Written and maintained by CASRAI Editorial Board
Published
On this page
Process tracing is an empirical methodology designed to unpack and test causal mechanisms within case studies. Researchers employing a case study research method face the challenge of determining not merely whether an independent variable correlates with an outcome, but through which specific intermediate steps that cause generates the effect. By establishing an explicit chain of diagnostic events, process tracing links qualitative analysis with formal Bayesian probability.
Definition in One Paragraph
Process tracing is a qualitative and mixed-method within-case research method designed to identify, unpack, and empirically evaluate the causal mechanisms linking an initial condition or independent variable to a final outcome. Rather than measuring average treatment effects or cross-case statistical correlations across large sample sizes, process tracing investigates the continuous, interlocking chain of events, entities, and intermediate activities operating within a single bounded case. By systematically gathering diagnostic evidentiary traces—termed causal process observations (CPOs)—investigators test hypothesized mechanisms against competing explanations using formal logical tests and Bayesian updating principles, allowing researchers to evaluate whether an observed outcome was produced by a hypothesized causal sequence or alternative historical explanations.
Formula and Notation Explained
In formal methodology, process tracing is operationalized through Bayesian inference, treating probability as a degree of belief given available empirical information. When evaluating a hypothesized causal mechanism (H₁) against an alternative hypothesis (H₂) in light of an observed causal process observation (E), Bayes’ Rule in odds form is defined as:
$$frac{P(H_1 mid E, I)}{P(H_2 mid E, I)} = frac{P(H_1 mid I)}{P(H_2 mid I)} times frac{P(E mid H_1, I)}{P(E mid H_2, I)}$$
Every term in this formulation represents a distinct methodological component:
- H₁: The primary hypothesized causal mechanism connecting cause to outcome.
- H₂: The rival explanatory hypothesis for the observed outcome.
- E: The diagnostic causal process observation (CPO) uncovered during fieldwork or archival examination.
- I: Background information and domain knowledge preceding evidence E.
- P(H₁) / P(H₂): Prior odds ratio comparing the plausibility of H₁ versus H₂ prior to observing evidence E.
- P(E | H₁, I) / P(E | H₂, I): The Bayes Factor or Likelihood Ratio (LR), measuring the probability of observing evidence E under H₁ relative to H₂.
- P(H₁ | E, I) / P(H₂ | E, I): Posterior odds ratio representing updated relative support after observing E.
Converting posterior odds into posterior probability P(H₁ | E, I):
$$P(H_1 mid E, I) = frac{text{Posterior Odds}}{1 + text{Posterior Odds}} = frac{P(E mid H_1, I) P(H_1 mid I)}{P(E mid H_1, I) P(H_1 mid I) + P(E mid H_2, I) P(H_2 mid I)}$$
Methodologists classify evidentiary tests into four diagnostic standards based on necessity and sufficiency:
- Straw-in-the-Wind: Neither necessary nor sufficient. Passing slightly raises plausibility (LR > 1); failing slightly lowers it. Does not confirm or eliminate either hypothesis.
- Hoop Test: Necessary but not sufficient. Hypothesis H₁ must pass to survive (P(E | H₁, I) ≈ 1.0). Passing affirms viability; failing decisively eliminates H₁.
- Smoking-Gun Test: Sufficient but not necessary. Passing provides strong confirmation because the evidence is improbable under rivals (P(E | H₂, I) ≈ 0, LR >> 1). Failing does not eliminate H₁.
- Doubly Decisive Test: Both necessary and sufficient. Passing confirms H₁ and eliminates rivals; failing eliminates H₁.
Worked Example with Arithmetic Shown
Consider an empirical study evaluating why a regional government enacted a stringent decarbonization statute. A researcher employing a structured research approach investigates two competing hypotheses:
- H₁ (Advocacy Mobilization): Sustained coalition lobbying by local environmental groups directly drove legislative enactment.
- H₂ (Fiscal Grant Incentive): Capital funding from a national infrastructure grant prompted enactment independently of civic lobbying.
The investigation follows five computational steps:
Step 1: Specify Prior Odds
Equal prior plausibility is assigned based on preliminary literature:
$$P(H_1 mid I) = 0.50, quad P(H_2 mid I) = 0.50, quad text{Prior Odds} = frac{0.50}{0.50} = 1.00$$
Step 2: Evaluate Causal Process Observation (E)
Archival discovery yields internal cabinet minutes showing ministers directly adopted statutory clauses drafted by the environmental coalition:
- P(E | H₁, I) = 0.85: Under advocacy lobbying, finding deliberations adopting coalition clauses has an 85% probability.
