COLLEGE POLITICAL SCIENCE • FOUNDATIONS OF POLITICAL SCIENCE

Evaluating Causation — Evaluate causal claims and identify alternative explanations

Learn to distinguish genuine causal relationships from spurious correlations in political research and public discourse.

Historical Context & Motivation

Questions of causation have occupied political thinkers for centuries, from Thucydides asking what caused the Peloponnesian War to contemporary debates about whether economic inequality drives populist voting. The capacity to evaluate causal claims — assertions that one phenomenon produces, triggers, or influences another — is arguably the most consequential analytical skill in political science. Without it, scholars and citizens alike are vulnerable to misleading arguments, poorly designed policies, and ideological manipulation disguised as empirical reasoning.

The systematic study of causation in social inquiry gained momentum during the Enlightenment, when philosophers began insisting that claims about the political world required evidence rather than appeals to divine will or tradition. Over the following centuries, political science gradually adopted more rigorous standards for causal inference, borrowing tools from philosophy, statistics, and experimental science while grappling with the unique challenges of studying human behavior and institutions.

1739
Hume's Problem of Induction
David Hume argued in A Treatise of Human Nature that we never directly observe causation — only constant conjunction. This philosophical challenge remains foundational to how political scientists think about inferring causes from observed patterns.
1843
Mill's Methods of Causal Reasoning
John Stuart Mill formalized five methods for identifying causes, including the Method of Difference and the Method of Agreement. These became essential tools for comparative political analysis and case-study research.
1935
Fisher's Randomized Experiments
R.A. Fisher's The Design of Experiments established randomization as the gold standard for causal inference, influencing how social scientists would later approach field experiments in political behavior.
1986
King, Keohane & Verba's Framework
The publication of Designing Social Inquiry (1994, building on ideas from the 1980s) brought formal causal inference logic to qualitative and quantitative political research, sparking productive methodological debates across the discipline.
2010s
The Credibility Revolution
Political science experienced a surge in natural experiments, regression discontinuity designs, and instrumental variable strategies — collectively known as the credibility revolution — demanding higher evidentiary standards for causal claims.

This intellectual history leaves us with a central question: When someone claims that X causes Y in politics, how do we determine whether that claim is credible, and what alternative explanations might account for the observed relationship? This lesson equips you with a structured framework for answering that question.

Core Principles of Causal Reasoning

Before evaluating any specific causal claim, it is essential to understand the foundational criteria that distinguish genuine causation from mere association. Political scientists generally require that a causal relationship satisfy several conditions simultaneously, and the failure of even one condition opens the door to alternative explanations — competing accounts that could explain the observed pattern without invoking the proposed cause. The following principles structure rigorous causal evaluation across both quantitative and qualitative political research.

1

Covariation

The proposed cause (X) and effect (Y) must vary together systematically. If democratization is claimed to reduce armed conflict, we should observe that as countries become more democratic, conflict rates decline. Covariation is necessary but not sufficient for causation.
2

Temporal Precedence

The cause must precede the effect in time. If a policy was enacted after the outcome already changed, the policy cannot have caused that change. Establishing time order is especially tricky in political science, where policies, institutions, and behaviors evolve simultaneously.
3

Non-Spuriousness

The observed relationship between X and Y must not be the product of a third variable (a confound) that independently influences both. Eliminating confounders is the core challenge of causal inference, because in observational political data, confounders are pervasive and often unmeasured.
4

Causal Mechanism

A plausible causal mechanism — a theory of how and why X produces Y — strengthens the causal claim. Correlations without mechanisms raise suspicion. For example, claiming that education reduces political extremism is more convincing if we can specify how (e.g., exposure to diverse viewpoints, improved critical thinking).
5

Robustness

The relationship should hold across different samples, time periods, and methodological approaches. A causal claim that survives only under one specific set of statistical controls or in one narrow context warrants skepticism and further investigation.
KEY TAKEAWAY
Think of evaluating a causal claim like diagnosing a patient in medicine. A fever (covariation) is a symptom, but you need to rule out alternative diagnoses (confounders), confirm the infection preceded the fever (temporal order), understand the biological pathway (mechanism), and verify that the diagnosis holds across multiple tests (robustness). Jumping from a single symptom to a diagnosis is exactly the mistake people make when they conflate correlation with causation in political arguments.

