Hi Julia, if you are having troubles, the easiest thing to do would be to
run Clarify separately on each of the Amelia imputed data sets and to
combine the results afterwards. Best of luck with your research,
Gary
--
*Gary King* - Albert J. Weatherhead III University Professor - Director,
IQSS <http://iq.harvard.edu/> - Harvard University
GaryKing.org - King(a)Harvard.edu - @KingGary <https://twitter.com/kinggary> -
617-500-7570 - Assistant <king-assist(a)iq.harvard.edu>: 617-495-9271
On Mon, Sep 25, 2017 at 3:21 AM, Julia CAGE <julia.cage(a)sciencespo.fr>
wrote:
> Hi all,
>
> I have an issue with the combined use of Clarify/Amelia and I don't know
> how to solve it.
>
> Basically, I am estimating a sureg model for electoral data with district
> and election FE.
>
> If I run the following model, everything works well:
>
> xi : estsimp sureg (lvotesP1 controls i.yearst i.district)
> (lvotesP2 controls i.yearst i.district)
> (lvotesP3 controls i.yearst i.district), sims(1000)
>
> However, as soon as I add the mi() option (I multiply imputed the missing
> voting data and I want to use the information from all the datasets) then I
> got an error message linked to the FE (I have no problem when no FE are not
> included):
>
> variable _Iyearst_2 not found
>
> (and I cannot enter the FE manually because Stata only supports a limited
> number of variables)
>
> Note that this is not linked to my yearst variable; if I only use the
> district FE, I got the same error message but for _Idistrict_2
>
> Could you please let me know what I should do to fix this issue?
>
> Thanks!
> Best,
> Julia
>
> --------------------------------------------
> Julia Cagé
> Assistant Professor of Economics
> Sciences Po Paris, Department of Economics
> https://sites.google.com/site/juliacagehomepage/
> Email: julia.cage(a)sciencespo.fr
>
Hi all,
I have an issue with the combined use of Clarify/Amelia and I don't know
how to solve it.
Basically, I am estimating a sureg model for electoral data with district
and election FE.
If I run the following model, everything works well:
xi : estsimp sureg (lvotesP1 controls i.yearst i.district)
(lvotesP2 controls i.yearst i.district)
(lvotesP3 controls i.yearst i.district), sims(1000)
However, as soon as I add the mi() option (I multiply imputed the missing
voting data and I want to use the information from all the datasets) then I
got an error message linked to the FE (I have no problem when no FE are not
included):
variable _Iyearst_2 not found
(and I cannot enter the FE manually because Stata only supports a limited
number of variables)
Note that this is not linked to my yearst variable; if I only use the
district FE, I got the same error message but for _Idistrict_2
Could you please let me know what I should do to fix this issue?
Thanks!
Best,
Julia
--------------------------------------------
Julia Cagé
Assistant Professor of Economics
Sciences Po Paris, Department of Economics
https://sites.google.com/site/juliacagehomepage/
Email: julia.cage(a)sciencespo.fr
Dear all,
My name is Berta Barbet and I am a PhD Student at the University of Leicester, United Kingdom. I am trying to analyse whether or not the impact of different issues at explaining voters decision changes through time in several countries. In order to do so I am performing several MNL regressions with issue positions as IV and vote as DV. And now I would like to plot the results into graph to make the visualisation of the importance of the different issues more clear. I was using the prchange command for the SPost package. But this package does not offer confidence intervals for the predicted values, so I decided to use the CLARIFY package.
However, I have not managed to find a way to save these results with this package and I have not been able to find anyone who can help me. So far I have manage to calculate the changes in predicted values using:
estsimp mlogit vote Welfare Defense Constitutionalism Decentralization European_Integration Morality Freedom Multiculturalism Labour Market Green if year==1992
setx mean
simqi, fd(prval(1 2 3 5)) changex(Defense min max) genpr(defense92_1 defense92_2 defense92_3 defense92_5)
Although I would prefer to calculate the change in predicted values by moving one SD above and below the mean, instead of min-max, in general the results that show in the screen are enough. However, I cannot save them. esttab commands do not work with the package. And the package only seems to save monte carlo simulations for the predicted values not for anything different.
I have thought about the possibility of calculating the predicted probabilities at the different points and then just compute the difference and their variances. The problem is that unless I find a way to estimate the predicted probability at the highest and lowest value together, something that I have not been able to do (I only know who to do it by asking it to estimate them separated), the values of the Monte Carlo simulations saved come from different simulations. Hence, I should not substract them because they are not comparable.
I have a lot of results, so saving them manually would imply a lot of work and risks making mistakes.
Thank you very much for your help and patience.
Kind regards,
Berta Barbet Porta
I noted that Clarify's -nbreg- routine simulates alpha (the dispersion
parameter) by simulating ln(alpha) and then exponentiating each draw.
However, the results (predicted values) are different if one simulates
directly from alpha.
Is the first method preferable than the latter?
Thanks in advance.
Kind regads,
Javier
Dear all,
Does anyone know how to use clarify estsimp command with xtpcse models? I saw that people used estsimp_pcse command, but I simply could not get it work in my Stata 12. Any suggestions?
Thanks,
Ping
Hi,
I am trying to use Clarify to analyze some imputed data from Amelia.
The problem is that my preliminary analyses indicate I need to use
xtreg and Clarify doesn't appear to support this model. Any
suggestions?
Thanks,
Jenifer
--
Jenifer Whitten-Woodring, PhD
Assistant Professor
University of Massachusetts Lowell
Department of Political Science
978-934-4242
jenifer_whittenwoodring(a)uml.edu
Hi all,
if you want to test for first difference in an interaction model
(logistic regression) of the following sort: y=a+b1x1+b2x2+b3x1x3 where
both x1 and x2 are dummy variables you would normally write:
simqi , prval(1) fd(prval(1)) changex(x1 0 1). Right?
but would the program than know for itself that when x1==0 the
interaction effect x1x2 is also 0? can one take a significant difference
to be indicative or does the model not make any sense? or is there an
alternative command for this situation?
Thanks in advance,
a sociologist
--
Oshrat Hochman, PhD
Department of Sociology and Anthropology
Tel Aviv University
Dear Clarify users,
I have a question. Is there a way with CLARIFY to calculate the impact of a change in the LEVEL of an explanatory variable (X) on the LEVEL of the dependent variable(Y)?
With OLS, the way to do it is simply to make the following calculation:
Delta Y = b Delta X, where b is the unstandardized regression coefficient (Achen 1982).
With logistic regression, the above formula is inappropriate. Possible solutions are:
(1) To use weighted least-squares or generalized least-squares (Finkel, JoP, 1993: 8). In this case, the regression coefficient can be used directly to estimate the impact of a change in the LEVEL of X on the change in the LEVEL of Y.
(2) To use the (quite demanding) approach proposed by Denk and Finkel in the 1992 AJPS piece.
Could Clarify perform a similar estimation with logistic regressions?
Many thanks for your help,
Thomas Didier
McGill University