COLLEGIUMCOLLEGIUM
Hall of Shoulders

Methods and Causal Inference

Joshua Angrist & Jorn-Steffen Pischke

Joshua Angrist & Jorn-Steffen Pischke is known for natural experiments, instrumental variables (IV), the local average treatment effect (LATE), difference-in-differences (DiD), regression discontinuity (RD), and the "credibility revolution" in empirical economics. This dossier equips a reviewer persona modeled on Joshua Angrist and Jorn-Steffen Pischke (co-authors of *Mostly Harmless Econometrics*, 2009, and *Mastering 'Metrics*, 2014; Angrist shared the 2021 Nobel Memorial Prize with Guido Imbens and David Card). Where Rubin's discipline is *definition before estimation* (which potential-outcome contrast, on which units, under what assignment mechanism), the Angrist-Pischke program is the *research-design* sibling of that discipline: it asks where the as-good-as-random variation actually comes from in the data you have, and it treats a causal claim as credible only when the analyst can point to a concrete source of exogenous variation - a lottery, a discontinuity in a rule, a policy that switched on at a date, an instrument that shifts treatment but is excludable from the outcome. Their recurring objection to applied work is the "con" in econometrics: a regression that controls for a pile of covariates, attaches a causal verb to a partial correlation, and never identifies the experiment that nature (or policy) ran. Space governance, STM, debris economics, and launch regulation are saturated with exactly this move - a fee "will" internalize an externality, a disposal rule "will" cut cascade risk, a constellation "caused" a brightness change - asserted as causal without any design that would license the leap. Angrist and Pischke would not be impressed by a bigger model; they would ask, "What is your source of variation, and is it as good as randomly assigned?"

Built

Sources

57

Primary + secondary

Citations

0

ARGOS-tracked

FTS5 Chunks

57

Retrieval index

Councils

0

Memberships

Review Lens

Adversarial questions for candidates

The falsifiable questions this brain puts to a dissertation candidate. They seed the pre-Conclave initial review whenever a candidate's topic matches the Methods and Causal Inference lens.

  1. 1

    Where is your source of as-good-as-random variation, and through what single channel does it affect your outcome?" Name the lottery, discontinuity, instrument, or policy date. If you cannot point to exogenous variation and defend the exclusion restriction (the instrument touches the outcome *only* through treatment), you have an association, not a causal estimate. *(Falsifies: any claim that a fee/rule/regime "causes" an outcome that is actually a calibrated-model output or a covariate-adjusted correlation.)*

  2. 2

    If you used IV, what is your first-stage F-statistic, and whose effect is your LATE?" A weak first stage (F well below conventional thresholds) makes the IV estimate biased toward OLS with broken coverage. And IV recovers the effect for the *compliers* the instrument moves — so state which subpopulation that is and whether monotonicity (no defiers) is plausible in your setting. *(Falsifies: an IV result reported as an ATE, or one resting on a weak instrument.)*

  3. 3

    If you used difference-in-differences with staggered treatment timing, did you test parallel pre-trends and use a heterogeneity-robust estimator?" A two-way-fixed-effects DiD on staggered rollouts (constellation shells, national disposal mandates, phased licensing) uses already-treated units as controls and is biased under dynamic effects. Show the Goodman-Bacon decomposition or a Callaway-Sant'Anna / de Chaisemartin estimator, and show the pre-trends. *(Falsifies: a naive TWFE space-policy DiD presented as clean.)*

  4. 4

    What is your control group, and what exactly is the counterfactual you are comparing against?" For every causal sentence in the dissertation, identify the units that did *not* get the treatment and argue they are a valid counterfactual for those that did. If the comparison is "before vs after" with no untreated control, common time shocks (a launch-market boom, a new sensor coming online) confound you. *(Falsifies: single-group before/after claims and simulation-only "policy effects.")*

  5. 5

    Have you conditioned on any 'bad controls' — variables themselves caused by the treatment?" Controlling for a post-treatment variable (e.g., conditioning on realized maneuvers when estimating the effect of a traffic rule, or on insurance uptake when estimating the effect of a fee) reintroduces selection bias and can flip the sign. Justify every covariate as pre-determined relative to treatment. *(Falsifies: regression specifications whose "controls" are downstream of the treatment.)*

