Concept Architecture
Concept
Theoretically, Instrumental Variable (IV) is a statistical method used to estimate causal treatment effects when treatment allocation is confounded by unmeasured variables. An instrumental variable is associated with treatment assignment but influences the outcome only through its effect on treatment and is independent of unmeasured confounders. The method exists to reduce bias in observational studies where conventional regression adjustment cannot fully address unmeasured confounding.
Mathematically, instrumental variable analysis estimates causal effects by separating the variation in treatment explained by the instrument from variation attributable to confounding. The classical implementation uses two-stage least squares (2SLS), in which treatment is first predicted from the instrumental variable and then the predicted treatment values are used to estimate the causal effect on the outcome. Valid inference depends on the assumptions of instrument relevance, independence and exclusion restriction.
In practice, instrumental variable methods are widely applied in comparative effectiveness research, real-world evidence studies and health economics when randomisation is unavailable. Common instruments include physician prescribing preference, geographic variation and policy changes. Instrumental variable estimates are frequently used to inform health technology assessment and economic evaluation when observational evidence is required.
Purpose
Used to estimate causal treatment effects from observational data while reducing bias arising from unmeasured confounding.
Mathematical Formulae
Primary Formula
Second-stage model:
Y = ?? + ??X? + �
where:
- Y = outcome
- X? = predicted treatment from the first-stage regression
- ?? = causal treatment effect
- � = random error
Supporting Formulae
First-stage regression:
X = �? + �?Z + u
where:
- X = observed treatment
- Z = instrumental variable
- �? = association between instrument and treatment
Wald estimator (binary instrument):
?IV = (E[Y�Z = 1] ? E[Y�Z = 0]) / (E[X�Z = 1] ? E[X�Z = 0])
Related Mathematical Methods
- Two-Stage Least Squares (2SLS)
- Causal Inference
- Propensity Score Methods
- Difference-in-Differences
- Regression Analysis
- Local Average Treatment Effect (LATE)
Example
A real-world study evaluates the effectiveness of two antihypertensive drugs using physician prescribing preference as an instrumental variable. The first-stage regression predicts treatment assignment from prescribing preference, and the second-stage regression estimates the causal effect on cardiovascular events. The instrumental variable estimate indicates a 12% relative reduction in hospitalisation after accounting for unmeasured confounding, providing evidence for a subsequent cost-effectiveness analysis.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| LINEST | =LINEST(B2:B500,A2:A500,TRUE,TRUE) | Estimate first-stage and second-stage regression coefficients |
| FORECAST.LINEAR | =FORECAST.LINEAR(A2,$B$2:$B$500,$C$2:$C$500) | Predict treatment values from the instrumental variable |
| SLOPE | =SLOPE(B2:B500,C2:C500) | Estimate regression coefficients |
| INTERCEPT | =INTERCEPT(B2:B500,C2:C500) | Estimate regression intercepts |
VBA (Optional)
Automate two-stage least squares estimation by performing sequential first-stage and second-stage regressions and generating causal effect estimates.
Sources
- Angrist JD, Imbens GW, Rubin DB. Identification of Causal Effects Using Instrumental Variables. Journal of the American Statistical Association. 1996.
- Wooldridge JM. Econometric Analysis of Cross Section and Panel Data.
- Hern�n MA, Robins JM. Causal Inference: What If.
- NICE. Health Technology Evaluation Manual.
- ISPOR Good Practice Reports.
Related Concepts (2)
Library
Publications
1
Good Practices for Real-World Data Studies of Treatment and/or Comparative Effectiveness: Recommendations from the Joint ISPOR-ISPE Special Task Force on Real-World Evidence in Health Care Decision Making — Berger, Sox, Willke, Brixner, Eichler, Goettsch, Madigan, Makady, Schneeweiss, Tarricone, Wang, Watkins & Mullins, Vol. 20, No. 8 ed., 2017 (Value in Health)
The joint ISPOR-ISPE recommendations on good procedural practice for real-world data studies (observational studies and registries) used to inform healthcare decisions — study registration, replicability and stakeholder involvement — the reference for RWE credibility in HTA.
Journal ArticleView source →
Frequently Asked Questions (6)
What is an instrumental variable?
A variable associated with the exposure of interest but with no direct effect on the outcome except through that exposure, used to address confounding.
Source: Angrist, Imbens & Rubin 1996
How does an instrumental variable help untangle confounding?
An instrumental variable is a factor linked to the exposure of interest but with no path to the outcome except through that exposure. It helps untangle confounding by isolating the part of the exposure's variation that is driven by the instrument, which is unrelated to the hidden factors that otherwise muddy an observational comparison. Because that slice of variation is effectively as good as random, the effect estimated from it is protected from the confounders that plague ordinary analysis. Borrowing a natural source of randomness is its trick. Hernan and Robins (2020) describe this method.
Source: Hernan & Robins 2020
How does an instrumental variable work?
An instrumental variable works by providing variation in the exposure that is independent of confounders, so that its effect on the outcome, which can only operate through the exposure, reveals the causal effect of the exposure. The estimated effect is derived by relating the instrument's association with the outcome to its association with the exposure. This isolates the part of the exposure's variation driven by the instrument. So an instrumental variable works by using an external source of variation in the exposure, unrelated to confounding, to identify the exposure's causal effect on the outcome, effectively mimicking the variation a randomised experiment would provide.
Source: Angrist & Pischke 2009
What conditions must an instrumental variable satisfy?
An instrumental variable must satisfy three main conditions: relevance, meaning it is associated with the exposure; the exclusion restriction, meaning it affects the outcome only through the exposure and not by any other path; and independence, meaning it is not associated with confounders of the exposure-outcome relationship. Violations of these conditions bias the estimate. So an instrumental variable is valid only when it genuinely influences the exposure, has no direct or other effect on the outcome, and is unrelated to confounders, and because the exclusion restriction in particular cannot be fully tested, the validity of an instrument relies substantially on argument and subject-matter knowledge.
Source: Angrist, Imbens & Rubin 1996
When are instrumental variables used?
Instrumental variables are used when confounding, especially by unmeasured factors, threatens the estimation of a causal effect from observational data, and a valid instrument is available. They are common where randomisation is not possible but a natural source of exposure variation exists, such as policy changes, geographic variation, or, in genetic studies, genetic variants. So instrumental variables are used to estimate causal effects despite unmeasured confounding, which methods relying on measured confounders cannot address, making them valuable in observational research when a credible instrument can be found, though their use depends on the strong and largely untestable assumptions that define a valid instrument.
Source: Angrist & Pischke 2009
What are the limitations of instrumental variables?
The limitations of instrumental variables include that a valid instrument is hard to find and its key assumption, the exclusion restriction, cannot be fully tested, so validity rests on argument; that weak instruments, only weakly associated with the exposure, give imprecise and potentially biased estimates; and that the estimated effect may apply only to the part of the population whose exposure is influenced by the instrument. So instrumental variable estimates are interpreted cautiously, with attention to instrument strength and the plausibility of the assumptions, since a poor or invalid instrument can mislead, and the method addresses confounding only under demanding, not fully verifiable, conditions.
Source: Angrist, Imbens & Rubin 1996
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 8 Dec 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-ES-RWE-005
Stable URI · Machine-readable · Resolvable · CC BY 4.0