-Values for Matched Observational Studies with Unmeasured ConfoundingRosenbaum, 1987b () introduced a sensitivity analysis technique for matched observational studies. Although it works for general matched studies (Rosenbaum, 2002b, ), the theory is most elegant for one-to-one matching. Different from Chapters 17 and 18, Rosenbaum-type sensitivity analysis works best for matched observational studies for testing the sharp null hypothesis of no individual treatment effect.
Consider a matched observational study, with
indexing unit
in pair
.
With exactly matched pairs, unit
and unit
have the same covariates
.
Assume IID sampling, and extend Definition 11.1 to define the propensity score as
![]() |
Let
denote the set of all potential outcomes within pair
.
Conditioning on the event that
, we have
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
![]() |
|||
![]() |
![]() |
Define
as the odds of the treatment for unit
, and we have
![]() |
Under ignorability,
is only a function of
, and therefore,
and
. Thus the treatment assignment mechanism conditional on covariates and potential outcomes is equivalent to that from an MPE with equal treatment and control probabilities. This is a strategy to analyze matched observational studies we discussed in Chapter 75.
In general,
is also a function of the unobserved potential outcomes, and it can range from
to
.
Rosenbaum, 1987b ()’s model for sensitivity analysis imposes bounds on the odds ratio
.
The odds ratios are bounded by
![]() |
for some pre-specified
. Equivalently,
![]() |
for some pre-specified
.
Under Assumption 19.1, we have a biased MPE with unequal and varying treatment and control probabilities across pairs. When
, we have
and thus a standard MPE. Therefore,
measures the deviation from the ideal MPE due to the omitted variables in matching.
-values under Rosenbaum’s sensitivity analysis modelConsider testing the sharp null hypothesis
![]() |
based on the within-pair differences
Under
,
is fixed but
is random if
. Consider the following class of test statistics:
![]() |
where
is a function of
Special cases include the sign statistic, the pair
statistic (up to some constant shift), and the Wilcoxon signed-rank statistic:
![]() |
where
are the ranks of 
What is the null distribution of the test statistic
under Assumption 19.1 with a general
? It can be quite complicated because Assumption 19.1 does not fully specify the exact values of the
’s. Fortunately, we do not need to know the exact distribution of
but rather the worst-case distribution of
that yields the largest
-value under Assumption 19.1. This worst-case distribution corresponds to
![]() |
The corresponding distribution of
has mean
![]() |
and variance
![]() |
with a Normal approximation
![]() |
in distribution.
In practice, we can report a sequence of
-values as a function of
.
We conduct Rosenbaum-style sensitivity analysis in the matched LaLonde data. Using the Matching package, we can construct the matched dataset.
We consider using the test statistic
. Under the ideal MPE with
, we can simulate the distribution of
and obtain the
-value
, as shown in the first subfigure in Figure 23. With a slightly larger
, the worst-case distribution of
shifts to the right, and the
-value increases to
. If we further increase
to
, then the worst-case distribution of
shifts further and the
-value exceeds
. Figure 24 shows the histogram of the
’s and the
-value as a function of
;
measures the maximum confounding that we can still reject the null hypothesis at level
.
We can also use the senmw function in the sensitivitymw package to obtain a sequence of
-values against
.
Figure 24 shows the plot of the
-value against
.
with
IID Bernoulli
, based on the matched LaLonde data.
-value as a function of
, based on the matched LaLonde data.The erpcp dataset is from the R package sensitivitymw. It contains
matched pairs of a welder and a control, based on observed covariates age and smoking. The outcome is the DNA elution rate. Figure 25(a) shows the histogram of the within-pair differences of the outcomes and the
-value against
based on the pair
-statistic. The following R code generates Figure 25(a).
The lead250 dataset is from the R package sensitivitymw. It contains
matched pairs of a daily smoker and a control of nonsmoker, based on observed covariates gender, age, race education level, and household income from NHANES. The outcome is the blood lead level in
. Figure 25(b) shows the histogram of the within-pair differences of the outcomes and the
-value against
based on the pair
-statistic. The following R code generates Figure 25(b).
’s and
-value as a function of 
Assumption 19.1 has the following equivalent form.
The propensity score satisfies the following model
![]() |
where
is an unknown function and
is a a bounded unobserved covariate for unit
.
Remark: Rosenbaum, 2002b (, page 108) also shows that if Assumption 19.1 holds, then Assumption 19.2 must hold for some
’s. Assumption 19.2 may be easier to interpret because
measures the log of the conditional odds ratio of
on the treatment; see Chapter 31.F for the interpretation of the coefficients in the logistic regression.
Re-analyze Example 10.3 using Rosenbaum’s approach based on matching.
Rosenbaum, (2015) provides a tutorial for his two R packages for sensitivity analysis with matched observational studies.
本书:A First Course in Causal Inference,作者 Peng Ding。原文:arXiv:2305.18793v2,按 CC BY 4.0 许可发布。本站(yuanens.com)把 arXiv 网页版改成分章的纯 HTML,把公式渲染成图片,方便在 Kindle 等电子书阅读器上看,内容没有改动。转换难免有错,以原文为准。本站与作者和出版社无关。