Matchit October 2015

matchit@lists.gking.harvard.edu

14 participants
11 discussions

Adjusting SEs after matching with replacement

by Pavelle, Bridget J (DSHS/RDA)

Hi all, Planning on matching with replacement in MatchIt then bringing data back into SAS for regression analysis of matched pairs. Could someone please remind me of the standard approach to adjusting standard errors in post-matching analysis where some controls are used more than once? Thank you, Bridget ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Bridget Pavelle, Ph.D. Senior Research Manager Program Research and Evaluation Section Research and Data Analysis Division WA Department of Social and Health Services PO Box 45204 Olympia, WA 98504-5204 Fax: (360) 902-0705 Email: bridget.pavelle(a)dshs.wa.gov<mailto:bridget.pavelle@dshs.wa.gov> Telephone: (360) 902-0741

8 years, 6 months

Fwd: Specify groups in matchit function

by Susan Kamal

Hi Liz, here is the output from match.data(). As you can see the pairs are from the same centre "50". Also sometime spairs have boh Interv=1 or Inter=0. I want that for each pair, one has Centre=50 and Interv=1 or Centre=40 and Interv=0. The intervention was only done at centre =50. Idnum Centre Interv LTFU_6m LTFU_12m VL_BL log_VL_BL CD4_BL Dur_HIV SEX ETHN EDUC hepC hepB IVD 1 18209 50 1 . . 330 2.52 304 2976 1 0 0 1 0 898989 3 19276 50 1 3/2/2012 . 48700 4.69 491 1122 1 0 1 0 0 0 4 25339 50 0 4/24/2006 . 730000 5.86 378 2598 1 0 1 0 0 898989 6 42002 50 0 3/29/2010 . 252000 5.40 320 2321 1 1 0 0 1 898989 8 50331 50 1 2/16/2010 . 53 1.72 316 6708 1 0 1 0 1 898989 9 50435 50 0 11/19/2010 . 2190 3.34 322 8058 1 0 0 1 1 898989 11 50455 50 1 . . 54 1.73 960 8454 2 0 0 1 0 898989 13 50895 50 0 7/7/2010 . 47100 4.67 202 898989 2 1 0 0 1 898989 14 50964 50 1 11/7/2008 . 10000 4.00 63 898989 1 0 0 0 1 898989 15 51082 50 1 . . 29400 4.47 292 898989 2 0 0 0 1 898989 17 51544 50 0 3/3/2008 . 78 1.89 352 3058 2 1 0 0 1 898989 18 51559 50 0 2/29/2008 . 48 1.68 241 3296 1 0 0 0 0 898989 19 51613 50 0 . . 31200 4.49 268 3831 2 0 0 1 0 898989 20 51651 50 0 6/18/2008 . 645000 5.81 219 2963 2 1 0 0 1 898989 21 51654 50 0 5/4/2011 . 253 2.40 386 4471 2 0 0 1 0 898989 23 51673 50 0 10/28/2009 . 1390 3.14 350 2134 1 0 0 1 1 898989 25 51692 50 1 10/11/2007 10/11/2007 227000 5.36 253 2290 1 1 0 0 0 898989 27 51705 50 1 . . 56800 4.75 450 3671 2 0 0 1 0 898989 28 51708 50 0 . . 139000 5.14 301 7496 1 0 0 1 1 898989 30 51736 50 0 . . 312000 5.49 133 3652 1 0 0 0 0 898989 On Tue, Oct 27, 2015 at 7:24 PM, Elizabeth Stuart <estuart(a)jhu.edu> wrote: > Hi Susan, > I’m not sure I understand your question but a few thoughts: > > - The function should be giving you pairs, one of whom has Interv=0 and > the other with Interv=1; I wonder if you are looking somehow only at the > first member of each pair? If you use the match.data() command after the > matchit call it should give you back all of the subjects that were put into > a pair (those with interv=1 and interv=0). > > - I’m not sure I understand how you want to deal with center, but if you > want each matched pair to have 2 people that came from the same center > (what I would call an “exact match” on center; i.e., two individuals can > only be matched if they are from the same center), you can do that using > the “exact=“ option in the matchit call. > > Liz > > > > > > On Oct 27, 2015, at 3:54 AM, Susan Kamal <susan.kamal(a)gmail.com> wrote: > > Hello, > > How can I indicate in my matchit function that my matched pairs have to be > one pair from the intervention group and another from the control group? > > In my dataset I have a variable= Interv which is a binary variable that > indicates that a person is assigned to an intervention or not. > > I have anotehr variablecalledcentre, also binary, and I want that my > matched pairs each come from one of the 2 cetres. > > m.out <- matchit(Interv ~ SEX + ETHN + EDUC + log_VL_BL + CD4_BL + Dur_HIV > + hepC + hepB + IVD, data = mydata, method= "nearest") > > This function just gave out all 195 paris form the same centre and with > Interv=1. > > Thanks, > Susan > > -- > Regards, > Susan Kamal > - > --- > 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 > > > -- Regards, Susan Kamal -- Regards, Susan Kamal

