Dear Prof King, Prof Tomz and Prof Wittenberg:
> thank you for posting the software on the web. it is very useful!
>
> i am using your "clarify" for seemingly
> unrelated regression. can you kindly let me know the following problem on
> syntax?
>
> since i have some explantory variables are common in those equations, i
> have the following questions of analysis specifications:
>
> 1. how can i specify different value of those same variables
> simultanously appear in different equations when using "setx" ?
>
> 2. is it meant that i am making all the explantory variables set to
> zero when i typed "setx" only.
>
> 3. is it meant that all other variables are set to their "mean" while i
> just specified only one variable?
>
> 4. how can i perform the first differencing and the graph of the seeming
> uncorrelated regression.
yours
yiu por
Below...
On Mon, 5 Aug 2002, George Krause wrote:
> Gary:
>
> (1) Can one use the standard set seed command performed when
> doing bootstrapping and MC simulations in STATA before using
> estsimp command of CLARIFY so that one can reproduce the
> same estimates in subsequent runs of a simulation on the same model?
I think so, but I haven't tried it. You can tho. Do set seed, run
clarify and look at the first (say) 10 simulations of some quantity. Then
do set seed again and run it again. If all the simulations are the same,
then it works (with very high probability). If they differ, then there's
probably something else you can do (which perhaps someone else on this
list I'm CCing can suggest).
>
> (2) Can one account for Robert Friedrichlike (AJPS1982)
> conditional marginal effects in CLARIFY? If so, how can this be done
> using CLARIFY within STATA? From examining the CLARIFY documentation on
> the web, it seems that to me that interaction terms can only be analyzed
> in isolation from the linear term that conditions the formers impact
> on the dependent variable (e.g., the FAQ on interaction terms leads me
> to think this way plus the CLARIFY documentation that I examined -
> though maybe I am overlooking something). In other words, if we have the
> following model:
>
> Turnout = alpha + Beta_1*Education + Beta_2*Race
> + Beta_3*(Education *Race) + e
>
> can we perform CLARIFY analysis on Beta_1 + Beta_3 (or Beta_2 + Beta_3)
> i.e., the conditional (marginal) effect of Beta_3 on Y and its
> corresponding conditional standard errors? Or is it that one can only
> analyze Beta_3 separately from Beta_1 using CLARIFY? If the latter is
> true, then this suggests that analyzing Beta_3 via CLARIFY only provides
> information on the deviation from the baseline effect Beta_1 (i.e.,
> partial effect) and not the conditional (full) effect of Beta_1 +
> Beta_3.
With Clarify, you can decide on the quantity of interest (beta_1 or
beta_1+beta3 or sqrt(beta_1+log(beta_3)) or anything else. then the
procedure is to use clarify to make that calculation for the simulations
so that you wind up with simulations of your quantity of interest. at
that point, you can summarize your M simulations any way you like, such as
with se's or confidence intervals.
Best of luck,
Gary
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (928) 832-7022 :
>
> Any thoughts/advice on these matters is greatly appreciated. Thank
> you for your time and consideration of my queries.
>
> Best Regards,
>
> George Krause
>
> George A. Krause
> Associate Professor of Political Science
> Department of Government and International Studies
> 337 Gambrell Hall
> University of South Carolina
> Columbia, South Carolina 29208
> (803) 777-4545/3109 (office/department phone)
> (803) 777-8255 (fax)
> George.Krause(a)sc.edu (e-mail)
> http://www.cla.sc.edu/GINT/facbio/krause.html (web bio)
>
clarify mailing list served by Harvard-MIT Data Center
List Address: clarify(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/listis.cgi?info=clarify
I think the names of these things vary by field and I don't know the term
you're using. But it sounds like you're talking about the analytical
approximation. What Clarify will do is to calculate this exactly, without
the approximation. They'll probably be close, but to the extent that they
differ, Clarify's should be better.
Gary
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (928) 832-7022 :
On Mon, 29 Jul 2002, mganz wrote:
> Hello,
>
> First let me thank you for a very nice piece of software. My question has to
> do with retransforming a logged dependent variable to the original scale. How
> do the results from Clarify compare to retranforming using the smearing
> estimate (with or without bootstrapping to get the standard error)?