Mapping Causal Claims Visually

Causal reasoning in political science can be made far more transparent when we represent claims as causal diagrams (sometimes called directed acyclic graphs or DAGs). These diagrams use arrows to show the direction of proposed causal influence and make it immediately visible where confounding variables, mediating variables, and spurious pathways might lurk. The diagram below illustrates three common structural patterns that political scientists must distinguish when evaluating causal claims.

Panel A shows the ideal scenario where X directly causes Y with no confounders. Panel B illustrates spurious correlation, where confounder Z independently causes both X and Y, creating an illusion of causation (shown by the dashed line). Panel C shows mediation, where X causes Y through an intermediate mechanism M. The bottom panel applies the confounding pattern to a classic political science puzzle: the relationship between campaign spending and electoral victory.

The political example in the lower panel illustrates why the confounding pattern (Panel B) is so important for political science. Numerous studies have found a positive correlation between campaign spending and electoral victory, leading many commentators to conclude that money buys elections. However, candidate quality — including charisma, policy positions, and party support — independently attracts donations (more spending) and wins votes (more victories). When researchers like Steven Levitt controlled for candidate quality using incumbent fixed effects, much of the apparent spending effect evaporated. This demonstrates how identifying confounders is not merely a methodological exercise; it has direct implications for campaign finance policy and democratic theory.

The Logic of Causal Inference

While political science does not always rely on mathematical formulas in the same way that physics or economics does, the logic of causal inference can be formalized using the potential outcomes framework (also called the Rubin causal model or the counterfactual framework). This framework provides a precise language for defining what a causal effect actually is, and why it is so difficult to measure in political settings.

The Fundamental Problem of Causal Inference

For any individual unit of analysis — a voter, a country, a legislative district — the causal effect of a treatment (X) on an outcome (Y) is defined as the difference between the outcome that would occur under treatment and the outcome that would occur without treatment. The problem is that we can only ever observe one of these two states for any given unit. This is the fundamental problem of causal inference: the counterfactual outcome is inherently unobservable.

INDIVIDUAL CAUSAL EFFECT
τᵢ = Yᵢ(1) − Yᵢ(0)
Where τᵢ is the causal effect for unit i, Yᵢ(1) is the outcome under treatment (X = 1), and Yᵢ(0) is the outcome without treatment (X = 0). We can never observe both Yᵢ(1) and Yᵢ(0) for the same unit.

Because individual causal effects are unobservable, researchers typically estimate the Average Treatment Effect (ATE) across a population, which compares the mean outcome in treated and untreated groups. However, this comparison only yields a valid causal estimate if the treated and untreated groups are comparable in every respect except the treatment — a condition that randomization achieves but observational data rarely satisfies.

AVERAGE TREATMENT EFFECT
ATE = E[Yᵢ(1)] − E[Yᵢ(0)]
The ATE is the expected value of the individual treatment effect across the population. In a randomized experiment, the difference in group means provides an unbiased estimate. In observational studies, selection bias — driven by confounders — can cause the observed difference to diverge substantially from the true ATE.

Selection Bias and Confounding

OBSERVED DIFFERENCE DECOMPOSITION
E[Y | X=1] − E[Y | X=0] = ATE + Selection Bias
This decomposition reveals a critical insight: the observed difference between treated and untreated groups equals the true causal effect plus any selection bias. Selection bias arises when units that receive treatment differ systematically from those that do not — precisely the confounding problem illustrated in Section 3.
⚙️ Why This Matters for Political Science
Political scientists rarely get to randomly assign treatments. We cannot randomly make some countries democratic and others authoritarian, randomly assign voters to different media diets, or randomly impose economic sanctions on selected nations. This means that virtually every observational study in political science must explicitly address the selection bias term. Methods like matching, regression controls, instrumental variables, and difference-in-differences are all strategies for reducing or eliminating selection bias, but each makes assumptions that must be carefully scrutinized.

