GlossaryStatistics and interpretation
Causation
Causation is a relationship in which a change in one factor produces a change in another.
Causation is a relationship in which a change in one factor produces a change in another. A causal claim goes beyond describing a pattern or sequence. It requires evidence supporting the proposed effect and addressing other plausible explanations, with conclusions limited to what the study can establish.
Reading claims about impact
A report may say that a program “reduced” incidents when its data show only that incidents fell after the program began. Time order is necessary context, but it does not isolate the program's effect. Seasonal changes, changes in reporting, population shifts, or other programs may also matter.
The CDC's interpretation guidance distinguishes causal assessment from merely finding an association. Its framework calls for considering chance, selection and information bias, confounding, and error. Confounding means another factor helps explain the apparent relationship.
Example
Hypothetical example: A neighborhood begins a prevention program in spring. Reported incidents decline during the next three months. The same months historically have lower counts, and a reporting-system change also occurred.
The decline is a descriptive finding. Attributing all of it to the program requires evidence that addresses the seasonal pattern and reporting change. A suitable comparison or stronger study design could help, but the before-and-after totals alone do not provide that answer.
Common confusion
Correlation describes association. Causation attributes an effect. Statistical significance can describe evidence under a test without proving that the relationship is causal.
Causation in an evaluation is also different from a legal finding of responsibility. A public program study does not decide an individual case.
Scope of a conclusion
Look for the study design, comparison group, outcome definition, time period, and remaining limitations. Prefer wording that matches the evidence: “declined after” describes chronology; “caused a decline” requires additional support. A well-supported effect in one setting may still require care before generalizing to another.