- P(E | H₂, I) = 0.10: If fiscal grants were the sole driver, deliberations centering on civic draft clauses has only a 10% probability.
Step 3: Calculate Likelihood Ratio (Bayes Factor)
$$text{Likelihood Ratio (LR)} = frac{P(E mid H_1, I)}{P(E mid H_2, I)} = frac{0.85}{0.10} = 8.50$$
Step 4: Compute Posterior Odds
$$text{Posterior Odds} = text{Prior Odds} times text{LR} = 1.00 times 8.50 = 8.50$$
Step 5: Convert to Posterior Probability
$$P(H_1 mid E, I) = frac{8.50}{1 + 8.50} = frac{8.50}{9.50} approx 0.8947, quad P(H_2 mid E, I) = 1 – 0.8947 = 0.1053$$
Finding the executive minutes updates the probability that advocacy mobilization drove policy enactment from 50.00% to 89.47%, substantially ruling in H₁ over H₂.
Interpretation and Reading Guidance
Interpreting process tracing findings requires assessing the diagnostic power of individual evidentiary tests across the causal chain:
| Test Type | Logical Criterion | Likelihood Ratio (LR) | If Evidence Passes | If Evidence Fails |
|---|---|---|---|---|
| Straw-in-the-Wind | Neither necessary nor sufficient | Moderate (1.0 < LR < 3.0) | Affirms plausibility; does not confirm | Weakens plausibility; does not eliminate |
| Hoop Test | Necessary, not sufficient | Constrained (LR > 1.0; P(E|H₁) ≈ 1.0) | Affirms viability; does not prove H₁ | Decisively eliminates hypothesis H₁ |
| Smoking-Gun | Sufficient, not necessary | Substantial (LR >> 1.0; P(E|H₂) ≈ 0) | Substantially confirms hypothesis H₁ | Leaves H₁ viable; does not eliminate |
| Doubly Decisive | Both necessary and sufficient | Maximal (P(E|H₁) ≈ 1.0, P(E|H₂) = 0) | Proves H₁ and eliminates rival H₂ | Decisively eliminates H₁ |
Passing a hoop test confirms only that a hypothesis survives an essential condition; it does not warrant claiming that the mechanism has been proven.
Assumptions and Common Mistakes
Rigorous application of process tracing relies on key analytical assumptions:
- Assumption of Mechanismic Continuity: The causal chain must possess continuous temporal and conceptual links between cause and outcome. Gaps where intermediate steps are asserted without evidence undermine inferential validity.
- Evidentiary Independence: Bayesian updating assumes observations are conditionally independent. Treating multiple reports originating from one press release as separate observations artificially inflates posterior probabilities.
- Equifinality: Multiple causal pathways may produce identical outcomes. Demonstrating that mechanism A operated does not automatically disprove that mechanism B contributed concurrently.
- Conflating Necessity with Sufficiency: A common error involves treating a passed hoop test as definitive proof, mistaking baseline survival for causal confirmation.
- Confirmation Bias in Evidence Search: Searching only for documents affirming the favored hypothesis while neglecting checks for disconfirming records distorts likelihood assessments.
How to Report It (APA Example)
When reporting process tracing analyses in scholarly publications, authors must document prior probabilities, the evidential basis for likelihood assignments, and resulting posterior calculations. In studies combining fieldwork with participant observation or archival analysis, systematic reporting ensures transparency.
An exemplary reporting format in APA 7th style is structured as follows:
To evaluate the causal mechanism linking civic coalition lobbying (H₁) to municipal policy adoption against the rival fiscal incentive explanation (H₂), we conducted Bayesian process tracing following Collier (2011) and Fairfield and Charman (2017). Setting neutral prior probabilities (P(H₁) = .50, P(H₂) = .50; prior odds = 1.00), we evaluated internal cabinet strategy minutes (Observation E-1). Because deliberations explicitly referenced coalition statutory text and mapped electoral pressure, we estimated P(E-1 | H₁) = .85 and P(E-1 | H₂) = .10, yielding a likelihood ratio (Bayes factor) of 8.50. Under Bayesian updating, the posterior odds of 8.50 correspond to a posterior probability of P(H₁ | E-1) = .89, indicating strong empirical support for the advocacy mechanism over the fiscal incentive hypothesis.