Threats to Causal Inference & Alternative Explanations

When evaluating a causal claim in political science, the analyst's task is to identify specific threats to internal validity — reasons why the observed relationship between X and Y might not reflect genuine causation. Each threat corresponds to a distinct type of alternative explanation. Understanding these categories transforms causal evaluation from vague skepticism into structured, productive critique. The following taxonomy, drawing on Campbell and Stanley's influential classification as well as subsequent refinements, organizes the most common threats encountered in political research.

This taxonomy organizes the five major families of threats to causal inference in political science. Each box represents a category of alternative explanations, with specific sub-types and a concrete political example. When evaluating any causal claim, systematically working through these categories helps ensure that no plausible alternative has been overlooked.
Framework for systematically identifying alternative explanations
Threat CategoryCore Question to AskClassic Political Example
ConfoundingIs there a third variable that drives both X and Y?National wealth confounds the relationship between democracy and peace (the democratic peace theory).
Reverse CausationCould Y actually be causing X instead?Media coverage is correlated with terrorism — but does coverage cause terrorism, or does terrorism cause coverage?
Measurement ErrorAre X and Y measured accurately and consistently?Self-reported party identification may not capture actual voting behavior, biasing studies of partisan effects.
Selection BiasAre we only observing a non-representative subset of cases?Studying only countries that experienced civil war to identify causes ignores countries with similar conditions that avoided war.
Temporal / ContextualWas the change in Y already occurring independently of X?Crime rates were falling nationwide before New York's 'broken windows' policing — the policy may have coincided with, rather than caused, the decline.

Worked Example: Evaluating a Causal Claim

Consider the following claim, frequently encountered in political science and public discourse: "Foreign aid causes economic growth in developing countries." We will systematically evaluate this claim using the framework developed in previous sections, identifying each criterion for causation and surfacing plausible alternative explanations.

Does Foreign Aid Cause Economic Growth?
1
Step 1 — Identify the Causal ClaimThe independent variable (X) is the level of foreign aid a country receives; the dependent variable (Y) is economic growth (typically measured as GDP per capita growth). The claim asserts that increasing X leads to increasing Y — that is, more aid produces more growth.
Claim: X (foreign aid) → Y (economic growth)
2
Step 2 — Check CovariationSome cross-national studies find a positive correlation between aid receipts and growth, but others find no relationship or even a negative one. William Easterly's research, for instance, shows that many heavily aided African countries experienced stagnant or declining growth. The evidence for covariation is mixed, which weakens the causal claim at its most basic level.
Covariation: Inconsistent across studies — neither strongly positive nor reliably present.
3
Step 3 — Assess Temporal PrecedenceAid disbursements typically precede measured growth periods, suggesting the time-order criterion is at least superficially met. However, aid allocation decisions are often made in response to anticipated economic conditions (e.g., aid surges after crises), which can blur temporal sequencing. If countries receive aid precisely because they are experiencing economic decline, the temporal relationship between aid and subsequent recovery may be misleading.
Temporal precedence: Partially satisfied, but complicated by endogenous aid allocation.
4
Step 4 — Identify Confounders and Alternative ExplanationsThis is the critical step. Several confounders threaten the claim: (1) Institutional quality — countries with strong governance may attract more aid AND grow faster independently. (2) Geopolitical strategic importance — donors give more aid to strategically important countries, which may also receive trade and investment benefits. (3) Reverse causation — fast-growing countries may attract more aid from donors seeking to back winners, inverting the causal arrow. (4) Selection bias — the poorest countries receive the most aid but face the most structural barriers to growth, creating a misleading negative correlation.
At least four plausible alternative explanations identified: institutional quality, strategic importance, reverse causation, and selection bias.
5
Step 5 — Evaluate the MechanismFor the claim to be persuasive, we need a plausible mechanism linking aid to growth. Proponents argue that aid finances infrastructure, education, and health improvements that raise productivity. Critics counter that aid may fuel corruption, create dependency, or be diverted to non-productive uses. Craig Burnside and David Dollar (2000) found that aid promotes growth only in countries with sound fiscal, monetary, and trade policies — suggesting the mechanism is conditional rather than universal. This conditionality further undermines a blanket causal claim.
Mechanism is plausible but conditional — aid may only cause growth under specific institutional conditions, not universally.
6
Step 6 — Assess RobustnessThe aid-growth relationship has proven notoriously fragile across different specifications, time periods, and country samples. Easterly, Levine, and Roodman (2004) showed that the Burnside-Dollar finding collapsed when additional data years were added. This lack of robustness is a major strike against the causal claim and suggests that the observed relationship is highly sensitive to methodological choices.
Overall assessment: The claim that foreign aid universally causes economic growth fails several criteria for causation. A more defensible claim would be conditional and narrowly specified.
KEY TAKEAWAY
This worked example demonstrates that evaluating a causal claim is not about proving it right or wrong in a single step — it is about systematically interrogating each criterion and asking what specific evidence would be needed to rule out each alternative explanation. The most credible causal claims in political science are those that have survived this gauntlet of scrutiny.

