If you're saying that your explanatory variable is education and you'd
like to know what happens to your dependent variable (earnings?) if
education is increased by 2 years, then you simply use setx to increase
education from its observed mean to the mean plus 2 years. Your chosen
quantity of interest will then change and you can observe what your model
is saying about that change.
if instead you have a two-stage model, with education the dependent
variable in one stage and then somehow the predicted values from that are
put into the next stage, then you could use clarify in the two stages
separately, with the first stage just producing the predicted values and
then clarify in the 2nd stage adding to that predicted education value.
Gary
---
Gary King
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, email: King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Mon, 25 Apr 2005, Nicolas EPELE wrote:
> Gary: Sorry for bother you again. Let me give you an example: Ferreira
> and Leite in Educational expansion and income distribution. A
> Micro-Simulation for Cear� analyse what would happend if the mean
> education would be 2 years greater (an expanding educational policy)
> than the real mean. To generate this, they estimate an ordered probit to
> know the relationship among years of schooling and some people
> characteristics. They increase the education level respecting this
> relationship (the distribution of education from this characteristics
> resulting from the ordered probit) and they obtain a new vector of
> education (with the same characteristics). So they can apply this new
> education vector to an equation of earnings and estimate a
> counterfactual poverty and inequality measures. This is what I'm looking
> for, and that is my question to you if Clarify (that is a very
> interesting package) can do this. Sorry for having at one's disposal
> your time. Thank you very much again. Sincerely yours, Nicolas
>
> Gary King <king(a)harvard.edu> wrote:
>
> If I understand what you're saying, you have an ordinal probit
> specification and you want to change your explanatory variable M such that
> the predicted value (or expected category) of your dependent variable e
> increases by 2. I think with Clarify, you'd need to do this 'by hand',
> increasing M gradually until the prediction came out as you wanted it.
> Doing this via a bisection search would be the quickest strategy I'd
> guess.
>
> Gary
>
> ---
> Gary King
> Institute for Quantitative Social Science
> Harvard University, 34 Kirkland St, Cambridge, MA 02138
> http://GKing.Harvard.Edu, email: King(a)Harvard.Edu
> Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
>
>
> On Mon, 25 Apr 2005, Gary King wrote:
>
>>
>>
>> ---------- Forwarded message ----------
>> Date: Mon, 25 Apr 2005 08:28:59 -0500 (CDT)
>> From: Nicolas EPELE
>> To: King(a)Harvard.Edu
>> Subject: Professor Gary King, request
>>
>>
>> Professor Gary King:
>>
>> My name is Nicolas Epele. I'm writing from
>> Argentina, I'm economist from Universidad Nacional de La Plata and I'm trying
>> to use the Clarify package in Stata. I'm trying to do the next simulation: I
>> have a dataset of micro-data with years of schooling, e, an characteristics,
>> M. I'm running the ordered probit model OP(e | M). So would like to generate
>> an e' vector of years of schooling with a mean 2 years greater than the mean
>> of vector e, but preserving the relation established in the model. So if I
>> run OP(e' | M), the coefficient will be similar to OP(e | M) model. I'm
>> writing to know if Clarify can do that.
>>
>> Thanks in advance and thank you very much for
>> sharing the Clarify package. Sorry for my english.
>>
>> Faithfully yours, Nicolas
>>
>>
>>
>> ---------------------------------
>> Do You Yahoo!?
>> Todo lo que quieres saber de Estados Unidos, Am�rica Latina y el resto del
>> Mundo.
>> Vis�ta Yahoo! Noticias.
>
>
> ---------------------------------
> Do You Yahoo!?
> Todo lo que quieres saber de Estados Unidos, Am�rica Latina y el resto del Mundo.
> Vis�ta Yahoo! Noticias.
>
---------- Forwarded message ----------
Date: Mon, 25 Apr 2005 08:28:59 -0500 (CDT)
From: Nicolas EPELE <e_nico77(a)yahoo.com>
To: King(a)Harvard.Edu
Subject: Professor Gary King, request
Professor Gary King:
My name is Nicolas Epele. I'm writing from Argentina, I'm economist from Universidad Nacional de La Plata and I'm trying to use the Clarify package in Stata. I'm trying to do the next simulation: I have a dataset of micro-data with years of schooling, e, an characteristics, M. I'm running the ordered probit model OP(e | M). So would like to generate an e' vector of years of schooling with a mean 2 years greater than the mean of vector e, but preserving the relation established in the model. So if I run OP(e' | M), the coefficient will be similar to OP(e | M) model. I'm writing to know if Clarify can do that.
Thanks in advance and thank you very much for sharing the Clarify package. Sorry for my english.
Faithfully yours, Nicolas
---------------------------------
Do You Yahoo!?
Todo lo que quieres saber de Estados Unidos, Am�rica Latina y el resto del Mundo.
Vis�ta Yahoo! Noticias.
Dear Professor King,
I am student of political science at the University of Mannheim,
Germany. In my current research I am concerned with the explanation of
vote over reporting and therefore I have a question regarding Clarify.
Since you are one of its authors, I hope you have an idea to solve my
problem.
Using sample data I estimated a logistic regression to model whether or
not the respondents voted in the last three elections. Afterwards, I
used Clarify to receive predicted values for three specific
combinations of the independent variables (the difference is the year
of election). To identify the size of vote over reporting, I subtracted
the real turnout of each election year from my predicted values. As the
values of the true turnout lie within the confidence interval of the
predicted values, these differences are not significant.
The next step should be a test, if the differences between the true
turnout and my predicted turnout for the three election years are
statistical significant different. Therefore, my idea was to perform a
t-test for dependent samples. To do this, I need the variance of each
estimator and the covariance of the ones I would like to compare.
However as far as I know, Clarify offers the variance of a predicted
value, but not the covariance between two predicted values (for two
election years). Do you have any idea how I could get this covariance?
Thanks in advance.
Sincerely yours,
Tobias Stark
________________________________________________________________
Tobias Stark
Bismarckstr. 49
67059 Ludwigshafen
Germany
0049 (0)621 181 3433
--
Clarify mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=clarify
Hi all,
I have been reading the instructions to graph the clarify results in
stata. However, it is not clear how to keep the confidence intervals
low
and high values for a simple OLS regression instead of a probabilistic
model. When I try to do it:
estsimp reg dependent set of independents, sims(1000);
gen axis=_n + 1 in 1/204;
setx mean;
local a=1;
while `a'<=204 {;
setx keyindependent `a';
simqi, ev genpr(margins);
_pctile margins, p(.5,99.5);
replace plo=r(r1) if axis==`a';
replace phi=r(r2) if axis==`a';
drop margins;
local a = `a' + 1;
};
sort axis;
graph plo phi axis, s(ii) c(||);
It returns this message "Probabilities are not allowed with
regression", as I understand it when we say genpr(margins) we are just
generating the confidence interval and in the next line we define the
confidence level, such that we can replace those values to be graphed
later.
If I try without generating the margins I obtain the values but,
obviously, this are not kept by stata and thus I cannot do the graph
directly in stata.
I wonder if any of you have faced the same problem and how did you
solved it.
Thanks a lot
Yours
Jose Merino
--
Clarify mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=clarify
Dear all,
I am trying to use the Clarify for my two separate
models (one with the zinb model and the other with the
mlogit model).
First, it looks like the program does not support
either the zip or zinb model; what are some
alternatives other than using the nbreg or poisson
model?
Second, for my mlogit analysis, it looks like the
program does not support the cluster option (the error
message �matrix not positive definite�); what are some
alternatives?
Thanks so much in advance.
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
--
Clarify mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=clarify