%PDF-1.6
%
573 0 obj
<>
endobj
602 0 obj
<>stream
False
11.0
8.5
24
2019-04-24T14:00:08.826-04:00
SASLaTeX
Lamm, Thompson, Yung
13d249f4fa415440f478f820b8c6e47d864373b5
1605347
<p>Propensity score matching is an intuitive approach that is often used in estimating causal effects from observational data. However, all claims about valid causal effect estimation require careful consideration, and thus many challenging questions can arise when you use propensity score matching in practice. How to select a propensity score model is one of the most difficult questions that you are likely to encounter when you do matching. The propensity score model should consider both the tenability of the assumption of no unmeasured confounding and the covariate balance in the matched data. This paper discusses how you can use the PSMATCH procedure in conjunction with other procedures in SAS/STAT® software to tackle some of these practical challenges. In particular, the paper describes how you can use causal graphs to investigate questions that are related to unmeasured confounding and how you can use the PSDATA statement in PROC PSMATCH to incorporate propensity scores that are computed using approaches other than logistic regression. The paper also illustrates features of PROC PSMATCH that you can use to try to improve covariate balance and control properties of the final matched data set.</p>
<p>Michael Lamm, SAS<br>
Clay Thompson, SAS<br>
Yiu-Fai Yung, SAS</p>
<p style="font-family: Arial,Helvetica,sans-serif; font-size: 90%; line-height: 300%; margin: -3em 0em 2em 0em;"><a href="https://github.com/sascommunities/sas-global-forum-2019">Access sample code</a></p>
Session 3056
en
jeff
<p>Propensity score matching is an intuitive approach that is often used in estimating causal effects from observational data. However, all claims about valid causal effect estimation require careful consideration, and thus many challenging questions can arise when you use propensity score matching in practice. How to select a propensity score model is one of the most difficult questions that you are likely to encounter when you do matching. The propensity score model should consider both the tenability of the assumption of no unmeasured confounding and the covariate balance in the matched data. This paper discusses how you can use the PSMATCH procedure in conjunction with other procedures in SAS/STAT® software to tackle some of these practical challenges. In particular, the paper describes how you can use causal graphs to investigate questions that are related to unmeasured confounding and how you can use the PSDATA statement in PROC PSMATCH to incorporate propensity scores that are computed using approaches other than logistic regression. The paper also illustrates features of PROC PSMATCH that you can use to try to improve covariate balance and control properties of the final matched data set.<br>
</p>
<p><a href="https://github.com/sascommunities/sas-global-forum-2019">Access sample code files now</a></p>
<p>Michael Lamm, SAS<br>
Clay Thompson, SAS<br>
Yiu-Fai Yung, SAS</p>
pdflatex
2019-04-12T11:50:14.000-04:00
2019-04-12T11:50:14.000-04:00
2019-04-07T14:05:10.000-04:00
application/pdf
2019-04-24T14:00:09.129-04:00
Lamm, Thompson, Yung
Propensity score matching is an intuitive approach that is often used in estimating causal effects from observational data. However, all claims about valid causal effect estimation require careful consideration, and thus many challenging questions can arise when you use propensity score matching in practice. How to select a propensity score model is one of the most difficult questions that you are likely to encounter when you do matching. The propensity score model should consider both the tenability of the assumption of no unmeasured confounding and the covariate balance in the matched