Look for a Third Variable Behind the Association
In “Why Correlation Does Not Imply Causation,” you learned that even a strong association does not establish that changing one variable causes a change in another. A useful next question is whether another variable could help explain the pattern. In an observational study, a variable that was not included in the analysis may be connected to both measured variables and contribute to their association.
For example, suppose a city finds a strong positive association between the number of streetlights in a neighborhood and the number of police-reported nighttime incidents. It would be premature to conclude that streetlights cause more incidents. Neighborhood population or nighttime activity might be related to both: busier, more densely populated areas may have more streetlights and more opportunities for incidents to be reported. That is a possible explanation to investigate, not a conclusion proved by the correlation.
A useful way to evaluate a proposed lurking variable is to check two connections. First, could it be related to the explanatory variable? Second, could it also be related to the response variable? If either connection is missing or implausible, the variable has not yet offered a convincing explanation for the association.
This two-link check does not prove that the third variable is responsible. It helps you decide whether it is a plausible explanation that the study should consider. Other explanations may also be possible, and the data may not provide enough information to distinguish them.
How a Lurking Variable Can Contribute to Confounding
Suppose \(x\) is the explanatory variable and \(y\) is the response variable. If a third variable \(z\) tends to vary with \(x\) and is also connected to \(y\), the observed association between \(x\) and \(y\) may reflect some combination of these relationships. The study may not be able to tell how much of the pattern is associated with \(x\), how much with \(z\), or whether other factors also matter.
This is why it is important to name the measured variables precisely. If a study records recreational screen time, for instance, “schoolwork” is not automatically an adequate explanation for an association involving that measurement. You would need a plausible reason that schoolwork is connected to the specific screen-time variable being measured, not just to screen use in general.
A lurking variable might make an observed association look stronger, weaker, or different from the relationship between the measured variables alone. The direction and size of any such influence depend on the setting. Do not assume that identifying one possible lurking variable tells you exactly what the association would be without it.
In an observational study, researchers record what happens rather than assigning the explanatory variable. As in the earlier tutorial on causation, this limits conclusions about cause and effect. A plausible lurking variable is a reason to be cautious about a causal interpretation, not proof that the measured explanatory variable has no effect.
Worked Example: Streetlights and Nighttime Incidents
Imagine a fictional city comparing neighborhoods. Its observational records show a strong positive association between the number of streetlights and the number of police-reported nighttime incidents. The city did not assign streetlights to neighborhoods for a controlled study.
State the variables: The explanatory variable is the number of streetlights in a neighborhood. The response variable is the number of police-reported nighttime incidents there.
Propose a lurking variable: Neighborhood population or nighttime activity could be a lurking variable. The two-link check asks whether it could plausibly be connected to both measured variables.
Check the links: More densely populated or busier neighborhoods may have more people and activity after dark, which could be associated with more reported incidents. Those neighborhoods may also have more streetlights because more streets and public spaces need lighting. Thus, the proposed variable has a plausible connection to the streetlight count and to the incident count.
Conclude in context: Population or nighttime activity could contribute to the positive association between streetlights and reported incidents. The records do not establish that streetlights cause incidents, nor do they prove that population or activity explains the pattern. The city would need additional information and a suitable study design to investigate those explanations.
Use the Two-Link Check, Not Just a Plausible-Sounding Story
A possible explanation can sound reasonable but still fail as a confounding explanation if it does not connect to both measured variables. A careful answer identifies the proposed variable, states how it could be related to the explanatory variable, and states how it could be related to the response variable. The link must fit the actual variables and context in the question.
For example, if researchers find an association between hours spent studying and exam scores, saying “students have different schedules” is incomplete unless you explain how schedules might be related to both study hours and scores. The label alone does not show how the variable could contribute to the observed pattern.
Also be cautious about the direction of a proposed relationship. A lurking variable need not make the association stronger in every setting. The strongest answer describes plausible links without claiming more than the scenario supports.
Worked Example: Study Time and Exam Scores
A fictional school surveys students about hours spent studying for a science exam and records their scores. In the observed students, more study time tends to go with higher scores. The school did not assign students how long to study, so this is an observational study.
State the variables: The explanatory variable is hours spent studying for the exam, and the response variable is the exam score.
Propose a lurking variable: Prior preparation in science could be a lurking variable. Students with stronger preparation may be more comfortable studying for longer, and they may also be better prepared to earn higher scores.
Check the links: The first link is from prior preparation to study time: students who already know more may have the skills or motivation to spend more time studying. The second link is from prior preparation to exam score: stronger preparation may help students answer more questions correctly. Both are plausible in this fictional setting.
Conclude in context: Prior preparation could help explain some of the positive association between study hours and exam scores, so the observed pattern does not by itself show that more studying caused higher scores. It also does not show that studying had no effect. The observational data alone cannot separate the possible contributions of study time and prior preparation.