When structuring a case study paper, researchers should accompany narrative text with an evidence summary matrix detailing each causal process observation, archival source, test type, and assigned likelihood ratio.
How to Calculate It in Excel, R, or Python
Researchers can implement Bayesian process tracing updating routines using standard computational tools.
Excel Formula Implementation
B1: P(H₁) (0.50);B2: P(H₂) (0.50)B3: P(E|H₁) (0.85);B4: P(E|H₂) (0.10)B5(Prior Odds):=B1/B2;B6(Likelihood Ratio):=B3/B4B7(Posterior Odds):=B5*B6;B8(Posterior P(H₁|E)):=B7/(1+B7)
R Snippet
bayes_process_tracing <- function(prior_h1, p_e_h1, p_e_h2) {
prior_odds <- prior_h1 / (1 - prior_h1)
lr <- p_e_h1 / p_e_h2
post_odds <- prior_odds * lr
post_p <- post_odds / (1 + post_odds)
cat(sprintf("Prior Odds: %.2f | LR: %.2f | Post Odds: %.2f | Post P: %.4fn",
prior_odds, lr, post_odds, post_p))
}
bayes_process_tracing(prior_h1 = 0.50, p_e_h1 = 0.85, p_e_h2 = 0.10)
Python Snippet
def bayes_process_tracing(prior_h1: float, p_e_h1: float, p_e_h2: float) -> dict:
prior_odds = prior_h1 / (1.0 - prior_h1)
lr = p_e_h1 / p_e_h2
post_odds = prior_odds * lr
post_p = post_odds / (1.0 + post_odds)
return {
"prior_odds": round(prior_odds, 4),
"likelihood_ratio": round(lr, 4),
"posterior_odds": round(post_odds, 4),
"posterior_probability_h1": round(post_p, 4),
}
print(bayes_process_tracing(prior_h1=0.50, p_e_h1=0.85, p_e_h2=0.10))
Frequently Asked Questions
What is the difference between process tracing and regression analysis?
Regression evaluates cross-case covariation across sample datasets to estimate average effects. In contrast, process tracing investigates within-case mechanisms linking cause to outcome using diagnostic qualitative evidence.
Can process tracing be conducted on a single case study?
Yes. Process tracing was developed specifically for within-case inference in single case studies, testing discrete causal steps using hoop and smoking-gun tests to confirm or eliminate rival hypotheses without cross-case comparisons.
What is a causal process observation (CPO)?
A causal process observation (CPO) is diagnostic data providing insight into context, process, or mechanism. Unlike spreadsheet dataset rows, CPOs include archival records, interviews, or minutes verifying intermediate steps.
How do hoop tests differ from smoking gun tests in process tracing?
A hoop test evaluates a necessary condition: passing affirms viability, while failing eliminates the hypothesis. A smoking-gun test evaluates a sufficient condition: passing confirms the hypothesis, while failing does not eliminate it.
How does Bayesian process tracing handle multiple pieces of evidence?
Bayesian updating is applied sequentially. The posterior odds calculated from the first causal process observation serve as the prior odds for evaluating the second, assuming conditional independence.
Sources
- Beach, D., & Pedersen, R. B. (2019). Process-Tracing Methods: Foundations and Guidelines (2nd ed.). University of Michigan Press. https://www.press.umich.edu/10015525/process_tracing_methods_second_edition (Accessed 2026-11-09).
- Collier, D. (2011). Understanding Process Tracing. PS: Political Science & Politics, 44(4), 823–830. https://polisci.berkeley.edu/sites/default/files/people/u3827/Understanding%20Process%20Tracing.pdf (Accessed 2026-11-09).
- Fairfield, T., & Charman, A. (2017). Explicit Bayesian Analysis for Process Tracing: Guidelines, Opportunities, and Caveats. Political Analysis, 25(3), 363–380. https://doi.org/10.1017/pan.2017.14 (Accessed 2026-11-09).
- George, A. L., & Bennett, A. (2005). Case Studies and Theory Development in the Social Sciences. MIT Press. https://mitpress.mit.edu/9780262572224/case-studies-and-theory-development-in-the-social-sciences/ (Accessed 2026-11-09).







