Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Propensity score matching after multiple imputation
From
Adam Olszewski <[email protected]>
To
[email protected]
Subject
Re: st: Propensity score matching after multiple imputation
Date
Sat, 22 Mar 2014 20:36:24 -0400
Hi,
I have never attempted to MI treatments or outcomes. I think it's becoming
very tricky at that point, because the number of units in each arm will
vary. I think the only way would be to estimate the treatment effect in
each dataset and then average, but you are in very uncharted territory
here, especially with the estimation of error.
AO
On Sat, Mar 22, 2014 at 8:21 PM, Adam Olszewski
<[email protected]> wrote:
> Hi,
> I have never attempted to MI treatments or outcomes. I think it's becoming
> very tricky at that point, because the number of units in each arm will
> vary. I think the only way would be to estimate the treatment effect in each
> dataset and then average, but you are in very uncharted territory here,
> especially with the estimation of error.
> AO
>
>
> On Sat, Mar 22, 2014 at 3:54 AM, natalia malancu <[email protected]>
> wrote:
>>
>> Just realized that there is a scenario in which things do not work out
>> - that in which you are MI on the treatment variable (since psmatch2 /
>> teffects do not work with mi). Any thoughts on how one could go around
>> that?
>>
>> Once again thanks,
>> Natalia
>>
>> On Sat, Mar 22, 2014 at 5:25 PM, natalia malancu
>> <[email protected]> wrote:
>> > Thanks Adam. I'll give it a try and see how to it works.
>> >
>> > On Sat, Mar 22, 2014 at 6:38 AM, Adam Olszewski
>> > <[email protected]> wrote:
>> >> The technicalities depend on how you are storing your MI data (I use
>> >> the wide format) and what kind of propensity score adjustment you are
>> >> envisaging (matching, stratification, weighting). You could also
>> >> perhaps:
>> >> 1) -mi estimate: logistic- your PS model in all datasets.
>> >> 2) -mi predict- the propensity score.
>> >> 3) use the predicted score in a -psmatch2, pscore()- call.
>> >> You may need to -mi unset- the dataset before #3, I believe that at
>> >> least -psgraph- does not work with MI data.
>> >> I'm not guaranteeing though that this is a statistically sound
>> >> procedure.
>> >> Adam
>> >>
>> >> On Fri, Mar 21, 2014 at 11:02 AM, natalia malancu
>> >> <[email protected]> wrote:
>> >>> The references (totally skipped my mind, apologizes):
>> >>>
>> >>> Mitra, R. and Reiter, JP. (2011) Propensity score matching with
>> >>> missing covariates via iterated, sequential multiple imputation
>> >>> [Working Paper]
>> >>>
>> >>> Hill, J (2004) Reducing Bias in Treatment Effect Estimation in
>> >>> Observational Studies Suffering from Missing Datap [ISERP Working
>> >>> Papers]
>> >>>
>> >>>
>> >>> Adam: the paper I am referring to seems to be the earlier version of
>> >>> the one you are mentioning.
>> >>>
>> >>> a. I completely share your concern and I cannot come up with a
>> >>> fix-maybe others have some suggestions
>> >>> b. On the technical end I presume the scenario to deal with things
>> >>> would be (please do correctly if I am wrong): mi extract to get the
>> >>> datasets, psmatch2 to obtain the PS in each of the datasets,
>> >>> reconstructing a master containing all PS variables, constructing a
>> >>> variable containing the average PS, estimating the treatment effect.
>> >>>
>> >>> Thanks,
>> >>> Natalia
>> >>>
>> >>> On Sat, Mar 22, 2014 at 1:49 AM, natalia malancu
>> >>> <[email protected]> wrote:
>> >>>> The references (totally skipped my mind, apologizes):
>> >>>>
>> >>>> Mitra, R. and Reiter, JP. (2011) Propensity score matching with
>> >>>> missing
>> >>>> covariates via iterated, sequential multiple imputation [Working
>> >>>> Paper]
>> >>>>
>> >>>> Hill, J (2004) Reducing Bias in Treatment Effect Estimation in
>> >>>> Observational
>> >>>> Studies Suffering from Missing Datap [ISERP Working Papers]
>> >>>>
>> >>>>
>> >>>> Adam: the paper I am referring to seems to be the earlier version of
>> >>>> the one
>> >>>> you are mentioning.
>> >>>>
>> >>>> a. I completely share your concern and I cannot come up with a
>> >>>> fix-maybe
>> >>>> other have some suggestions
>> >>>> b. On the technical end I presume the scenario to deal with things
>> >>>> would be
>> >>>> (please do correctly if I am wrong): mi extract to get the datasets,
>> >>>> psmatch2 to obtain the PS in each of the datasets, reconstructing a
>> >>>> master
>> >>>> containing all PS variables, constructing a variable containing the
>> >>>> average
>> >>>> PS, estimating the treatment effect.
>> >>>>
>> >>>> Thanks,
>> >>>> Natalia
>> >>>>
>> >>>>
>> >>>> On Sat, Mar 22, 2014 at 1:34 AM, Adam Olszewski
>> >>>> <[email protected]>
>> >>>> wrote:
>> >>>>>
>> >>>>> In their most recent paper:
>> >>>>> Mitra R1, Reiter JP. A comparison of two methods of estimating
>> >>>>> propensity scores after multiple imputation. Stat Methods Med Res.
>> >>>>> 2012
>> >>>>> they recommend:
>> >>>>> 1) calculating PS in each imputed dataset
>> >>>>> 2) averaging PS accross the imputations
>> >>>>> 3) estimating treatment effect using the averaged PS
>> >>>>> I am not sure how this addresses the problem of uncertainty of
>> >>>>> estimates though. I am not aware of a method that would estimate the
>> >>>>> treatment effect taking into consideration the uncertainty about the
>> >>>>> propensity score.
>> >>>>> AO
>> >>>>>
>> >>>>> On Fri, Mar 21, 2014 at 8:41 AM, natalia malancu
>> >>>>> <[email protected]> wrote:
>> >>>>> > Hi guys!
>> >>>>> >
>> >>>>> > After reading Mitra, Robin and Reiter, Jerome P. (2011) and
>> >>>>> > Hill's
>> >>>>> > 2004 paper, I was wondering whether there is a way to:
>> >>>>> > a. compute and then
>> >>>>> > b. average propensity scores after multiple imputation. Causal
>> >>>>> > inference to follow
>> >>>>> >
>> >>>>> > In STATA 12, which I am using, this is not possible with psmatch2.
>> >>>>> > Is
>> >>>>> > is possible in STATA 13 with teffects? Are there are options I am
>> >>>>> > missing on?
>> >>>>> >
>> >>>>> > Any suggestions are much appreciated,
>> >>>>> > Natalia
>> >>>>> > *
>> >>>>> > * For searches and help try:
>> >>>>> > * http://www.stata.com/help.cgi?search
>> >>>>> > * http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/