Statalist


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

=?UTF-8?Q?R=C3=A9p.=20:=20Re:=20st:=20Interpretation=20of=20-mfx?==?UTF-8?Q?-?=


From   "Herve STOLOWY" <stolowy@hec.fr>
To   <statalist@hsphsun2.harvard.edu>
Subject   =?UTF-8?Q?R=C3=A9p.=20:=20Re:=20st:=20Interpretation=20of=20-mfx?==?UTF-8?Q?-?=
Date   Thu, 19 Jun 2008 07:34:55 +0200

Dear Maarten:

I sincerely appreciate your detailed reply (including the other e-mail). It's very clear and helps me a lot. 

I have though a secondary question of interpretation. I use the auto file and make a logit regression followed by mfx.

sysuse auto
logit  foreign mpg weight price
mfx

Here are the outputs:

------------------------------------------------------------------------------
     foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -.1210918   .0956825    -1.27   0.206    -.3086261    .0664424
      weight |  -.0068497   .0019993    -3.43   0.001    -.0107682   -.0029312
       price |   .0009264   .0003073     3.01   0.003      .000324    .0015287
       _cons |   14.42237   5.413925     2.66   0.008     3.811276    25.03347
------------------------------------------------------------------------------


Marginal effects after logit
      y  = Pr(foreign) (predict)
         =  .04198211
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
     mpg |  -.0048703      .00566   -0.86   0.390   -.01597  .006229   21.2973
  weight |  -.0002755      .00021   -1.33   0.184  -.000682  .000131   3019.46
   price |   .0000373      .00003    1.36   0.175  -.000017  .000091   6165.26


Here is my question: in the logit ouput, the coefficients on weight and price are significant. In the mfx output, the corresponding coefficients of dy/dx are NOT significant. How should I interpret this result?

Best regards

Herve





***********************************************************
Professeur/Professor
President of the French Accounting Association (AFC)
HEC Paris
Departement Comptabilite Controle de gestion / Dept of Accounting and Management Control
1, rue de la Liberation
78351 - Jouy-en-Josas
France
Tel: +33 1 39 67 94 42 - Fax: +33 1 39 67 70 86
mail: stolowy at hec dot fr
web: http://www.hec.fr/stolowy

>>> Maarten buis <maartenbuis@yahoo.co.uk> 12:18:36 pm 6/18/2008 >>>
--- Herve STOLOWY <stolowy@hec.fr> wrote:
> I run -mfx- after -logit- and have some difficulties to interpret
> some of the elements displayed in the mfx output. 
<snip>
> I understand the meaning of dy/dx, 95% CI and X. However, I don't
> know how to read the z and associated p-value. What is the underlying
> test? 

The standard error is the standard error of dy/dx. The z is the test
statistic of the test that dy/dx equals zero, and the reported p-value
is its p-value.

> Is the Std. Err. an important element to display? In other
> terms, if you want to present some of the elements of the output in a
> paper, which ones would you chose? dy/dx? Std. Err? z? p(z)? CI? X?

This is largely a matter of style. What you want is report the point
estimate (dy/dx) and some measure of uncertainty, this could be either
the standard error, the z, or the CI. 

It sounds a bit weird to report the z, but in fact it usually more
convenient for the reader than reporting the standard error. Think
about how you would interpret a table with coefficients and standard
errors: you would look for each coefficient whether the coefficient is
less than twice the standard error or not. If you report the z-value
all they need to do is look at whether the reported z-value is less
than 2 or not. It does not really matter which one of these three you
choose because it is pretty easy to recover the other statics from any
one of these (in combination with the point estimate).

Some people would only report the p-value. The problem here is that
recovering the other statistics from the p-value is much harder, as
usually substantial rounding takes place. Even worse is only reporting
whether or not a coefficient is significant or not (so-called "gazing
at the stars"). Now it is completely impossible to recover any of the
other statistics.

Anyhow, the choice of what to report is usualy not yours to make, as
most journals have pretty strict guidelines on what you should report.

-- Maarten 

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

http://home.fsw.vu.nl/m.buis/ 
-----------------------------------------


      __________________________________________________________
Sent from Yahoo! Mail.
A Smarter Email http://uk.docs.yahoo.com/nowyoucan.html 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html 
*   http://www.stata.com/support/statalist/faq 
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index