data. This paper discusses how you can use the PSMATCH procedure in conjunction with other procedures in SAS/STAT® software to tackle some of these practical challenges. In particular, the paper describes how you can use causal graphs to investigate questions that are related to unmeasured confounding and how you can use the PSDATA statement in PROC PSMATCH to incorporate propensity scores that are computed using approaches other than logistic regression. The paper also illustrates features of PROC PSMATCH that you can use to try to improve covariate balance and control properties of the final matched data set.
Access sample code files now
Michael Lamm, SAS
Clay Thompson, SAS
Yiu-Fai Yung, SAS
Building a Propensity Score Model with SAS/STAT® Software: Planning and Practice
uuid:c75fe12b-58ba-425d-89b0-6f048203bc28
uuid:a4ff36d9-7844-4153-9b3d-8b0cb5089ab3
sas
This is pdfTeX, Version 3.14159265-2.6-1.40.18 (TeX Live 2017) kpathsea version 6.2.3
SASLaTeX
false
support:sgf-papers/topic/analytics/statistics
support:sgf-papers
software:STAT
support:skill-level/intermediate
year:2019
event-type:180/session-type/breakout
support:customer-roles/statistician
endstream
endobj
588 0 obj
<>
endobj
574 0 obj
<>
endobj
566 0 obj
<>
endobj
567 0 obj
<>
endobj
568 0 obj
<>
endobj
569 0 obj
<>
endobj
570 0 obj
<>
endobj
131 0 obj
<>
endobj
142 0 obj
<>
endobj
154 0 obj
<>
endobj
162 0 obj
<>
endobj
169 0 obj
<>
endobj
172 0 obj
<>
endobj
173 0 obj
<>stream
xXYs8~ϯ#UAxiqlq#ZxvTęڪ- ju7_h|p{[:LDa4X<A,?a\Ńv`MmNÑ
o(
beClPl1" <-BϋCP=x%b(4LNJuSE6(Dd|7| i|Jdp$tg_mlM(+ڜM͚xsˀ4yhDY[gH&"8̲
#[pbk`@Rw0aB%>UҘS5!>p2xPy4lt4-2ґ>y+fMh?<zNM~ǹ\^ˢ1MhYuUG')T_j̊CwHM/f S. E.*wSoU(cpKL9>DaVɥ2%*1ݺ%I\udK>qj wmV(}SDY!˵ɋ[FO0.0ة
?`2jþKXC"
&SҽCEw Aҳ2+'4cNAz4{`69#DEdR>ﺮ%SһigT4.ϭi0ao]Xcen1m55%f]j4gDvkޑ3w> qYQU`4!:{wEN sieJEsL`rJ~Iӎc$gZ;r]`硇
wRZV{EίEyvOy.#{v#pj+C'(
4bk *K\bKDWd~*xt)Inǣe?,VgUځԁ+JhQ6o?a0ҤoX@*PraJa[pn()lrي-oY\;o ÔEq_K|$s&)}T.`q`?eCs(|~榩AcWV*Rh.C7ꞥPB]Zmq ^
$22
gώvhcv:AL1ha*HyVZX>;ߎ~ ŤWW7Kz?M>Ob[u_t]8wooq9M( ÌmsQblr.]-}av֡lJ8v1)O~c3
3HhNP,Gfg2Jl@ĺI%QL! fWǫ2ge8JF㙈F0@F<~!N1&Mt_B!Y]t-%AqcRNG T.%t: "A z3FG9
];t.˲2NיFqsM8iwf9u$ dK, 6fqA9KRC?#sFw 147=B04z\-C#AӄJ!#) ,h `X=HONBO-{r`Y4$Gqveհ:ٲha܃;e;S`aAhtyc+sMEDPRv[0]j[Q({01
W/fkSC3X6:I&|mLwG7\c
7%]6!K1u17,.FRCh2JISA:\|QV+Lړvnlv(~Cጬ|_T"cd gӛ0wWW?+80%R{ S,,chl4f߭׆W~0x ':~PZ#L4ٲ!
W~-UM1@A7v/יq*D pɆ1l$ ة-T ,'Dj~nw$@JI[v:wQlkuZѫ(D.Nn0AGu Io6 2A~ȓy <
~C*Cfaz:5vТh& '3*ʰm)ag7E5r*H%vEvl/uA4|'I,9 Dÿ'G lCn m,I:,b;F XI,lʼ:ϲ}qf ~QDM(*271 'M;\ZT'
0BI 3]
endstream
endobj
437 0 obj
<>/ProcSet[/PDF/Text]>>
endobj
597 0 obj
<>
endobj
600 0 obj
<>
endobj
469 0 obj
<>
endobj
472 0 obj
<>
endobj
594 0 obj
<>
endobj
473 0 obj
<>
endobj
470 0 obj
[600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600 600]
endobj
192 0 obj
<>stream
xڭweTݒ.\w
4N7nI H]k9sf;kSOS{bN@i'GWfv6~2
ȬvP%@3W+H- v>>>jdm
Rסgdd?=o;! kG ۇ;?ި\m += ',Q `3{= :B +'0% ,oXb mzZ r1`
A `3G:@nx[9M{SuB, gW[VUIt1s+78YEZ:YU߾77pzAͼr9Apŀ Z-_WRO Wފ-[nk#
_"h`g>w 53o$, @+Ve' ,{"/H"FvW6sx\0 뎱7_f {fÿ A#j1G7AXaAA@KU
S۵-`{#Mѿ `fgc7
stwo"MU\WJ\YԿTߴwr~#(9Y/qq'O dp%ego
+ 0x+ѿH9Z8Y5+fon`{Ё@O@mJzk5^ฤAo7;`sQf~n`SW@J_kUK$_{ݸ]Ib?J\u6`V"S>spljƅ$m`@J@l{gtXvzy'4 G#C]W={DfxN]Luy<B0aYN?
]
5
F6SPw-) $[ng \!#Sso5Ձ62c=}ul^ !ۿj?skbн]a.S!v}^4;4lve~cPHƨxDŽr 1D qLFP)R54Z|V:4Q8'lJ@'8Edᙒ)>j^^}SPm@{N⠪7תw2asb] kb˗cK1$ҧƽZ{B9}/@أO'e
8",[$(5ˏh=)"ne&ݴ2ط0ǭPlʱ?ʉ)
P߲P!IV3r^R+Ԛ-o=}ׇkbYX .|BZ5j,XC\SO$Z'Yr÷l<>B0%L5}w:)^WlDRc HyOuz]ߒghgD'Gi27g$JVH&U5&uW{kt>GXQUp4HL*x*KP
l, ~F ng^Lj6g%;vE
>a.x=֖aڎLllFѵrkILH|*n"6'j=ba\:>k!}}əUuLV=冺lCC\7r'OnVtG堜@3M8óIPZƣ]WAA#N{E\!^@lT/opbNT+CG6-1{n ݊K"՝NN\9]CWa:4۫DwPs&D?a7=2T[bW9lowmP* %iw0|vVH/L-'j@,aPWؒ3`wfHz|hr3wM$CG}bµsΟ^o.`/.}sO+Ԡg=㲎/;$Eƛv4vfd
א //$Oġ}, cVBCzYWʯ]1W۪,srMB|1#736͚Fׂ
@2c|lt@Dxy)=vgx7|t.}wO 麍
](D%yMKQ|Lq8Zm=%JU4!#[1i CMq^9ʲq0ygZ'-Q"0`f1iSP{n7>W"9p# >ˁ91iS7#6#E22qK+aLq`FQl})~IهPkԅm}9x)
*nZ)d0k$S\p+'J&ZޝNM ̀"ßIa(K=s*ܼ
|5o۳o[vJT:ͥ+GO~؟,Lf+QAx5s}NRMpZˡ}