in tobit: same significance across mfx?

From   Ryan Edwards <[email protected]>
To   [email protected]
Subject   st: Mfx in tobit: same significance across mfx?
Date   Mon, 4 Aug 2008 20:29:08 -0400

I'm using Stata 10's "svy: tobit" routine, but I'd like to look at different marginal effects than the default, which are the beta coefficients in the latent variable model.

The standard solution would be to run "mfx" afterward. The problem is that running mfx after svy: tobit (still) takes an extremely long time to report the standard errors. mfx, nose is much quicker, of course. So my question is whether I have to pretend I'm a grad student in the 1970s and run my large number of models overnight, or if there's a convenient shortcut.

Interestingly, in the old "dtobit" routine from STB56, for each independent variable the ratios of coefficients to their standard errors across all the various marginal effects are exactly the same.

Presumably the various scaling factors, which I think are things like the cdf, pdf, and sigma, were treated as constants in that code, so the ratio of coefficient to standard error is always the same regardless of the particular marginal effect of interest. If true, this would be a nice shortcut to getting transformed standard errors indeed.

But I don't see quite the same thing when actually running svy: tobit and then mfx afterward (once it finally cranks everything out). As I think is typical, the primary svy: tobit returns a t-stat while mfx returns a z-stat, but more importantly, the ratios of coefficients to standard errors are decidedly different between the default tobit output and the mfx output.

Could this be due to the svy augmentation? Or is it erroneous? Or does it reveal a problem in the original dtobit code?

Ryan Edwards
Assistant Professor of Economics
Queens College and the Graduate Center
City University of New York
[email protected]
tel: 718-997-5189
fax: 718-997-5466

* For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index