Table 9 summarizes the observed groups and the corresponding latent groups under the monotonicity assumption.
![]() |
![]() |
![]() |
latent groups |
|---|---|---|---|
![]() |
![]() |
![]() |
or a |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
or n |
Interestingly, Assumptions 21.1–21.3 together have some testable implications. Balke and Pearl, (1997) called them the instrumental variable inequalities. This chapter will give an intuitive derivation of a special case of these inequalities. The proof is a direct consequence of identifying the means of the potential outcomes for all latent groups defined by
.
We summarize the main results in Theorem 22.1 below. Define
![]() |
as the proportion of type
, and
![]() |
as the mean of the potential outcome
for type
.
Exclusion restriction implies that
and
. Let
and
denote them, respectively.
Proof of Theorem 22.1: Part I: We first identify the proportions of the latent compliance types. We can identify the proportion of the never takers by
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
![]() |
and the proportion of the always takers by
![]() |
![]() |
![]() |
||
![]() |
![]() |
Therefore, the proportion of compliers is
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
![]() |
|||
![]() |
![]() |
|||
![]() |
![]() |
which is coherent with our previous discussion. Although we do not know individual latent compliance types for all units, we can identify the proportions of never-takers, always-takers, and compliers.
Part II: We then identify the means of the potential outcomes within latent compliance types.
The observed group
only has never takers, so
![]() |
The observed group
only has always takers, so
![]() |
The observed group
has both compliers and always takers, so
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
![]() |
|||
![]() |
||||
![]() |
![]() |
Solve the linear equation above to obtain
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
||||
![]() |
![]() |
The observed group
has both compliers and never takers, so
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
![]() |
|||
![]() |
||||
![]() |
![]() |
Solve the linear equation above to obtain
![]() |
![]() |
![]() |
||
![]() |
![]() |
|||
![]() |
||||
![]() |
![]() |

Based on the formulas of
and
in Theorem 22.1, we can simplify
as
![]() |
which is the same as the formula in Theorem 21.1 before.
Theorem 22.1 focuses on identifying the means of the potential outcomes,
. Imbens and Rubin, (1997) derived more general identification formulas for the distribution of the potential outcomes; I leave the details to Problem 22.3.
Is there any additional value of the detour to derive the formula of
through Theorem 22.1? The answer is yes. For binary outcome, the following inequalities must be true:
![]() |
which implies four inequalities
![]() |
![]() |
![]() |
||
![]() |
![]() |
![]() |
||
![]() |
![]() |
![]() |
||
![]() |
![]() |
![]() |
Rearranging terms, we obtain the following unified inequalities.
Under the IV assumptions 21.1–21.3, the difference in means for
and
must all be non-negative. Importantly, these implications only involve the distribution of the observed variables. Rejection of the IV inequalities leads to rejection of the IV assumptions.
Balke and Pearl, (1997) derived more general IV inequalities with and without assuming monotonicity. The proving strategy above is due to Jiang and Ding, (2020) for a slightly more complex setting. Theorem 22.2 states the testable implications only for a binary outcome. Problem 22.4 gives an equivalent form, and Problem 22.5 gives the result for a general outcome.
For a binary outcome, we can estimate all the parameters by the method of moments below.
We then re-visit two canonical examples with binary data.
Investigators et al., (2014) assess the effectiveness of the emergency endovascular versus the open surgical repair strategies for patients with a clinical diagnosis of ruptured aortic aneurism. Patients are randomized to either the emergency endovascular or the open repair strategy. The primary outcome is the survival status after 30 days. Let
be the treatment assigned, with
for the endovascular strategy and
for the open repair. Let
be the treatment received.
Let
be the survival status, with
for dead, and
for alive. Table 10(a) summarizes the observed data.
Using the IVbinary function above, we can obtain the following estimates:
There is no evidence of violating the IV assumptions.
In Hirano et al., (2000), physicians are randomly selected to receive a letter encouraging them to inoculate patients at risk for flu. The treatment is the actual flu shot, and the outcome is an indicator of flu-related hospital visits. However, some patients do not comply with their assignments. Let
be the indicator of encouragement to receive the flu shot, with
if the physician receives the encouragement letter, and
otherwise. Let
be the treatment received.
Let
be the outcome, with
if for a flu-related hospitalization during the winter, and
otherwise. See Problem 21.7 for more details of the data. Table 10(b) summarizes the observed data.
Using the IVbinary function above, we can obtain the following estimates:
Since
, there is evidence of violating the IV assumptions.
![]() |
![]() | ||||
|---|---|---|---|---|---|
![]() |
![]() |
![]() |
![]() | ||
![]() |
107 | 68 | 24 | 131 | |
![]() |
42 | 42 | 8 | 79 | |
![]() |
![]() | ||||
|---|---|---|---|---|---|
![]() |
![]() |
![]() |
![]() | ||
![]() |
31 | 85 | 30 | 99 | |
![]() |
424 | 944 | 237 | 1041 | |
With a binary outcome, we can define the risk ratio for compliers as
![]() |
Remark: Using Theorem 22.1, we can identify any comparisons between
and
.
