Matchit September 2013

matchit@lists.gking.harvard.edu

4 participants
4 discussions

by yashar atefi

Hi, I receive this message when I try to use my own vector of propensity score in matchit. Error in weights.subclass(psclass, treat) : No units were matched In addition: Warning messages: 1: In if (distance %in% c("GAMlogit", "GAMprobit", "GAMcloglog", "GAMlog", : the condition has length > 1 and only the first element will be used 2: In max(pscore[treat == 0]) : no non-missing arguments to max; returning -Inf 3: In max(pscore[treat == 1]) : no non-missing arguments to max; returning -Inf 4: In min(pscore[treat == 0]) : no non-missing arguments to min; returning Inf 5: In min(pscore[treat == 1]) : no non-missing arguments to min; returning Inf The code is something like mHigh.out <- matchit(training ~ sed+ebrdqual+eprestige+ efullpart,data=cov_H, method="subclass", distance= cov_H$FITVAL,subclass=5, sub.by="all") I really appreciate your help. Best, -- *Yashar Atefi* Doctoral Student in Marketing University of Houston | C. T. Bauer College of Business Department of Marketing & Entrepreneurship 334 Melcher Hall Houston, Texas 77204-6021 Tel: (713) 743 4577 Emails: yatefi(a)bauer.uh.edu Web Page: *http://www.bauer.uh.edu/yatefi/*

10 years, 7 months

Multilevel Propensity Score Matching

by yashar atefi

Hi, I have a data set with three levels: stores, salespeople, and customers. I want to use a multilevel propensity score matching to include covariates at both store and salespeople levels, so I need to use a multilevel logisitic regression to estimate the propensity score. Is there an option that I can specify this type of matching? Thanks a lot, Yashar -- *Yashar Atefi* Doctoral Student in Marketing University of Houston | C. T. Bauer College of Business Department of Marketing & Entrepreneurship 334 Melcher Hall Houston, Texas 77204-6021 Tel: (713) 743 4577 Emails: yatefi(a)bauer.uh.edu Web Page: *http://www.bauer.uh.edu/yatefi/*

10 years, 7 months

Re: [matchit] Dear, Professor Imai

by Kosuke Imai

You can feed your own propensity score into the matchit function. Kosuke Imai Princeton University http://imai.princeton.edu Sent from my iPhone On Sep 10, 2013, at 9:09 AM, "Seungwon Song" <seungwonsong12(a)gmail.com<mailto:seungwonsong12@gmail.com>> wrote: Dear, Professor Imai Allow me to introduce myself. My name is Seungwon Song and I am a Graduate student at Yonsei University in Seoul Korea. I write today to ask some questions about the "MatchIt" package you developed for the R program. I am currently majoring in Evaluation & Assesment and am in the middle of my graduation thesis. The subject of my research is related to the Propensity Score method. The title is <Comparison of Logistic Multilevel Model and Logistic Regression Model in Estimating Propensity Scores and Causal Inference>. The main purpose of my research is to test the effects of the hierarchical data structure on the estimation of the propensity score and the causal effect inference using the Propensity Score Matching Method. To accomplish this task I propose to estimate propensity score with Logistic regression model ignored nested data structure and Logistic multilevel model reflected data properties, after that, to infer causal effect using matching method. Through this procedure, I want to compare the results(estimated propensity score, matching diagnosis, causal effect inference) between logistic regression model and logistic multilevel model to clarify relatively appropriate model. In my study I am utilizing your program matching package the “MatchIt”. But as I have noted above, my study contains the propensity score estimated by the use of logistic multilevel model, one of Generalized linear mixed models. As you may know, Logistic multilevel model can be analyzed by using the glmmPQL function with the “Mass” package and glmer function with “lme4” package in R.(Or using HLM, SAS program) But in your 2011 research on <MatchIt: Nonparametric Preprocessing for Parametric Causal Inference> the glmmPQL & glmer cannot be applied to estimating the Propensity Score. I cannot choose those functions as the “distance” options. During my research I by chance became aware of the fact that it is possible to use another program or another R Package only to estimate the propensity score and put this estimated propensity score in “MatchIt” with the matching process. That is, I want to use “MatchIt” from not estimating propensity score, but matching procedure. I wish to apply this method to my study but find I have no tangible way to go about this task. So here I am. If this is appliable I am very curious as to how this will take shape. Your reply will be a great help in my endeavor to accomplish this task. I eagerly await for your reply. Your's Sincerely, Seungwon Song

10 years, 7 months

problem when matching several controls (ratio>1) and requesting exact match on some variables

by Gilles Hejblum

Hello, We found a problem in the matchit procedure when both 'exact' and 'caliper' features are used with more than 1 match per case (ratio >1). Matchit version : 2.4-21 R version : 3.0.1 The following example allows to reproduce the problem : Here, rows 1/2/3 should only be matched with rows 4/5/6 as an exact match on "sex" is requested. > tmp = data.frame(exposure=c(1,1,1,0,0,0,0,0,0), sex=c(0,0,0,0,0,0,1,1,1), age=c(1,2,3,4,5,6,1,2,3)) exposure sex age 1 1 0 1 2 1 0 2 3 1 0 3 4 0 0 4 5 0 0 5 6 0 0 6 7 0 1 1 8 0 1 2 9 0 1 3 However, the output is as follows : > tmp.match = matchit(exposure~age, replace=TRUE, ratio=3, exact="sex", data=tmp, method = "nearest", caliper = 0.2, verbose = TRUE, calclosest=T) > tmp.match$match.matrix 1 2 3 1 "4" "9" "5" 2 "4" "9" "7" 3 "4" "5" "6" The first match complies with the exact request, but the second and third do not. We traced the error to function match2mearest, line 205 : mindev <- clabels[!clabels%in%match.matrix[itert,(1:r-1)]][min(deviation)==deviation] should be mindev <- clabels[!clabels%in%match.matrix[itert,(1:r-1)] & (matchedc2 == 0)][min(deviation)==deviation] Otherwise, the exact condition in matching is not taken into account when r>1 (i.e. after the first match when several controls are requested). Best regards, -- _______________________________________________ Gilles Hejblum, Inserm-UPMC UMR S 707 / Unite de Sante Publique, Batiment Caroli (Porte 2, rez-de-chaussee), Hopital Saint-Antoine, 184, Rue du Faubourg Saint Antoine, 75571 PARIS CEDEX 12, FRANCE Tel : (33) (1) 49-28-32-28 Fax : (33) (1) 49-28-32-33 Email : gilles.hejblum(a)u707.jussieu.fr _______________________________________________

10 years, 8 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Matchit September 2013