Econometrics Instrumental Variables Calculate Compliers

Estimate the complier share in an instrumental variables framework using the first stage, then optionally compute the intent-to-treat effect and the Local Average Treatment Effect for compliers. This calculator is designed for researchers, students, policy analysts, and applied econometricians who need a fast, transparent way to interpret compliance behavior under the standard IV assumptions.

Interactive IV Complier Calculator

How to calculate compliers in an instrumental variables design

In econometrics, the phrase calculate compliers usually refers to estimating the share of units whose treatment status changes because of the instrument. Under the standard IV framework with a binary instrument Z and binary treatment D, compliers are people who take the treatment when Z = 1 but do not take it when Z = 0. If the instrument is an encouragement, an offer, a policy eligibility rule, or a random lottery, compliers are exactly the group induced into treatment by that instrument.

The practical value of this concept is enormous. In a randomized encouragement design, not everyone offered treatment actually takes it, and some people not offered treatment may still obtain it. Ordinary treatment versus control comparisons can then mix together causal effects, selection, and noncompliance. Instrumental variables solve this by focusing on the exogenous shift created by the instrument. The first stage identifies how strongly the instrument changes treatment uptake, and under monotonicity that first stage is interpreted as the proportion of compliers in the sample.

Complier Share = P(D = 1 | Z = 1) – P(D = 1 | Z = 0)

This equation is simple, but the interpretation depends on a set of assumptions. First, the instrument must be relevant, meaning it moves treatment. Second, the instrument must be as good as randomly assigned or conditionally exogenous. Third, the exclusion restriction must hold, meaning the instrument affects the outcome only through treatment. Fourth, monotonicity rules out defiers, the units who do the opposite of what the instrument encourages. When these conditions hold, the first stage equals the complier proportion, and the Wald estimator recovers the Local Average Treatment Effect for compliers.

Why the first stage equals the complier share

To see the intuition, classify units into four latent compliance types:

• Always-takers: take treatment whether Z = 1 or Z = 0.
• Never-takers: never take treatment regardless of Z.
• Compliers: take treatment if and only if Z = 1.
• Defiers: take treatment if and only if Z = 0.

If monotonicity holds, the defier group is absent. In that case, the only reason treatment uptake is higher when Z = 1 is that compliers move into treatment. Always-takers are treated in both instrument states and never-takers are untreated in both states, so neither group contributes to the difference in treatment rates across instrument values. That means the observed first stage directly reveals the complier share.

Example:

• If 62% are treated when Z = 1, then P(D=1|Z=1) = 0.62.
• If 38% are treated when Z = 0, then P(D=1|Z=0) = 0.38.
• The first stage is 0.62 – 0.38 = 0.24.
• Under monotonicity, about 24% of the sample are compliers.

That estimate can also be translated into counts. In a sample of 10,000 observations, a complier share of 0.24 implies roughly 2,400 compliers. This is often useful when explaining results to nontechnical audiences because a percentage alone can feel abstract, while a headcount makes the treatment-inducement margin easier to understand.

How the calculator works

This calculator asks for the sample sizes in the Z = 1 and Z = 0 groups and the number treated in each group. It computes:

1. Treatment rate when Z = 1: treated with Z = 1 divided by sample size with Z = 1.
2. Treatment rate when Z = 0: treated with Z = 0 divided by sample size with Z = 0.
3. First stage / complier share: treatment rate difference across instrument groups.
4. Estimated number of compliers: complier share multiplied by total sample size.
5. Optional ITT and LATE: if you also enter mean outcomes by instrument status.

The optional outcome inputs are useful because many researchers want to move from compliance to causal interpretation immediately. Once you have the complier share, you can estimate the Local Average Treatment Effect using the Wald ratio:

LATE = [E(Y | Z = 1) – E(Y | Z = 0)] / [P(D = 1 | Z = 1) – P(D = 1 | Z = 0)]

If the instrument raises treatment take-up by 0.20 and raises the average outcome by 4 units, then the estimated treatment effect for compliers is 4 / 0.20 = 20 outcome units. This is not necessarily the average treatment effect for everyone. It is the average causal effect for the subset induced into treatment by the instrument.

Step-by-step interpretation of your results

1. Compare treatment rates across instrument groups

The first thing to check is whether treatment take-up is meaningfully higher when Z = 1. If the difference is close to zero, the instrument may be weak. Weak instruments create unstable IV estimates and wide confidence intervals. In applied work, researchers usually complement the first-stage difference with a first-stage regression and an F-statistic, especially in linear settings.

