Notes on Correlation, Illusory Correlation, Experimental Design, and Validity
Correlation Basics
When researchers look at relationships between variables, they typically use scatter plots to display the data. Other formats exist, but scatter plots are the most common (90%+ of cases).
Correlation outcomes can be:
Positive correlation: the scatter plot shows an upward trend (as X increases, Y increases).
Negative correlation: the scatter plot shows a downward trend (as X increases, Y decreases).
No correlation: the data points are scattered with no clear pattern.
Three main possibilities for the relationship in a dataset:
Positive correlation
Negative correlation
No correlation
The correlation coefficient ranges from to , with strength increasing as the value approaches either extreme.
The closer the score is to or , the stronger the correlation.
Example of strong negative correlation:
Example of a weak correlation: or
Important caveat: correlation does not imply causation. Just because two variables move together does not mean one causes the other.
Examples discussed:
Smoking and mental illness: there can be a correlation, but it does not establish causation. It is unclear whether smoking causes mental illness, mental illness causes smoking, or whether some third factor influences both.
Self-esteem and depression: a negative correlation might be assumed, but low self-esteem does not guarantee depression, and depression does not guarantee low self-esteem; individuals vary.
Accelerated sexual behavior and depressive symptoms: correlation does not establish a causal direction.
A cautionary note: correlation does not imply a fixed causal chain; multiple factors can influence observed patterns.
A practical example of potential confounding factors:
If you see a link between phone text messaging failures and low phone memory, don’t assume memory is the sole cause. Other factors (network issues, down towers, room location, interference) could explain the pattern.
Illusionary Correlation and Regression toward the Mean
Illusionary correlation (illusory correlation): the belief that two events are related when they are not, often driven by chance patterns or selective attention.
Common in gambling, sports outcomes, or performances where people feel a personal sense of control over random events (e.g., wearing a certain hat or shirt before a win).
Regression toward the mean: extreme outcomes are likely to be followed by more typical ones. Normal patterns are more common than extreme events.
If a particular event is extreme, future measurements tend to revert toward the average, which can be mistaken for a causal effect if not accounted for.
Consequences: illusionary correlations can foster superstitious behavior and reward-predictive rituals (e.g., a preferred outfit yielding perceived wins).
In parenting or education, premature reinforcement of a behavior based on a perceived link can reinforce the mistaken belief that a noncausal factor caused the change. Real change often requires systematic methods rather than superstition.
Experimental Terms: Groups, Bias, and Controls
Experimental group: the group exposed to the treatment or manipulation being studied.
Control group: a comparison group that does not receive the treatment; used to validate the efficacy of the treatment.
Double-blind study: a design that minimizes bias by ensuring that neither the researchers nor the participants know who is in the experimental vs. control group.
Purpose: reduces researcher expectations (observer bias) and participant expectations (placebo or demand characteristics).
Placebo effects: improvements attributed to the belief that a treatment is beneficial, rather than to the treatment itself.
Example: medication study where the experimental group receives the active drug and the control group receives a sugar pill; sometimes the control group shows similar improvements due to expectations.
Emphasizes the power of the mind–body connection and the importance of properly blinding and using placebos when feasible.
Variables in Experimental Research
Independent variable (IV): the factor being studied; deliberately manipulated or varied by the researcher.
The IV is the cause or condition being tested.
Dependent variable (DV): the outcome that is measured; expected to change as a result of manipulations to the IV.
A strong DV has precise and specific operational definitions that align with the research question.
Confounding variables: alternative factors that can influence the DV and obscure the true relationship between IV and DV.
Examples: gender differences (e.g., hormonal effects), age, weight, socioeconomic background, environment, or prior experiences.
In educational studies, factors like home environment and prior knowledge can confound results when comparing instructional methods.
Operational definitions: precise, observable definitions of the variables to ensure consistency and replicability across researchers and studies.
Validity and Reliability in Research
Validity: does the study test what it is intended to test?
Key questions:
Does the study actually address its stated purpose (test what it's supposed to test)?
Does the study account for bias (minimize bias in design, data collection, and analysis)?
Are ethical concerns considered and addressed?
Reliability: consistency and stability of the data and measurements.
Related to the quality of the data collected and the methods used to obtain it; higher reliability contributes to stronger inferences.
The notes imply that reliability will be discussed further in relation to quantitative data and measurement accuracy.
Practical Takeaways and Connections
Always examine whether a reported relationship could be due to confounding variables or shall we attribute causation without proper experimental controls.
When evaluating studies, check for the use of experimental and control groups, blinding, and the presence of a placebo condition when appropriate.
Understand that correlation strength is not a measure of causation; even a near-perfect correlation can be misleading if confounding variables or reversed causality exist.
In practice, use precise operational definitions to ensure that the independent and dependent variables are clearly measurable and replicable.
Consider ethical implications and bias when designing studies or interpreting results.
$