27 Jan
2011
27 Jan
'11
8:09 p.m.
Hello,
I am using nearest matching with ratio of 20 or 50 and a caliper. When I do
this I get the message that not every treated observation has the requested
number of matches. This is OK for my application, but the weights of the
controls are all 1.
I would like the weights to reflect the number of controls for each treated
observation, That is, control weight=1/K where K is the number of non-NAs in
the match.matrix in the row where the control obsID occurs. Otherwise the
weights do not properly reflect the balance achieved by matching.
Is there a simple way to compute this weight vector from the match matrix?
Thanks,
Chris
27 Jan
27 Jan
8:15 p.m.
I'd have a look at CEM, which produces weights like these.
Gary
--
*Gary King* - Albert J. Weatherhead III University Professor - Director,
IQSS - Harvard University
GKing.Harvard.edu <http://gking.harvard.edu/> - King(a)Harvard.edu -
@kinggary<http://twitter.com/kinggary>- 617-500-7570 - Asst 495-9271 -
Fax 812-8581
On Thu, Jan 27, 2011 at 8:09 PM, Chris Hane <christopher.a.hane(a)gmail.com>wrote;wrote:
Hello,
I am using nearest matching with ratio of 20 or 50 and a caliper. When I
do this I get the message that not every treated observation has the
requested number of matches. This is OK for my application, but the weights
of the controls are all 1.
I would like the weights to reflect the number of controls for each treated
observation, That is, control weight=1/K where K is the number of non-NAs in
the match.matrix in the row where the control obsID occurs. Otherwise the
weights do not properly reflect the balance achieved by matching.
Is there a simple way to compute this weight vector from the match matrix?
Thanks,
Chris
9:02 p.m.
You would have to do a bit of R programming here. First, for each treated unit, you can
compute the number of matched controls. This can be done by apply(!is.na(match.matrix),
1, sum). Then, for each control unit, you check which treated unit it was matched to (you
probably need to use a loop and "%in%").
Kosuke
On Jan 27, 2011, at 8:09 PM, Chris Hane wrote:
Hello,
I am using nearest matching with ratio of 20 or 50 and a caliper. When I do this I get
the message that not every treated observation has the requested number of matches. This
is OK for my application, but the weights of the controls are all 1.
I would like the weights to reflect the number of controls for each treated observation,
That is, control weight=1/K where K is the number of non-NAs in the match.matrix in the
row where the control obsID occurs. Otherwise the weights do not properly reflect the
balance achieved by matching.
Is there a simple way to compute this weight vector from the match matrix?
Thanks,
Chris
--
MatchIt mailing list served by Harvard-MIT Data Center
List Address: matchit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit
MatchIt Software and Documentation: http://gking.harvard.edu/matchit/
Browse/Search List Archive: http://lists.hmdc.harvard.edu/lists/matchit/
28 Jan
28 Jan
2:59 p.m.
thanks, yes it was easier than I thought.
There may be a simpler method but flattening the match.matrix to a vector
and merging with row sums seemed easiest.
makeMatchItWeights <- function(x)
{
stopifnot(inherits(x, "matchit"))
m <- x$match.matrix # each row is 1 treated person, elements are the matched
control IDs
# columns in matrix are maximum # of matches per treated
max.matched <- ncol(m)
# make matrix into vector of controlIDs
# each consecutive block of max.matched are the controls for 1 treated
# treated IDs as repeated vector to align with control IDs
mdf <- data.frame(tid=rep(rownames(m),
each=max.matched),cid=as.vector(t(m)))
mdf<-mdf[complete.cases(mdf),]
# make df of the row counts
rowmm <- ifelse(is.na(x$match.matrix), 0,1)
r<- rowSums(rowmm)
rms <- data.frame(rid=names(r), s=r)
mdfs<- merge(mdf, rms, all.x=T, by.x="tid", by.y="rid")
mdfs$w<-1/mdfs$s
mdfs
}
On Thu, Jan 27, 2011 at 6:02 PM, Kosuke Imai <kimai(a)princeton.edu> wrote:
You would have to do a bit of R programming here.
First, for each treated
unit, you can compute the number of matched controls. This can be done by
apply(!is.na(match.matrix), 1, sum). Then, for each control unit, you
check which treated unit it was matched to (you probably need to use a loop
and "%in%").
Kosuke
On Jan 27, 2011, at 8:09 PM, Chris Hane wrote:
Hello,
I am using nearest matching with ratio of 20 or 50 and a caliper. When I
do this
I get the message that not every treated observation has the
requested number of matches. This is OK for my application, but the weights
of the controls are all 1.
I would like the weights to reflect the number of controls for each
treated
observation, That is, control weight=1/K where K is the number of
non-NAs in the match.matrix in the row where the control obsID occurs.
Otherwise the weights do not properly reflect the balance achieved by
matching.
Is there a simple way to compute this weight vector from the match
matrix?
Thanks,
Chris
4859
days inactive
4859
days old
matchit@lists.gking.harvard.edu
3 comments
3 participants
participants (3)
-
Chris Hane -
Gary King -
Kosuke Imai