Thank you all for your helpful comments. I think that -margins- may be a
new command in Stata 11 and so I do not have access to it (I'm still
using Stata 10).
The effects coding may be a way forward for me, I'll have to look into
that a bit more.
Just to clarify my original question - what is the correct adjust command
to get back to the original coefficients in the regression? The first
-adjust- leaves the categorical rep78 'as is', but the resulting
predictions by foreign do not agree with the coefficients estimated from
the regression. However, the second -adjust- holds the categorical rep78
at its values in the dataset, which Martin points out are the proportions
that emerge for -proportion rep78-, and the resulting predictions this
time correspond to the estimated regression coefficients. So the second
-adjust- command seems to be the 'correct' way to treat the categorical
rep78. Is this correct? If so, I am struggling to understand what this
means for variables that only take values of zero and one. I know that it
sets them to their proportions, but how does this make sense when you have
a categorical variable such as sex, where you are either male or female?
In this instance what would setting sex to a specific value, say 0.25,
mean? Also, if you are setting the categorical variable according to
their proportions, what happens to the reference category? This does not
have a coefficient and so cannot be set to a specified value?
I hope that all this is understandable, I am just not fully grasping why
the second -adjust- command seems to work, and exactly what it is
predicting.
Thank you all for your input so far.
Gillian
"Martin Weiss" <[email protected]>
Sent by: [email protected]
28/10/2009 12:18
Please respond to
[email protected]
To
<[email protected]>
cc
Subject
st: AW: Adjust after regression involving categorical variables
<>
Maybe this FAQ will assist you:
http://www.stata.com/support/faqs/stat/adjust.html
I am not sure what the question really is. You are conducting two
different
prediction exercises, and they lead to different outcomes, just as you
would
expect: In the first one, Stata holds -rep78- at its values in the
dataset,
in the second one it assigns the mean to them. Note that these means are
the
proportions that emerge for - proportion rep78 -...
HTH
Martin
-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von
[email protected]
Gesendet: Mittwoch, 28. Oktober 2009 12:44
An: [email protected]
Betreff: st: Adjust after regression involving categorical variables
Dear all,
I am struggling to understand the -adjust- command after regression
involving categorical variables. My aim in using -adjust- is to obtain
the predicted values adjusted for the categorical variable, but I am not
explicitly interested in the categorical variable and so do not want it
appearing in the -by()- option of -adjust-. I have been unable to find
any examples of this kind of use of -adjust-. I have reproduced my query
using the auto dataset below. I am using Stata 10.1 SE.
sysuse auto, clear
descr
** just for this example, assume that rep78 is categorical
xi: regress price weight turn i.rep i.foreign
** output
price Coef. Std. Err. t P>t [95%
Conf. Interval]
weight 4.243125 .6699849 6.33 0.000 2.903407
5.582842
turn -208.6987 125.9326 -1.66 0.103 -460.5164
43.11914
_Irep78_2 822.0914 1691.818 0.49 0.629 -2560.907
4205.09
_Irep78_3 710.281 1560.7 0.46 0.651 -2410.531
3831.093
_Irep78_4 341.2531 1631.858 0.21 0.835 -2921.848
3604.355
_Irep78_5 876.4049 1740.224 0.50 0.616 -2603.387
4356.197
_Iforeign_1 3239.838 859.1453 3.77 0.000 1521.871
4957.805
_cons -32.54137 4097.528 -0.01 0.994 -8226.054
8160.972
** want the predicted values by foreign - not specifically interested in
rep78 but wanted to adjust for it, but I am unsure as to how to treat
rep78
** option 1 - set continuous values to mean but leave rep78 as is
adjust weight turn, by(foreign)
** output
----------------------
Car type | xb
----------+-----------
Domestic | 5164.18
Foreign | 8390.29
----------------------
** However, you see that 8390.29-5164.18=3226.11, and not 3239.838 as
predicted by the model above
** option 2 - treat dummies created by -xi- as continuous, and also set
them to their mean
adjust weight turn _Irep78_2 _Irep78_3 _Irep78_4 _Irep78_5, by(foreign)
** output
----------------------
Car type | xb
----------+-----------
Domestic | 5160.01
Foreign | 8399.84
----------------------
You see that the final -adjust- command gives 8399.84-5160.01=3239.83, as
given by the regression model above. So it appears that the second
treatment of the categorical gives the 'correct' predictions. However, I
am struggling to interpret exactly what this means for rep78, and does it
make sense to set variables that are 0/1 to their mean?
I would be extremely grateful for any assistance with this.
Many thanks,
Gillian
------------------------------------------------------------------------
ATTENTION:
This message contains privileged and confidential information intended
for the addressee(s) only. If this message was sent to you in error,
you must not disseminate, copy or take any action in reliance on it and
we request that you notify the sender immediately by return email.
Opinions expressed in this message and any attachments are not
necessarily those held by the Health and Safety Laboratory or any person
connected with the organisation, save those by whom the opinions were
expressed.
Please note that any messages sent or received by the Health and Safety
Laboratory email system may be monitored and stored in an information
retrieval system.
------------------------------------------------------------------------
Think before you print - do you really need to print this email?
------------------------------------------------------------------------
------------------------------------------------------------------------
Scanned by MailMarshal - Marshal's comprehensive email content security
solution. Download a free evaluation of MailMarshal at www.marshal.com
------------------------------------------------------------------------
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
------------------------------------------------------------------------
ATTENTION:
This message contains privileged and confidential information intended
for the addressee(s) only. If this message was sent to you in error,
you must not disseminate, copy or take any action in reliance on it and
we request that you notify the sender immediately by return email.
Opinions expressed in this message and any attachments are not
necessarily those held by the Health and Safety Laboratory or any person
connected with the organisation, save those by whom the opinions were
expressed.
Please note that any messages sent or received by the Health and Safety
Laboratory email system may be monitored and stored in an information
retrieval system.
------------------------------------------------------------------------
Think before you print - do you really need to print this email?
------------------------------------------------------------------------
------------------------------------------------------------------------
Scanned by MailMarshal - Marshal's comprehensive email content security
solution. Download a free evaluation of MailMarshal at www.marshal.com
------------------------------------------------------------------------
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
