A confounding variable is a third variable, often unmeasured, that influences, or “confounds,” the relationship between an independent and a dependent variable by suggesting a spurious correlation.
Key Takeaways
- Confound: A confounding variable is a third factor that affects both the independent variable (IV) and dependent variable (DV), creating a rival explanation for the results.
- Vs Extraneous: Every confound is an extraneous variable (any factor besides the IV that could affect the DV), but only those that vary with the IV count as confounds.
- Correlation vs Causation: A correlation between two variables never proves that one causes the other, because a hidden confound could explain the link.
- Common Confounds: Participant expectations, the placebo effect, demand characteristics, and investigator effects are frequent sources of confounding.
- Control: Random allocation, control variables, within-subjects designs, and case-control matching all help rule out confounds.
Confounding Variables in Research
How Confounding Variables Distort Results
A confound works by changing alongside the independent variable (IV). An extraneous variable is any factor other than the IV that could affect the dependent variable (DV), the outcome a study measures. It becomes a confound once it varies systematically with the IV.
That is why a correlation between two variables does not prove that one causes the other.
Say the IV and DV look associated. If a third variable, C, actually affects both the IV and the DV, that association may be spurious rather than causal.
C is the confounding variable in this example.
To estimate the real effect of the IV on the DV, researchers must reduce the influence of this confound. A causal relationship that looks solid on the surface might not exist at all once it is taken into account.
Even when a real cause-and-effect relationship does exist, a confound can still make it look bigger or smaller than it really is. Researchers call this confounding bias. Positive bias overestimates the effect of the IV on the DV, and negative bias underestimates it.
The table compares the two.
| Extraneous variable | Confounding variable | |
|---|---|---|
| What it is | Any variable other than the IV that could affect the DV | An extraneous variable that has affected the DV and varies with the IV |
| Effect on results | Adds “noise” that may obscure a real effect | Offers a rival explanation and can produce a false effect |
| Threat | Reduces sensitivity | Directly threatens internal validity and the causal claim |
| Remedy | Standardise or hold constant | Control by design: randomisation, blinding, placebo, counterbalancing |
Confounds in Correlational and Quasi-Experimental Studies
Confounds are hardest to exclude when the researcher does not manipulate the IV. In a correlational study, two measured variables may both be driven by a third. This is the third-variable problem.
Quasi-experiments carry the same risk. Their IV is a subject variable, a trait participants already have, such as age or gender identity. Researchers cannot randomly allocate people to these traits, so a rival explanation can never be fully excluded.
Statistical control can help, but it needs care. Rohrer (2018) used graphical causal models to show how researchers should decide which third variables to adjust for in observational data.
Two types are risky, because controlling for either can move the estimate away from the true causal effect:
- Mediators: Variables on the causal path between the IV and DV.
- Colliders: Variables caused by both the IV and the DV.
Demand Characteristics and Investigator Effects
Two sources of confounding are common enough in psychology to name individually. Demand characteristics are cues in the experimental situation that reveal the hypothesis to participants. Orne (1962) argued that participants rarely stay passive.
Some try to confirm what they think the researcher wants, others deliberately do the opposite, and many act on evaluation apprehension, behaving as they think they should. A single-blind procedure controls for this by keeping participants unaware of their condition.
Investigator effects are the researcher’s version of the same problem. An experimenter’s own expectations can bias how they treat participants, record behaviour, or interpret ambiguous responses.
Rosenthal and Fode (1963) told students they were handling rats bred to be “maze-bright” or “maze-dull,” but the rats had actually been allocated at random. The bright rats performed better anyway.
Rosenthal and Jacobson (1968) found the same pattern in classrooms, an effect now known as the Pygmalion effect. Teachers were told certain randomly chosen pupils were about to bloom academically, and those pupils’ measured IQ rose more than their classmates’.
A double-blind procedure controls for this by keeping the experimenter unaware too.
Participant and Situational Confounds
Extraneous variables fall into two families, and each calls for a different control.
- Participant variables: Individual differences such as age, IQ, mood or prior experience. They matter most in independent-groups designs, where chance can place different people in each condition.
- Situational variables: Features of the environment or procedure, such as noise, time of day, instructions, or order effects (practice and fatigue) in repeated-measures designs.
Demand characteristics and investigator effects are situational confounds. Any of these factors becomes a confound only when it differs systematically between conditions. Testing one group in the morning and the other in the afternoon, for example, turns time of day into a confound.
Reducing Confounding Variables
Identify every possible confounding variable and consider its impact on your research design. Doing so protects internal validity, the confidence that the IV, not a confound, produced any change in the DV.
Here are some techniques to reduce the effects of these confounding variables:
- Random allocation: Randomly assigning participants to treatment and control groups ensures confounders affect both equally, so they cannot bias your independent variable.
- Control variables: Restrict your subject pool to people with similar characteristics, such as the same age range, sex, or education level, so these factors cannot confound your results.
- Within-subjects design: Every participant takes part in every condition, so participant differences cannot vary systematically between conditions.
- Case-control studies: These match confounders equally across the experimental and control groups, so the two groups stay comparable.
- Standardisation: Keep instructions, materials and procedure identical for every participant, so situational variables cannot differ systematically between conditions.
- Counterbalancing: Vary condition order to cancel practice and fatigue.
- Blinding and placebos: Single-blind procedures keep participants unaware of their condition, and double-blind procedures keep the experimenter unaware too.
- Statistical control: Measure the suspected confound as a covariate and adjust for it in the analysis, so its influence is separated out.
Match the control to the threat. Standardisation and counterbalancing target situational variables, random allocation and matching target participant variables, and blinding and placebos target expectation-based confounds.
Examples of Confounding Variables in Psychology
Suppose we wanted to measure the effects of caloric intake (IV) on weight (DV). We would have to try to ensure that confounding variables did not affect the results. These variables could include the following:
- Metabolic rate: A faster metabolism burns calories more quickly.
- Age: Younger people tend to burn calories faster than older adults, which affects weight gain.
- Physical activity: People who exercise more burn more calories and may weigh less, even if they eat more.
- Height: Taller individuals need to consume more calories to gain weight.
- Sex: Men and women have different caloric needs to maintain the same weight.
Psychology studies face confounds of their own:
- Exercise and stress: People who exercise more report lower stress, but better general health or more free time may drive both.
- Caffeine and stress: Caffeine consumed before the experiment, or participants’ expectations about the drink, can confound the result.
- Study time and test scores: A student’s inherent aptitude or previous knowledge can affect the score regardless of how long they study.
References
Orne, M. T. (1962). On the social psychology of the psychological experiment: With particular reference to demand characteristics and their implications. American Psychologist, 17(11), 776–783.
Rohrer, J. M. (2018). Thinking clearly about correlations and causation: Graphical causal models for observational data. Advances in Methods and Practices in Psychological Science, 1(1), 27–42. https://doi.org/10.1177/2515245917745629
Rosenthal, R., & Fode, K. L. (1963). The effect of experimenter bias on the performance of the albino rat. Behavioral Science, 8(3), 183–189.
Rosenthal, R., & Jacobson, L. (1968). Pygmalion in the classroom: Teacher expectation and pupils’ intellectual development. Holt, Rinehart & Winston.
