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 - King(a)Harvard.edu - @KingGary <https://twitter.com/kinggary> -
617-500-7570 - Assistant <king-assist(a)iq.harvard.edu>: 617-495-9271
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
>
>
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
i'm sure it is mentioned (probably in our paper somewhere). The costs and
benefits are not methodological; they are more of a choice about what
quantity of interest you are willing to try to estimate.
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 Wed, Jan 11, 2017 at 2:06 PM, Ignacio Martinez <ignacio82(a)gmail.com>
wrote:
> Thanks a lot Gary. Is there any literature that talks about this case? I
> imagine that there are plus and minuses to those approaches.
>
>
>
> On Wed, Jan 11, 2017 at 2:04 PM Gary King <king(a)harvard.edu> wrote:
>
>> one simple possibility is to switch 0s to 1s and 1s to 0s. if that
>> really won't work for you, then you could match with (a lot of)
>> replacement.
>>
>> 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 <(617)%20500-7570> -
>> Assistant <king-assist(a)iq.harvard.edu>: 617-495-9271 <(617)%20495-9271>
>>
>> On Wed, Jan 11, 2017 at 2:01 PM, Ignacio Martinez <ignacio82(a)gmail.com>
>> wrote:
>>
>> Hi everyone,
>>
>> Is there a paper that talks about matching when the sample has more
>> treatment observations than control observations? Is there an algorithm
>> that works better for this case? Can somebody explain to me why optimal
>> matching does not work at all in this case?
>>
>> Thanks,
>>
>> Ignacio
>>
>>
>>
one simple possibility is to switch 0s to 1s and 1s to 0s. if that really
won't work for you, then you could match with (a lot of) replacement.
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 <(617)%20500-7570> - Assistant <king-assist(a)iq.harvard.edu>:
617-495-9271 <(617)%20495-9271>
On Wed, Jan 11, 2017 at 2:01 PM, Ignacio Martinez <ignacio82(a)gmail.com>
wrote:
> Hi everyone,
>
> Is there a paper that talks about matching when the sample has more
> treatment observations than control observations? Is there an algorithm
> that works better for this case? Can somebody explain to me why optimal
> matching does not work at all in this case?
>
> Thanks,
>
> Ignacio
>
Hi everyone,
Is there a paper that talks about matching when the sample has more
treatment observations than control observations? Is there an algorithm
that works better for this case? Can somebody explain to me why optimal
matching does not work at all in this case?
Thanks,
Ignacio
for the first, you can take the 25 matches of any T to any C that are
smallest.
for the second, there is one graph it will calculate if it has the outcome
variable, but you don't need it or can ignore it.
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 Tue, Jan 3, 2017 at 10:50 PM, Juan Tellez <juan.f.tellez(a)gmail.com>
wrote:
> Thank you Gary. On the first solution, I imagine I would have to specify
> which 25 of the 50 treated units to find matches for? And on the second
> solution, since I am designing the survey I do not yet have an outcome to
> measure. The MakeFrontier() function call requires an outcome to run. Is
> there some workaround or is my thinking mistaken?
>
>
>
> On Tue, Jan 3, 2017 at 8:55 AM, Gary King <king(a)harvard.edu> wrote:
>
>> Hi Juan, You could ask matchit for the lowest imbalance on a "greedy"
>> basis, say 25 treated units with the closest controls or some such.
>> Alternatively, if you add one component -- a specific overall imbalance
>> metric -- you have a well defined mathematical problem. To rephrase, you'd
>> like the subset with 25 treated and 25 (or more) controls that has the
>> lowest level of imbalance among the (huge number of) all possible such
>> subsets. If so, this paper <http://j.mp/1dRDMrE> on the matching
>> balance frontier can calculate this.
>>
>> 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 <(617)%20500-7570> -
>> Assistant <king-assist(a)iq.harvard.edu>: 617-495-9271 <(617)%20495-9271>
>>
>> On Tue, Jan 3, 2017 at 9:26 PM, Juan Tellez <juan.f.tellez(a)gmail.com>
>> wrote:
>>
>>> Hello,
>>>
>>>
>>>
>>> I am planning a survey where I have 50ish treated municipalities and
>>> hundreds to choose from as potential controls. The tricky part of this
>>> matching exercise is that the survey will, in the end, only sample 25 of
>>> the 50 treated municipalities. I don't particularly care which 25 of the 50
>>> are chosen; what I am effectively looking for is the top 25
>>> treatment-control pairs from my sample.
>>>
>>> Is it possible to do this in MatchIt with a distance measure
>>> like mahalanobis? The MatchIt package is understandably conservative about
>>> discarding treatment observations, and when I use it to match I generally
>>> end up with around 50 matched treated units. How might I go about this?
>>> Thank you.
>>>
>>> --
>>> Best,
>>>
>>> Juan Fernando Tellez
>>> PhD Candidate
>>> Department of Political Science
>>> Duke University
>>>
>>
>>
>
>
> --
> Best,
>
> Juan Fernando Tellez
> PhD Candidate
> Department of Political Science
> Duke University
>
Hi Juan, You could ask matchit for the lowest imbalance on a "greedy"
basis, say 25 treated units with the closest controls or some such.
Alternatively, if you add one component -- a specific overall imbalance
metric -- you have a well defined mathematical problem. To rephrase, you'd
like the subset with 25 treated and 25 (or more) controls that has the
lowest level of imbalance among the (huge number of) all possible such
subsets. If so, this paper <http://j.mp/1dRDMrE> on the matching balance
frontier can calculate this.
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 Tue, Jan 3, 2017 at 9:26 PM, Juan Tellez <juan.f.tellez(a)gmail.com> wrote:
> Hello,
>
>
>
> I am planning a survey where I have 50ish treated municipalities and
> hundreds to choose from as potential controls. The tricky part of this
> matching exercise is that the survey will, in the end, only sample 25 of
> the 50 treated municipalities. I don't particularly care which 25 of the 50
> are chosen; what I am effectively looking for is the top 25
> treatment-control pairs from my sample.
>
> Is it possible to do this in MatchIt with a distance measure
> like mahalanobis? The MatchIt package is understandably conservative about
> discarding treatment observations, and when I use it to match I generally
> end up with around 50 matched treated units. How might I go about this?
> Thank you.
>
> --
> Best,
>
> Juan Fernando Tellez
> PhD Candidate
> Department of Political Science
> Duke University
>
Hello,
I am planning a survey where I have 50ish treated municipalities and
hundreds to choose from as potential controls. The tricky part of this
matching exercise is that the survey will, in the end, only sample 25 of
the 50 treated municipalities. I don't particularly care which 25 of the 50
are chosen; what I am effectively looking for is the top 25
treatment-control pairs from my sample.
Is it possible to do this in MatchIt with a distance measure
like mahalanobis? The MatchIt package is understandably conservative about
discarding treatment observations, and when I use it to match I generally
end up with around 50 matched treated units. How might I go about this?
Thank you.
--
Best,
Juan Fernando Tellez
PhD Candidate
Department of Political Science
Duke University