Research Design Strategies: Strengths & Limitations

Political scientists employ a variety of research designs to strengthen causal claims and address the threats identified in previous sections. No single method is universally superior; each design has characteristic strengths in eliminating certain alternative explanations and characteristic vulnerabilities to others. Understanding this trade-off landscape is essential for evaluating published research and designing one's own studies.

Comparative strengths and limitations of major research designs for causal inference
Research DesignKey StrengthsKey Limitations
Randomized Controlled Trial (RCT)Eliminates confounding by design; provides unbiased ATE estimates; strongest internal validity.Often ethically or practically infeasible in politics; limited external validity; Hawthorne effects possible.
Natural ExperimentExploits as-if-random variation in real-world settings; higher external validity than lab RCTs; ethically unproblematic.Researcher must justify that assignment is truly as-if-random; rare and hard to find; limited to specific contexts.
Regression with ControlsWidely applicable to observational data; can include many variables; familiar and accessible method.Cannot control for unobserved confounders; sensitive to model specification; temptation to interpret coefficients causally.
Difference-in-DifferencesControls for time-invariant confounders; intuitive before/after logic; useful for policy evaluation.Requires parallel trends assumption; vulnerable to time-varying confounders; sensitive to treatment timing.
Comparative Case Study (Mill's Methods)Rich contextual detail; can trace causal mechanisms; useful for rare events and process tracing.Small N limits generalizability; case selection bias; researcher degrees of freedom in interpretation.
KEY TAKEAWAY
Think of research designs as lenses with different focal lengths. An RCT provides a perfectly focused close-up of a specific causal effect but loses the broader landscape. A comparative case study captures the panoramic view — context, history, and mechanism — but may lack the sharpness to isolate one variable's effect. The best political science research often combines approaches, using qualitative methods to identify plausible mechanisms and quantitative methods to estimate effect sizes, each compensating for the other's blind spots.

Connecting to Advanced Causal Theory

The framework for evaluating causation presented in this lesson serves as a gateway to more sophisticated theoretical and methodological terrain. As political science has matured, scholars have increasingly engaged with advanced frameworks that refine, extend, and sometimes challenge the basic counterfactual approach. Understanding these connections allows students to appreciate the evolving frontier of causal reasoning in the discipline.

From foundational to advanced causal reasoning
Concept from This LessonAdvanced ExtensionKey Scholars / Works
Counterfactual definition of causationStructural Causal Models (SCMs) — formalize causal diagrams mathematically using do-calculus to determine when causal effects are identifiable from observational data.Judea Pearl, The Book of Why (2018)
Confounding and selection biasSensitivity analysis — quantifies how strong an unmeasured confounder would need to be to overturn a causal finding, rather than assuming all confounders are controlled.Cinelli & Hazlett (2020); Rosenbaum (2002)
Causal mechanismsProcess tracing — a structured qualitative method for testing causal mechanisms by examining within-case evidence for each link in a proposed causal chain.Beach & Pedersen (2019); Bennett & Checkel (2015)
Average Treatment EffectHeterogeneous treatment effects — recognizing that causal effects may differ across subgroups, requiring analyses that move beyond estimating a single average effect.Athey & Imbens (2016); Conditional Average Treatment Effects (CATE)

These advanced frameworks share a common ambition: to make causal reasoning more transparent, more rigorous, and more honest about uncertainty. As you progress in political science, you will encounter these tools in methodology courses and in the published literature. The foundational skills from this lesson — demanding evidence for each criterion of causation, systematically generating alternative explanations, and recognizing the limits of different research designs — will serve as the intellectual scaffolding on which these more technical approaches are built.

