Many thanks, Liz. I was able to replicate the weights following Dr.
King's explanation of weights applicable to matching with replacement (it
seems that works for 1-to-many matching without replacement as well):
j.mp/CEMweights
Thank you!
* 0.78 1 1.16 0 177 0 252 1 0
185 0*
(m_C/m_T)*Ws
m_C/m_T
*429/185=2.3189*
Ws
For matched pairs with 2 controls per 1 Treat (W 1/2) - *2.3189*1/2=1.159*
For matched pairs with 3 controls per 1 Treat (W 1/3) - *2.3189*1/3=0.772*
On Mon, Jan 30, 2017 at 3:47 PM, Elizabeth Stuart <estuart(a)jhu.edu> wrote:
Hi Katerina,
I believe this page will give you the details:
http://r.iq.harvard.edu/docs/matchit/2.4-20/How_Exactly_are.html
Liz
------------------------------------------------------
Elizabeth A. Stuart
Associate Dean for Education
Professor
Department of Mental Health
Department of Biostatistics
Department of Health Policy and Management
Johns Hopkins Bloomberg School of Public Health
estuart(a)jhu.edu
www.biostat.jhsph.edu/~estuart
On Jan 30, 2017, at 10:43 AM, Aikaterini Passa <ap7014(a)gmail.com> wrote:
Dear Liz,
Many thanks, again, for your input. If I can make one more question about
the weights obtained by the R program as a result of incomplete 1-to-many
matched pairs, how (what are the formulas) the program calculates
them? Using the Lalonde data, and do 1-to-3 matching without replacement
and without having an exact match, several matched pairs are incomplete
because there are not enough control units. In SAS, gmatch and PSmatching
are 2 macros to perform 1-to-many matching without replacement, but they do
not calculate weights for you when matched pairs are incomplete. If you
could share the formulas/rationale how one can manually calculate them it
would be very appreciated.
Many thanks for the help,
Katerina
m.out <- matchit(treat ~ re74+re75, data=lalonde, ratio=3,
m.order="random")
Warning message:
In matchit2nearest(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, :
Not enough control units for 3 matches for each treated unit when
matching without replacement. Not all treated units will receive 3 matches
table(m.out$treat, m.out$weights)
0.78 1 1.16
0 177 0 252
1 0 185 0
On Sun, Jan 29, 2017 at 1:26 PM, Elizabeth Stuart <estuart(a)jhu.edu> wrote:
> Hi Katerina,
> The reason those weights end up being 0.75 and 1.51 is that there aren’t
> two times as many unmarried people in the control group as in the treatment
> group; so 2:1 matching isn’t possible within both “married” groups, which
> means matchit then provides those weights to account for that. (In other
> words, because of the exact match, some people end up getting 1 match and
> others get 2 matches, depending on their marital status), and so then the
> weights account for that.
>
> > table(lalonde$treat, lalonde$married)
>
> 0 1
> 0 209 220
> 1 150 35
>
> So short answer is that yes, the weights should definitely be used;
> otherwise you won’t be reflecting the structure underlying the match and
> your analyses won’t be accurate.
>
> Liz
>
>
> ------------------------------------------------------
> Elizabeth A. Stuart
> Associate Dean for Education
> Professor
> Department of Mental Health
> Department of Biostatistics
> Department of Health Policy and Management
> Johns Hopkins Bloomberg School of Public Health
>
> estuart(a)jhu.edu
>
www.biostat.jhsph.edu/~estuart
>
>
>
>
>
>
>
>
>
> On Jan 29, 2017, at 12:22 PM, Aikaterini Passa <ap7014(a)gmail.com> wrote:
>
> Many thanks, Liz, for the response.
>
> In my case I also include an exact match on a variable when I perform the
> 1-to-many without replacement matching. I also tried to incorporate an
> exact match in your code, and as a result of that, the weights of the
> control matched cases take different values. While I was rying to find an
> explanation I came across Austin's 2008 article " *Assessing balance in
> measured baseline covariates when using many-to-one matching on the
> propensity-score". Because of the exact match there are several incomplete
> 1-to-many (in our case 1-to-2) matched pairs and based on Austin's article
> ignoring for incomplete matching can result into misleading balance
> diagnostics. I am wondering if these weights should be also used in the
> analyses. *
>
> *Many thanks for any feedback,*
>
> *Katerina*
>
> m.out <- matchit(treat ~ re74+re75, data=lalonde, ratio=2,
> m.order="random", exact=c("married"))
> 0 0.75 1 1.51
> 0 150 188 0 91
> 1 0 0 185 0
>
> On Sat, Jan 28, 2017 at 12:32 PM, Elizabeth Stuart <estuart(a)jhu.edu>
> wrote:
>
>> Dear Katerina,
>> I’m not sure why you are getting weights that are not 0 or 1 for that.
>> Here is some sample code using the lalonde data, which shows weights of
>> only 0 or 1.
>>
>> > m.out <- matchit(treat ~ re74+re75, data=lalonde, ratio=2)
>>
>>
table(m.out$treat, m.out$weights)
>>
>> 0 1
>> 0 59 370
>> 1 0 185
>>
>> Liz
>>
>>
>>
>>
>> On Jan 26, 2017, at 2:00 PM, Aikaterini Passa <ap7014(a)gmail.com> wrote:
>>
>> I have a similar question pertaining to the weights of matched groups.
>> When performing *1-to-many* *matching without replacement*, the program
>> gives weights to the matched controls other than 1. What these weights
>> resemble in this scenario (1-to-many matching without replacement), and
>> should they be incorporated in the analyses? Many thanks for your input.
>>
>> Katerina Passa
>>
>> On Thu, Jan 12, 2017 at 10:41 AM, Gary King <king(a)harvard.edu> wrote:
>>
>>> Hi Ignacio, here's the simplest explanation of weights I could come up
>>> with for one method (CEM), but it applies more generally to matching with
>>> replacement: j.mp/CEMweights
>>>
>>>
>>> Gary
>>> --
>>> *Gary King* - Albert J. Weatherhead III University Professor -
>>> Director, IQSS <http://iq.harvard.edu/> - Harvard University
>>>
GaryKing.org <http://garyking.org/> - King(a)Harvard.edu - @KingGary
>>> <https://twitter.com/kinggary> - 617-500-7570 <(617)%20500-7570>
-
>>> Assistant <king-assist(a)iq.harvard.edu>du>: 617-495-9271
<(617)%20495-9271>
>>>
>>> On Thu, Jan 12, 2017 at 9:59 AM, Ignacio Martinez
<ignacio82(a)gmail.com>
>>> wrote:
>>>
>>>> Hi everyone,
>>>>
>>>>
>>>> When you do matching with replacement you have to use weights because
>>>> some observations are used multiple times. Can somebody explain what
would
>>>> be the consequences of ignoring those weights when running OLS? My
>>>> intuition is that I would end up with bias estimator. Is this correct?
Is
>>>> it possible to sign the bias? Is there a paper that discuss this?
>>>>
>>>>
>>>> Thanks,
>>>>
>>>>
>>>> Ignacio
>>>>
>>>>
>>>
>>> -
>>> ---
>>> 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 <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
>>>
>>
>> -
>> ---
>> 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
>>
>>
>>
> <Austin-2008-Pharmacoepidemiology_and_Drug_Safety.pdf>
>
>
>
-
---
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