Be Precise About the Measured Variables
A lurking-variable explanation depends on what was actually measured. Consider a fictional survey that records the amount of sunscreen people apply during a beach outing and whether they experience sunburn. Suppose the observed association is positive: people who apply more sunscreen also tend to report more sunburn.
It would be incorrect to jump from this association to the claim that sunscreen causes sunburn. A plausible lurking variable is time spent outdoors in strong sunlight. People who spend longer outdoors may be more likely to apply sunscreen, perhaps more than once, and they also have more time for sun exposure that could result in sunburn. This variable is connected to both the measured sunscreen amount and the response, sunburn.
Notice the care needed in this explanation. It is not enough to say “sun exposure” if the link to sunscreen use is left unstated. The explanation is plausible here because longer outdoor exposure could lead people to use more sunscreen as well as increase their opportunity for sunburn. Even then, the observational association does not prove that this factor accounts for the whole pattern or determine the effect of sunscreen.
Worked Example: Sunscreen Use and Sunburn
A fictional beach-safety survey asks visitors how much sunscreen they applied during the day and whether they experienced sunburn. The survey finds a strong positive association between reported sunscreen amount and sunburn. Researchers observed visitors’ choices; they did not assign sunscreen use.
State the variables: The explanatory variable is the amount of sunscreen applied, and the response variable is whether a visitor experienced sunburn.
Propose a lurking variable: Time spent outdoors in strong sunlight could be a lurking variable.
Check the links: Visitors who spend longer outdoors may apply more sunscreen, including reapplying during the day. They also have longer exposure to sunlight, which could be associated with a greater chance of sunburn. Thus, outdoor time has a plausible connection to both the measured sunscreen amount and the response.
Conclude in context: Longer outdoor exposure could contribute to the positive association in the survey. The association does not show that sunscreen causes sunburn, and the proposed lurking variable is not proven to be the explanation. It is appropriate to report the observed association and identify outdoor time as one plausible factor that the survey should consider.
What the Observational Data Can Establish
Finding a plausible lurking variable does not erase the observed association. The data still show a pattern among the cases studied. What changes is how confidently we can interpret that pattern. If a third variable could be related to both measured variables, an observed association may not represent the relationship we would see if the third variable were held comparable across cases.
Sometimes researchers collect information about a suspected lurking variable and compare observations with similar values of that variable. This can help investigate whether the original association is present in those comparisons. But unless the study design and analysis justify a causal conclusion, measuring or accounting for an additional variable alone does not automatically prove causation.
When reading an observational study, ask what the researchers measured and what they left unmeasured. Then check whether a proposed lurking variable could be related to each of the two measured variables. Be careful to distinguish a plausible explanation from a demonstrated one, and to distinguish an observed association from a causal effect.
Common Mistakes and AP Exam Tips
- Naming a variable without explaining both links. Full-credit reasoning says how the proposed lurking variable could be related to the explanatory variable and how it could be related to the response variable, in context.
- Choosing a variable that is only related to one measured variable. That may be relevant to the study, but it does not complete a confounding explanation. Use the two-link check.
- Using a vague label instead of the actual measured quantity. If the study records recreational screen time, explain how a proposed factor could be related to recreational screen time specifically. Do not substitute total screen use or schoolwork without support from the scenario.
- Claiming that the lurking variable is proven to explain the association. In an observational study, say that it could contribute to or help explain the pattern. The association alone does not establish that it does.
- Concluding that the explanatory variable has no effect. A possible lurking variable makes a causal interpretation less secure; it does not prove that the explanatory variable has no causal effect.
- Calling every unmeasured variable a confounder. A variable is a useful confounding explanation only when its connections to the measured variables make sense in context. Being left out of the study is not enough.
A strong AP response identifies the explanatory and response variables, names one plausible lurking variable, and explains both links. It then states the limitation: because the study is observational, the observed association alone cannot separate the possible influence of the explanatory variable from the lurking variable or establish causation.
Check Your Understanding
For each situation, name a plausible lurking variable and explain both links before stating what the observational data do and do not show.
- A fictional town finds that neighborhoods with more public charging stations also have more reported electric-bicycle thefts. Propose a lurking variable and explain how it could be related to both measured variables.
- A survey finds that people who own more houseplants report better indoor air quality. Identify one possible lurking variable and explain the two links it would need to have.
- A school observes that students who attend more optional review sessions tend to earn higher exam scores. How could prior preparation be a lurking variable?
- In a beach survey, visitors who apply more sunscreen report more sunburn. Why is outdoor exposure time a more complete explanation than simply naming “weather” without further detail?
- Why does naming a plausible lurking variable not prove that it caused the observed association or that the explanatory variable had no effect?