2. Read the complier share carefully

A complier share of 0.05 means only 5% of the sample changed treatment status because of the instrument. That may still be policy-relevant, but it also means the IV estimate is local to a small margin of behavior. A complier share of 0.30 or 0.40 indicates a much broader behavioral response to the instrument. Larger complier shares generally make the local effect easier to communicate and often improve statistical precision, though design quality still matters more than size alone.

3. Distinguish complier share from treatment rate

Researchers sometimes confuse the treatment rate with the complier rate. They are not the same. You can have high treatment rates in both instrument states and still have a very small complier share if the difference between the two rates is small. Likewise, treatment rates can be modest in both states but still imply a sizable complier group if the gap is large.

4. If you compute LATE, remember it is local

The IV estimate applies to compliers, not necessarily to always-takers, never-takers, or the full population. That is the central insight of the Imbens and Angrist framework. In policy evaluation, this is often a feature rather than a bug because compliers are the people whose decisions can actually be shifted by the policy instrument.

Real-world examples and published statistics

The logic of compliers appears in many of the most influential IV studies in economics, public policy, and health economics. The table below summarizes a few widely cited applications and the kind of first-stage statistics that make complier interpretation possible. These figures are rounded summaries from published work and are intended to illustrate scale and interpretation rather than replace the original papers.

Study / Setting	Instrument	Approximate first-stage statistic	Interpretation for compliers
Oregon Health Insurance Experiment	Medicaid lottery selection	Lottery selection increased Medicaid coverage by about 0.25, or 25 percentage points	Roughly one quarter of the sample were induced into coverage by the lottery offer
Angrist and Evans fertility study	Same-sex composition of first two children	Probability of having a third child rose by roughly 0.06 to 0.07 in many samples	The IV estimate pertains to families whose fertility decisions changed because of sibling sex mix
Quarter-of-birth schooling studies	Compulsory schooling laws interacted with quarter of birth	Schooling increased by about 0.1 years in classic specifications	The return-to-schooling estimate is local to students whose education changed at the compulsory schooling margin

These examples show why calculating compliers matters. The same numerical IV estimate can imply very different policy stories depending on the first-stage size and on who the compliers are. A 10% earnings gain in a quarter-of-birth design speaks to marginal students affected by school-leaving laws; a utilization effect in the Oregon lottery speaks to adults who enroll in public coverage when randomly offered access.

Comparison of first-stage magnitudes

The next table translates first-stage magnitudes into the implied number of compliers in a common sample size of 10,000. This is a practical communication device that many applied researchers use in presentations, referee responses, and policy briefs.

First-stage difference	Complier share	Implied compliers in a sample of 10,000	Typical interpretation
0.05	5%	500	Narrow inducement margin, often raises weak-instrument concerns
0.15	15%	1,500	Moderate policy responsiveness with clearer local interpretation
0.25	25%	2,500	Large induced behavioral shift, common in strong encouragement designs
0.40	40%	4,000	Very strong first stage, usually easy to explain substantively

Common mistakes when trying to calculate compliers

• Ignoring monotonicity: without monotonicity, the first stage reflects compliers minus defiers, not compliers alone.
• Using raw treatment differences without checking instrument assignment: the relevant comparison is treatment uptake across instrument values, not simply treated versus untreated groups.
• Forgetting that LATE is local: the estimate does not automatically generalize to always-takers or never-takers.
• Overlooking weak instruments: a small first stage may produce noisy and biased finite-sample IV estimates.
• Assuming exclusion without justification: if the instrument affects outcomes through any channel besides treatment, the causal interpretation breaks down.

Practical checklist for applied researchers

1. State the instrument clearly and explain why it changes treatment uptake.
2. Report treatment rates by instrument group.
3. Compute and interpret the first stage as the complier share under monotonicity.
4. Provide the reduced form or ITT effect on outcomes.
5. Estimate the Wald ratio or 2SLS effect.
6. Discuss who the compliers are in substantive terms.
7. Report robustness checks, balance tests, and weak-instrument diagnostics.

Authoritative resources for deeper study

If you want to go beyond a calculator and understand identification, assumptions, and implementation in more depth, these resources are useful starting points:

• Penn State STAT course notes on instrumental variables
• National Institutes of Health overview on instrumental variable methods
• U.S. Census Bureau working paper discussing IV methods and causal inference

Bottom line

To calculate compliers in a binary IV setting, compute the treatment rate among those with Z = 1, subtract the treatment rate among those with Z = 0, and interpret that difference as the complier share under monotonicity. If you also know how the instrument shifts outcomes, divide the outcome difference by the first stage to recover the Local Average Treatment Effect. That is the core logic behind some of the most important empirical strategies in modern econometrics. A good calculator makes the arithmetic immediate, but the real value comes from careful interpretation: who is induced, why the instrument is credible, and whether the resulting LATE answers the policy question you actually care about.