This problem extends Theorem 22.1. Define
![]() |
as the density of
for latent stratum
, and define
![]() |
as the density of the outcome within the observed group
.
Exclusion restriction implies that
and
. Let
and
denote them, respectively.
Prove Theorem 22.3 below.
Remark: Imbens and Rubin, (1997) and Kitagawa, (2015) discussed similar results. We can test the first inequality based on an analog of the Kolmogorov–Smirnov statistic:
![]() |
Give an example in which all the IV inequalities hold and another example in which not all the IV inequalities hold. You need to specify the joint distribution of
with binary variables.
Theorem 21.1 relies on randomization, monotonicity, and exclusion restriction. The latter two are not testable even in randomized experiments. When they are violated, the IV estimator no longer identifies the CACE. This problem gives two cases below, which are restatements of Propositions 2 and 3 in Angrist et al., (1996). Recall
and
for
.
Prove Theorem 22.4.
In the process of deriving the IV inequalities in Section 110, we disentangled the mixture distributions by identifying the proportions of the latent strata as well as the conditional means of their potential outcomes. These results help to understand the drawbacks of other seemingly reasonable analyses. I review three estimators below and suppose Assumptions 21.1–21.3 hold.
The as-treated analysis compares the means of the outcomes among units receiving the treatment and control, yielding
![]() |
Show that
![]() |
The per-protocol analysis compares the units that comply with the treatment assigned in treatment and control groups, yielding
![]() |
Show that
![]() |
We may also want to compare the outcomes among units receiving the treatment and control, conditioning on their treatment assignment, yielding
![]() |
![]() |
![]() |
||
![]() |
![]() |
![]() |
Show that they reduce to
![]() |
Extend the discussion in Section 110 based on the notation in Section 108. With the potential outcome
, define the average causal effect of the treatment received on the outcome as
![]() |
and modify the definition of
as
![]() |
due to the change of the notation. They satisfy
![]() |
Section 110 identifies
,
,
,
,
,
and
. But the data do not contain any information about
and
. Therefore, we cannot identify
. With a bounded outcome, we can bound
.
Prove Theorem 22.5.
Remark: In the special case with a binary outcome, the bounds simplify to
![]() |
and
![]() |
Consider a randomized encouragement design where the units assigned to the control have no access to the treatment. For unit
, let
be the binary treatment assigned,
be the binary treatment received, and
be the outcome of interest. One-sided noncompliance happens when
![]() |
Suppose that Assumption 21.1 holds.
[(1)]
Does monotonicity Assumption 21.2 hold in this case?
How many latent strata defined by
are there in this problem? How do we identify their proportions by the observed data distribution?
State the assumption of exclusion restriction. Under exclusion restriction, show that
can be identified by the observed data distributions.
Give the formulas for all possible values of
and
. How do we identify the CACE in this case?
If we observe pretreatment covariates
for all units
, how do we use the covariate information to improve the estimation efficiency of the CACE?
Sommer and Zeger, (1991) provided the following dataset:
![]() |
![]() | ||||
|---|---|---|---|---|---|
![]() |
![]() |
![]() |
![]() | ||
![]() |
9663 | 2385 | 0 | 11514 | |
![]() |
12 | 34 | 0 | 74 | |
The treatment assigned
is whether or not the child was assigned to the vitamin A supplement, the treatment received indicator
is whether or not the child received the vitamin A supplement, and the binary outcome
is the survival indicator. The original RCT was conducted in Indonesia.
Re-analyze the data.
Remark: Bloom, (1984) first discussed one-sided noncompliance and proposed the estimator
, which is sometimes called the Bloom estimator. Bloom, (1984)’s notation is different from this chapter.
Sanders and Karim, (2021, Table 3) reported the following data from an RCT aiming to estimate the efficacy of smoking cessation interventions among individuals with psychotic disorders.
| group assigned | treatment received | group size | # positive outcomes |
|---|---|---|---|
| Control | None | 151 | 25 |
| Treatment | None | 35 | 7 |
| Treatment | Partial | 42 | 17 |
| Treatment | Full | 70 | 40 |
Three tiers of treatment received are defined as follows: “full” treatment corresponds to attending all
treatment sessions, “partial” corresponds to attending
to
sessions, and “none” corresponds to
sessions. The outcome is
defined as the binary indicator of smoking reduction of
or greater relative to baseline, measured at three months.
In this problem, the treatment assignment
is binary but the treatment received
takes three values
for “none”, “partial”, and “full.” The three-leveled
causes complications, but it can only be 0 under the control assignment. How many latent strata
do we have in this problem? Can we identify their proportions?
How do we extend the exclusion restriction to this problem? What can be the causal effects of interest? Can we identify them?
Analyze the data based on the questions above.
Balke and Pearl, (1997) derived more general IV inequalities.
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。