🔭 Looking Ahead
In courses on research methods and quantitative analysis, you will learn to formally implement strategies like instrumental variables, regression discontinuity, and synthetic control methods. Each of these is, at its core, an engineered solution to the specific alternative-explanation problems catalogued in this lesson. Mastering the logic of causal inference here prepares you to understand not just how those methods work, but why they are necessary.

Practice Problems

PROBLEM 1CONCEPTUAL
A political commentator claims: "Countries with more internet access are more democratic; therefore, the internet causes democratization." Identify which of the three criteria for causation (covariation, temporal precedence, non-spuriousness) this claim satisfies and which it does not. What is the most obvious alternative explanation?
PROBLEM 2BASIC APPLICATION
Using the potential outcomes framework, define the Average Treatment Effect (ATE) for the following scenario: A researcher wants to know whether exposure to a political debate (treatment) increases voter turnout (outcome). Write out the ATE formula and explain in one sentence why a simple comparison of turnout rates between viewers and non-viewers would not yield the ATE.
PROBLEM 3INTERMEDIATE
A study finds that U.S. states that adopted strict voter ID laws experienced lower turnout among minority voters in subsequent elections. The authors conclude that voter ID laws suppress minority turnout. A critic responds that states adopting strict voter ID laws may also have been passing other restrictive voting measures (e.g., reducing early voting days, closing polling stations) simultaneously. Using the threat taxonomy from this lesson, classify this criticism. Then propose one research design strategy that could help isolate the specific effect of voter ID laws from these co-occurring policies.
PROBLEM 4APPLIED
An international organization publishes a report claiming that UN peacekeeping deployments reduce the recurrence of civil war. Their evidence is that of the 50 post-conflict countries that received peacekeepers between 1990 and 2020, only 15% experienced renewed conflict, compared to 40% of post-conflict countries without peacekeepers. Conduct a full causal evaluation of this claim by (a) identifying at least three specific alternative explanations, (b) classifying each according to the threat taxonomy, and (c) suggesting what evidence would strengthen or weaken the causal claim.
PROBLEM 5CRITICAL THINKING
Some scholars argue that the emphasis on counterfactual causation in political science (the potential outcomes framework) systematically privileges certain types of causal questions — particularly short-term, manipulable treatments — over questions about structural, long-term, or constitutive causes (e.g., "Does capitalism cause inequality?" or "Does patriarchy cause gender gaps in representation?"). Evaluate this critique. Under what circumstances might the counterfactual framework be inadequate for political causal reasoning, and what alternative approaches might complement it?

Lesson Summary

Evaluating causal claims in political science requires checking five interconnected criteria: covariation between the proposed cause and effect, temporal precedence establishing that the cause precedes the effect, non-spuriousness ruling out confounding variables, a plausible causal mechanism explaining how the effect is produced, and robustness across different contexts and methods. The fundamental problem of causal inference — our inability to observe counterfactual outcomes — means that all causal claims in observational political research require explicit strategies for addressing selection bias and alternative explanations.

The five major families of threats — confounding, reverse causation, measurement error, selection bias, and temporal/contextual factors — provide a systematic checklist for generating alternative explanations. Different research designs (RCTs, natural experiments, difference-in-differences, process tracing) each address different subsets of these threats, and no single design eliminates all of them. The most persuasive causal arguments in political science are those that transparently acknowledge potential alternative explanations and marshal evidence — ideally from multiple methods — to rule them out.

Varsity Tutors • College Political Science • Evaluating Causation — Evaluate causal claims and identify alternative explanations