This paper talks about how after matching there should be less sensitivity to model
specification so in fact it shouldn’t matter too much which model you choose. This is
probably the closest literature I know of on this particular topic.
http://gking.harvard.edu/files/matchp.pdf
Liz
On Nov 17, 2015, at 8:34 PM, K Imran M
<drki.musa@gmail.com<mailto:drki.musa@gmail.com>> wrote:
Hi Liz,
Cool. And I guess adjusting the covariates individually also gives a
bigger and better picture of the model. By the way, have you come
across any literature comparing the two?
Many thanks Liz
Regards,
Kamarul
On Tue, Nov 17, 2015 at 11:53 PM, Elizabeth Stuart
<estuart@jhu.edu<mailto:estuart@jhu.edu>> wrote:
Hi Kamarul,
I personally prefer the model that adjusts for all of the covariates individually unless
you have a very small sample size or a rare outcome. Basically that is a much more
flexible model than is assuming that the outcome is a linear function of the propensity
score, and so I prefer it unless degrees of freedom are severely limited.
Thanks,
Liz
On Nov 16, 2015, at 10:07 AM, K Imran M
<drki.musa@gmail.com<mailto:drki.musa@gmail.com>> wrote:
Hi,
This is my first posting and I apologise if similar questions have
been asked before.
I have run these R codes (codes mostly taken from Matt Bogart blog).
My question would be:
1) (see the lowermost of the codes) Which model specification is the
'best' the describe the adjusted/matched predictive effect of
'treat'
upon 're78'?
2) What is the difference between the linear combination of 'treat +
pscore' against 'treat +
age+educ+black+hispan+married+nodegree+re74+re75' when both are using
the matched data generated by 'match.data'
###################################
library(MatchIt)
data1<-lalonde
#matching
m.out1<-matchit(treat~age+educ+black+hispan+nodegree+married+
re74+re75,data=data1,method = 'nearest',
distance = 'logit')
# create matched data based on matchit, giving the output ps as pscore
m.data1<-match.data(m.out1,distance='pscore')
# generate propensity score manually
m.data2<-lalonde
head(m.data2)
ps.mod<-glm(treat~age+educ+black+hispan+married+
nodegree+re74+re75,data=m.data2,family = binomial(link='logit'))
summary(ps.mod)
#predict for prosp score
head(m.data2)
m.data2$psore<-fitted(ps.mod)
head(m.data2)
dim(m.data2)
#do models
mod.reg<-lm(re78~treat+age+educ+black+hispan+married+nodegree+re74+re75,data
= lalonde)
mod.ps1<-lm(re78~treat+age+educ+black+hispan+married+nodegree+re74+re75,data
= m.data1)
mod.ps11<-lm(re78~treat+pscore,data = m.data1)
mod.ps2<-lm(re78~treat+age+educ+black+hispan+married+nodegree+re74+re75,data
= m.data2)
#restrict to ps between 0.1 to 0.9
m.data3<-m.data2[m.data2$psore>=0.1 & m.data2$psore<=.9,]
mod.ps3<-lm(re78~treat+age+educ+black+hispan+married+nodegree+re74+re75,data
= m.data3)
summary(mod.reg);summary(mod.ps1);summary(mod.ps11);summary(mod.ps2);summary(mod.ps3)
#####################
Thank you
Kamarul
--
Dr. Kamarul Imran Musa,
M.D. M.Community.Med.
Associate Professor (Epidemiology and Biostatistics) &
Public Health Physician
Dept of Community Medicine
School of Medical Sciences
16150 Universiti Sains Malaysia, Kbg Kerian
Kelantan
Thomson Reuters researchID:
http://www.researcherid.com/rid/G-4864-2010
Google-scholar:
http://scholar.google.co.uk/citations?user=aZyayMgAAAAJ&hl=en&authu…
blog:
http://designdataanalysis.wordpress.com
email : drkamarul(a)usm.my , k.musa(a)lancaster.ac.uk
-
---
MatchIt mailing list served by HUIT
List Address: matchit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe:
http://lists.gking.harvard.edu/mailman/listinfo/ei
MatchIt Software and Documentation:
http://gking.harvard.edu/matchit/
Browse/Search List Archive:
http://lists.gking.harvard.edu/mailman/private/matchit/
Matchit mailing list
Matchit(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/matchit
--
Dr. Kamarul Imran Musa,
M.D. M.Community.Med.
Associate Professor (Epidemiology and Biostatistics) &
Public Health Physician
Dept of Community Medicine
School of Medical Sciences
16150 Universiti Sains Malaysia, Kbg Kerian
Kelantan
Thomson Reuters researchID:
http://www.researcherid.com/rid/G-4864-2010
Google-scholar:
http://scholar.google.co.uk/citations?user=aZyayMgAAAAJ&hl=en&authu…
blog:
http://designdataanalysis.wordpress.com
email : drkamarul(a)usm.my , k.musa(a)lancaster.ac.uk
-
---
MatchIt mailing list served by HUIT
List Address: matchit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe:
http://lists.gking.harvard.edu/mailman/listinfo/ei
MatchIt Software and Documentation:
http://gking.harvard.edu/matchit/
Browse/Search List Archive:
http://lists.gking.harvard.edu/mailman/private/matchit/
Matchit mailing list
Matchit(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/matchit