After does not mean because
An outcome improves after a program begins, and the timeline is presented as the causal explanation. The missing question is what would have happened over the same period without the program. Before accepting the conclusion, identify what was observed, which assumptions connect it to the claim, and what practical judgment depends on that connection.
The relevant mechanism is this: time trends, concurrent changes, regression toward typical values, and selection can create before-and-after differences without a treatment effect. A disciplined analysis makes the chain explicit. Describe the sequence; identify other changes; estimate the untreated trajectory; compare similar units; test whether timing and mechanism fit. The result is that a temporal association becomes one clue within a causal argument rather than the entire argument. The chain supplies specific points at which a source, rival account, or later observation can challenge the argument.
The central claim is this: Causal inference asks about a comparison with an unobserved alternative history. The counterfactual cannot be observed for the same unit at the same time, so every design constructs it through assumptions. The aim is proportionate confidence: strong where the evidential link survives scrutiny and explicitly limited where it does not.
Read more closely
Apply the claim directly: write what you believe would have happened to your outcome if the proposed cause had been absent. Use Support to identify the observation doing the most work, Uncertainty to name the strongest live rival, and Revision to specify a result that would move confidence. Write the answer as a testable statement, not a declaration of intellectual virtue. Prioritize the uncertainty that matters most for the decision, then explain why your revision test is more informative than an easier confirming example. Name the expected direction beforehand.
Textual observation. The RCT guide describes the fundamental problem of causal inference as our inability to observe both potential outcomes for one unit. Read the opening source for the author’s actual distinction and note where the lesson goes beyond it. Paraphrase the reasoning, quote only the short phrase needed to anchor it, and mark any premise the text leaves undefended.
Empirical evidence. Interrupted series, comparison groups, and qualitative process evidence can test whether a change exceeds the background path and follows the proposed mechanism. This record bears on a defined process but does not settle the lesson’s question by itself. Check population, period, outcome, and comparison before treating the record as support for the broader claim.
Danger and counterargument. A dramatic, tightly timed effect can sometimes support causation without a formal control, especially when alternatives predict no such break. The counterfactual cannot be observed for the same unit at the same time, so every design constructs it through assumptions. Ask what observation would strengthen the objection and what would make it less plausible.
Course interpretation. Treat design strength as a matter of how well realistic alternatives are excluded, not as a ceremonial label. A fall in road deaths after a law may also reflect mobility changes, vehicle safety, enforcement intensity, weather, or reporting. This synthesis is the course author’s application, not a finding copied from one cited work. Preserve the distinction among source statement, empirical record, and interpretation. Use this transfer test. Replace ‘what happened next?’ with ‘compared with what would otherwise have happened?’
The practical stakes become clearer when two similar claims are compared directly. In the first, a temporal association becomes one clue within a causal argument rather than the entire argument. In the second, the analysis distinguishes adjustment that addresses bias from adjustment that creates it. Their wording may match, yet they rely on different observations, assumptions, and routes to correction. Compare the chains before the conclusions, and note whether one contains an independent test the other lacks. Keep that difference visible in both the prose and the workbook entry.
A useful challenge is to reverse the favored explanation. Suppose this objection holds: A dramatic, tightly timed effect can sometimes support causation without a formal control, especially when alternatives predict no such break. Would the next claim—a confounder is a causal role, not simply any variable correlated with the outcome—still hold for the same population and decision? The counterfactual does not prove the rival; it reveals which evidence must discriminate between them. Record the answer as observed, inferred, uncertain, and decision-relevant for this lesson.