>
> Thanks,
> Michael
>
> ________________________________________
> Michael Ganz, MS, PhD
> Assistant Professor
> Dept. of Maternal and Child Health
> Harvard School of Public Health
> mganz(a)hsph.harvard.edu
> http://www.hsph.harvard.edu/faculty/ganz
> Ph. 617-432-2382
> Fax 617-432-3755
>
clarify mailing list served by Harvard-MIT Data Center
List Address: clarify(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/listis.cgi?info=clarify
To whom it may concern:
I'm writing to seek help with a problem I've encountered with Clarify 2.0.
I am running three separate negative binomial regressions where the only
difference in the models are the dependent variables. In models 1 and 3 I
obtain interpretable predicted values of my DV using the simqi, prval(x)
command. However, in my second model, using the same code, I receive
predicted values that do not make sense (they are numbers in the millions
and greater) from the simqi command.
When I run the model in Stata without Clarify, Stata produces reasonable
predicted values. However, I like Clarify for its ability to hold
covariates at different levels and its SE's and CI's with the predicted
probabilities. I've used the program several times in the past and been
pleased with the output.
I appreciate any advice you might have regarding this problem.
Thanks very much,
Jennifer Victor
----------------------------------------
Jennifer Nicoll Victor
Ph.D. Candidate in Political Science
Washington University in St. Louis
E-mail: jnvictor(a)artsci.wustl.edu
Homepage: www.artsci.wustl.edu/~jnvictor
----------------------------------------
clarify mailing list served by Harvard-MIT Data Center
List Address: clarify(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/listis.cgi?info=clarify
Dear Paul,
Suppose you have 2 imputed datasets called ds1 and ds2. To get Clarify to
analyze these datasets, try the following
1. load ds1 or ds2 into memory by typing "use ds1"
2. run a model that clarify supports and specify the mi() option. If ds1
and ds2 are in your working directory, the command might look something
like this: "estsimp regress y x1 x2, mi(ds1 ds2)". If the datasets are
not in your working directory you will need to specify the full path, e.g.
"estsimp regress y x1 x2, mi(c:\mydata\ds1 c:\mydata\ds2)". Clarify will
report the parameter estimates and standard errors, based on the imputed
datasets.
3. To set the x's (or to get summary statistics for the imputed datasets),
use the setx command. Setx will remember the locations of the imputed
datasets that you specified at the estsimp stage. For example, "setx
mean" would set x1 and x2 to their means, again based on all the imputed
datasets. If you want to see actual numerical values of those means, type
"setx" by itself (after running "setx mean") and clarify will list out the
means for each x.
Hope this answers your question. Thanks for your interest in clarify.
- Mike
Michael Tomz, Assistant Professor Phone: 650-725-4031
Department of Political Science Fax: 650-745-2765
Encina Hall, Stanford University tomz(a)stanford.edu
Stanford, CA 94305-6044
=====================
On Wed, 3 Apr 2002, Paul Warwick wrote:
Hi Gary,
Thanks, but I'm still a little puzzled. I take it that estsimp will
calculate the means and SE's without specifying any modelname or any
variables. Is this correct? I can't find out for myself because I can't
get estsimp to find the imputed files. Without specifying a complete
path, I don't know how to get Stata to find them. I've tried everything I
can think of, but Stata seems to want something more or different. What's
the key?
Best wishes,
paul
clarify mailing list served by Harvard-MIT Data Center
List Address: clarify(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/listis.cgi?info=clarify
How do I access the quantities simulated by simqi? I had assumed that it
was a rclass program, but when I request the return list I only get
r(printed) and I don't seem to be able to access its value. I want to
post the results to a separate file for graphing purposes. The example
in the documentation uses the saved predicted values from genpr(), but I
am not clear on the relationship between the values set by setx and the
saved expected values (I am using a nbreg model). The expected values
are different for each case, so I assume they are based on the values of
the cases rather than the values set by setx.
Owen Abbe, Research Fellow
Center for American Politics and Citizenship
University of Maryland
1108 Tawes Building
College Park, Maryland 20742
301-405-9722 (Tel)
301-314-2532 (Fax)
www.capc.umd.edu
clarify mailing list served by Harvard-MIT Data Center
List Address: clarify(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/listis.cgi?info=clarify