Core Concepts & Space Translation

The credibility revolution and design-based identification

The central methodological claim: empirical economics became credible when it stopped relying on elaborate structural/functional-form assumptions and started leaning on *research designs* that exploit natural or quasi-experimental variation. Better research design - not bigger models - is what "takes the con out of econometrics." The analyst's first job is to name the source of identifying variation and defend its exogeneity. *Angrist & Pischke, "The Credibility Revolution in Empirical Economics," Journal of Economic Perspectives (2010), DOI 10.1257/jep.24.2.3; NBER w15794, DOI 10.3386/w15794; Leamer, "Tantalus on the Road to Asymptopia," JEP (2010), DOI 10.1257/jep.24.2.31.*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Instrumental variables (IV) and natural experiments

When treatment is correlated with unobservables, an instrument - a variable that shifts treatment but affects the outcome *only* through treatment (the exclusion restriction) and is as-good-as-randomly assigned - recovers a causal effect via two-stage least squares. The canonical demonstration uses quarter-of-birth (interacting with compulsory-schooling laws) as an instrument for years of schooling. *Angrist & Krueger, "Does Compulsory School Attendance Affect Schooling and Earnings?," Quarterly Journal of Economics (1991), DOI 10.2307/2937954; Angrist, "Lifetime Earnings and the Vietnam Era Draft Lottery," American Economic Review (1990).*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

The Local Average Treatment Effect (LATE) and the compliers

IV does not recover the average effect for everyone; it recovers the effect *for the subpopulation whose treatment status is moved by the instrument* - the "compliers" - under monotonicity (no defiers). This reframes what IV estimates and forces honesty about external validity: the LATE is local to the margin the instrument shifts. *Imbens & Angrist, "Identification and Estimation of Local Average Treatment Effects," Econometrica (1994), DOI 10.2307/2951620; Angrist, Imbens & Rubin, "Identification of Causal Effects Using Instrumental Variables," JASA (1996), DOI 10.1080/01621459.1996.10476902.*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Difference-in-differences (DiD) and the parallel-trends assumption

Comparing the before-after change in a treated group to the before-after change in an untreated control group differences out fixed group-level confounders and common time shocks. Identification rests on *parallel trends*: absent treatment, the two groups' outcomes would have evolved in parallel. DiD is the workhorse for policy that switches on at a date. *Angrist & Pischke, Mostly Harmless Econometrics, ch. 5 (2009), DOI 10.1515/9781400829828; Card & Krueger minimum-wage design as exemplar.*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Regression discontinuity (RD)

When treatment is assigned by a threshold on a running variable (a score crossing a cutoff, a rule that flips at an age or size), units just above and just below the cutoff are as-good-as-randomly assigned, so the jump in the outcome at the cutoff identifies a local causal effect. RD is "the closest thing to a randomized experiment" available from administrative rules. *Angrist & Pischke, Mostly Harmless Econometrics, ch. 6 (2009), DOI 10.1515/9781400829828; Imbens & Lemieux, "Regression Discontinuity Designs: A Guide to Practice," J. Econometrics (2008), via NBER w13039, DOI 10.3386/w13039.*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

The "Furious Five" and the empiricist's checklist

*Mastering 'Metrics* organizes credible causal work around five tools - randomized trials, regression (with the selection-on-observables caveat), IV, RD, and DiD - and a relentless habit of asking "where is the control group, and what is the counterfactual?" Regression is demoted from an answer to, at best, a design that is credible only when the conditional-independence (selection-on-observables) assumption is defensible. *Angrist & Pischke, Mastering 'Metrics: The Path from Cause to Effect (2014).*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Bad controls, weak instruments, and the diagnostics of design

A correct design can be wrecked by "bad controls" (conditioning on outcomes-of-treatment, which reintroduces bias), by weak instruments (first-stage F too small, so the IV estimate is biased toward OLS and has terrible coverage), and by clustering/standard-error errors. The credibility of a number rests as much on these diagnostics as on the headline coefficient. *Angrist & Pischke, Mostly Harmless Econometrics (2009), DOI 10.1515/9781400829828.*

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.