8 years, 6 months

Specify groups in matchit function

by Susan Kamal

Hello, How can I indicate in my matchit function that my matched pairs have to be one pair from the intervention group and another from the control group? In my dataset I have a variable= Interv which is a binary variable that indicates that a person is assigned to an intervention or not. I have anotehr variablecalledcentre, also binary, and I want that my matched pairs each come from one of the 2 cetres. m.out <- matchit(Interv ~ SEX + ETHN + EDUC + log_VL_BL + CD4_BL + Dur_HIV + hepC + hepB + IVD, data = mydata, method= "nearest") This function just gave out all 195 paris form the same centre and with Interv=1. Thanks, Susan -- Regards, Susan Kamal

8 years, 6 months

From SAS to MatchIt

by Susan Kamal

Hello, I am reworking an analysis that was done in SAS in R now using Matchit (I want to learn how to use MatchIt). Is there anyone on this list who used SAS to do propensity score matching and can help out? What is the equivalent of greedy matching macro in SAS in R/Matchit? Can non parcimonious logistic model be done using MatchIt? Thanks, Susan -- Regards, Susan Kamal

8 years, 6 months

from: Francois Maurice

by Francois Maurice

8 years, 6 months

Controlling for preexposure PS model covariates in final models for outcomes?

by Sharon Simonton

Hi All, I have a question regarding the final statistical analyses used to compare outcomes for matched cases and controls. I've noticed that several of you who have worked to develop and support MatchIt introduce the pre-exposure covariates used for the estimation of the propensity scores into the final models used to compare outcomes. I hadn't seen this done prior to reading several of your papers. If matching has indeed successfully eliminated all observable differences between cases and controls (ignorability) and all confounding and bias has truly been taken into account -- how would you explain the rationale for doing this? Would the "residual confounding" that this is supposed to address be the extent to which the propensity score pre-exposure covariates predict the outcome above and beyond their relationship with the treatment condition? I'll appreciate any thoughts you may have on this. Thank you in advance for your help. Best wishes, Sharon

8 years, 6 months

Dataset and code please

by Susan Kamal

Hello, I want to learn how to use MatchIt so can anyone send me any dataset and code to train myself? Thanks, Susan -- Regards, Susan Kamal

8 years, 6 months

Query pertaining to R-matchIT

by swapnil parkhe

Hi all, I am a graduate from *Indian Institute of Technology (IIT) Roorkee*, India ( http://www.iitr.ac.in/ <http://mailstat.us/tr/t/rj53gdczvvx6ri1gszrlj/4/http://www.iitr.ac.in/>). Currently, I am working in a company which is germane to the Data Science field. I am working on R-matchIT for one of our projects and while researching on the internet about R-matchIT, I came across your paper "MatchIt: Matching Software for Causal Inference" and found it extremely helpful. Although, I have some doubts regarding R-macthIT pertaining to you paper, which if you can help me resolve will be really helpful. Currently, the project that I deal with, only has categorical covariates (around 5-7 variables: nominal and ordinal only). Below are the doubts pertaining to that: 1. Is there any pre-requisite condition that needs to be adhered to? (like ratio of number of treated and control units, currently its almost 1:1 for me) 2. Can matchIT be used for categorical covariates? If not, what transformations need to be made? 2. Which method (nearest neighbour, subclass, optimal, etc.) will suit the best for the analysis? 3. How do I study the goodness of the match? I am really looking forward to hear back from you. Thanks in advance. Thanks, Swapnil Parkhe

8 years, 6 months

A question about jitter plots

by Ahmadreza Talebian

Hello all, I have a basic question about jitter plots in MatchIt. What does the vertical axis represent? Some points are located above some other points. What does it imply? Many thanks, Ahmadreza -- Ahmadreza Talebian PhD Candidate in Transportation Engineering Department of Civil and Materials Engineering University of Illinois at Chicago 950 S. Halsted St., #4266 Chicago, IL 60607

8 years, 6 months

Question about mahalanobis distance matching

by Vincent

Dear All, I am trying to replicate the results from Dr. King's paper: Why PS should not be used for Matching. Somehow I do not think Matchit with mahalanobis distance is giving me what I want Below is the code: n<-100 x1_contr<-runif(n,0,5) x2_contr<-runif(n,0,5) x_contr<-cbind(x1=x1_contr,x2=x2_contr) x1_treat<-runif(n,1,6) x2_treat<-runif(n,1,6) x_treat<-cbind(x1=x1_treat,x2=x2_treat) T<-c(rep(0,n),rep(1,n)) X.all<-rbind(x_contr,x_treat) my.data<-data.frame(T,X.all) rownames(my.data)<-paste("ID",1:dim(my.data)[1]) mdm.out<-matchit(T~x1+x2,data=my.data,distance=mahalanobis,mahvars =c("x1","x2"), caliper=0.05,replace=FALSE) summary(mdm.out) Sample sizes: Control Treated All 100 100 Matched 100 100 Unmatched 0 0 Discarded 0 0 How can all cases and controls be matched? Thanks, Vincent

8 years, 7 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Matchit October 2015