In AP Statistics, a confounding variable is an outside factor related to both the explanatory variable and the response variable, so you can't tell whether the treatment or the confounder caused the observed effect. Random assignment in experiments tends to balance out confounding variables.
A confounding variable is a third variable that's tangled up with both your explanatory variable and your response variable. Because it moves with both, you can't separate its effect from the treatment's effect. Picture a study that finds people who walk more have lower cholesterol. Maybe walking lowers cholesterol. Or maybe people who walk more also eat healthier, and diet is what's really driving the cholesterol numbers. Diet is the confounder, and the study can't tell the two explanations apart.
This is exactly why the CED says a well-designed experiment needs random assignment of treatments to experimental units (AP Stats 3.5.B). Random assignment tends to balance the effects of uncontrolled variables across treatment groups, so the healthy eaters end up spread roughly evenly between groups. Once the groups are balanced, any difference in responses can be attributed to the treatments themselves (AP Stats 3.5.C). Confounding is also the core reason observational studies can't establish causation. Nobody assigned the treatments, so confounders run wild.
Confounding lives in Topic 3.5 (Introduction to Experimental Design) in Unit 3, supporting learning objectives AP Stats 3.5.A, 3.5.B, and 3.5.C. It's the villain that the four principles of good design (comparison, random assignment, replication, control) exist to defeat. But it doesn't stay in Unit 3. When you reach inference in Unit 6, confounding determines what your conclusion is allowed to say. A confidence interval for a difference of proportions (AP Stats 6.9.B) can give you evidence that two groups differ, but if the data came from an observational study, confounding variables mean you can claim association, not causation. Knowing when you can and can't say "causes" is one of the most-tested judgment calls in the whole course.
Keep studying AP Statistics Unit 3
Randomization (Unit 3)
Random assignment is the cure for confounding. It doesn't eliminate confounders, but it spreads them roughly evenly across treatment groups so they can't systematically favor one group. That balance is what lets you attribute differences in the response to the treatments.
Independent Variable (Unit 3)
The independent (explanatory) variable is the one you deliberately manipulate. A confounder is an uninvited variable that travels along with it. If the confounder changes whenever your explanatory variable changes, their effects on the response get mixed together and you can't pull them apart.
Double Blind Experiment (Unit 3)
Blinding controls a specific human-flavored confounder, expectations. If subjects or researchers know who got the real treatment, their beliefs and behavior can influence the response. Double-blinding keeps that knowledge from contaminating the results.
Confidence Interval (Unit 6)
An interval for a difference of proportions that excludes zero tells you the groups likely differ. Whether you can say the treatment caused that difference depends on Unit 3 logic. Random assignment means causation is on the table; an observational study means confounding could explain the gap, so you stop at association.
Confounding shows up in two flavors. First, design questions ask you to fix it. Multiple-choice stems describe an experiment and a worrying outside factor (like nutrition affecting an athletic training study) and ask which design technique addresses it. Your toolbox is random assignment, blocking, comparison groups, and blinding. The 2023 FRQ on adding fibers to concrete and the 2021 FRQ on walking and cholesterol both required designing or evaluating studies where controlling outside variables was the point. Second, conclusion questions test your restraint. The 2022 FRQ comparing allergy treatment success rates at two clinics used random samples of patient records, an observational setup, so a causal claim would be wrong no matter what the interval showed. When you write a design FRQ answer, don't just name random assignment. Explain that it balances the effects of potential confounding variables across groups, because that explanation is what earns the point.
Both are unmeasured outside variables, but a confounding variable is specifically associated with both the explanatory variable and the response variable, which is what makes its effect impossible to separate from the treatment's. A lurking variable is the broader idea of any variable not included in the study that affects the relationship. On the exam, if you're explaining why an observed difference might not be caused by the treatment, "confounding" is the precise word, and you should describe how the variable connects to both the treatment groups and the response.
A confounding variable is associated with both the explanatory variable and the response variable, so its effect can't be separated from the treatment's effect.
Random assignment of treatments tends to balance confounding variables across groups, which is why well-designed experiments can support causal conclusions.
Observational studies can't establish causation because confounding variables are uncontrolled; the strongest claim you can make is association.
When a confounding variable is identifiable in advance (like nutrition in an athletic training study), blocking on it or holding it constant directly controls its effect.
On FRQs, naming random assignment isn't enough; you earn credit by explaining that it balances the effects of potential confounding variables so differences in responses can be attributed to the treatments.
A significant confidence interval for a difference of proportions tells you the groups differ, but confounding determines whether you're allowed to say why they differ.
It's an outside variable related to both the explanatory variable and the response variable, so you can't tell whether the treatment or the confounder produced the observed effect. The classic fix is random assignment, which balances confounders across treatment groups.
No, it doesn't eliminate them. Random assignment tends to balance their effects across treatment groups so they don't systematically favor one group. That balance is what lets you attribute differences in responses to the treatments, which is exactly how the CED phrases it in AP Stats 3.5.C.
A confounding variable is specifically tied to both the explanatory and response variables, which is what mixes its effect with the treatment's. A lurking variable is any unmeasured variable affecting the relationship. Confounding is the precise term to use when arguing a study can't support a causal claim.
Yes. Random sampling controls who gets into the study, not who gets which treatment. The 2022 FRQ comparing allergy success rates at two clinics used random samples of patient records, but because nobody randomly assigned patients to clinics, confounding variables like patient severity could still explain any difference.
If you can identify it before the experiment, block on it (group similar units together and randomize within blocks) or hold it constant for all units. If you can't identify it, rely on random assignment to balance its effects across treatment groups.