Shift-Share Instrumental Variable (Bartik Instrument)
Also known as: Bartik instrument, shift-share instrument, Shift-Share Araç Değişkeni (Bartik Instrument)
The shift-share instrumental variable, widely known as the Bartik instrument, is a causal-inference strategy that builds an instrument by interacting national or sector-level shocks (the shifts) with local composition weights (the shares). Its modern identification framework was set out by Goldsmith-Pinkham, Sorkin and Swift (2020) and Borusyak, Hull and Jaravel (2022).
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the shift-share IV when you want the causal effect of a local treatment (such as employment, trade, or migration exposure) on a continuous outcome across regions or panels, and you can construct credible national or sectoral shocks together with pre-period local composition shares. It needs at least about 50 units and rests on a clear identification claim: either the shares are exogenous or the shifts are exogenous, with a first-stage F-statistic above 10. It is not appropriate when neither shares nor shifts can be argued to be exogenous, or when the first stage is weak.
Strengths & limitations
- Turns aggregate national or sectoral shocks into location-level variation that is plausibly exogenous to local feedback.
- Comes with a transparent modern framework: Rotemberg weights (Goldsmith-Pinkham et al.) show which sectors drive identification, and the shock-level view (Borusyak et al.) clarifies when shifts can carry identification.
- Widely applicable to causal questions in labour, trade, and migration where local exposure mixes a common shock with local structure.
- Identification hinges entirely on either share exogeneity or shock exogeneity; if neither holds the estimate is not causal.
- With fewer than about 50 units the first stage is weak (F < 10) and the IV estimate becomes biased.
- When share exogeneity is violated, Rotemberg weights can turn negative, invalidating the identification story.
Frequently asked
What are the shifts and the shares?
The shifts are common, typically national or sector-level shocks (for example growth rates by industry), and the shares are each location's pre-period composition weights (for example the fraction of local employment in each industry). The instrument is the inner product of the two.
Do I need the shares or the shocks to be exogenous?
You need one of the two. Goldsmith-Pinkham et al. (2020) develop identification from exogenous shares and use Rotemberg weights to show which sectors drive the estimate. Borusyak et al. (2022) develop identification from exogenous shocks, where running the regression at the shock level is cleaner. You should commit to one route and defend it.
What does the first-stage F-statistic tell me?
It measures instrument strength. A first-stage F below 10 signals a weak instrument, which biases the two-stage least squares estimate toward the endogenous OLS result. The framework recommends an F above 10; small samples (under about 50 units) often fail this.
What if my Rotemberg weights are negative?
Negative Rotemberg weights indicate that the share-exogeneity assumption is failing, so the identification strategy is no longer valid. You would then need to argue identification from shock exogeneity instead, or consider an alternative design such as a standard 2SLS or matching approach.
Sources
- Goldsmith-Pinkham, P., Sorkin, I. & Swift, H. (2020). Bartik Instruments: What, When, Why, and How. American Economic Review, 110(8), 2586–2624. DOI: 10.1257/aer.20181047 ↗
- Borusyak, K., Hull, P. & Jaravel, X. (2022). Quasi-Experimental Shift-Share Research Designs. Review of Economic Studies, 89(1), 181–213. DOI: 10.1093/restud/rdab030 ↗
How to cite this page
ScholarGate. (2026, June 1). Shift-Share Instrumental Variable (Bartik Instrument). ScholarGate. https://scholargate.app/en/causal-inference/shift-share-iv
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Regression DiscontinuityCausal inference↔ compare
- Regression Kink DesignCausal inference↔ compare
- Two-Stage Least Squares (2SLS)Causal inference↔ compare