When you try to figure out why someone acted a certain way, you're performing a mental calculation. You look at what they did, think about how others act in the same situation, and try to decide if the cause stems from their internal personality or external circumstances. Social psychologist Harold Kelley formalized this everyday reasoning process into the covariation model, a foundational framework for understanding how people make causal attributions.
While web pages often mix up social attribution theory with modern mathematical statistics or confuse figures like Gena O'Kelley, the real roots of causal inference in behavioral science trace back to Kelley's work on how human minds assign causes to observed events. Grasping this model helps researchers, data analysts, and psychologists separate correlation from true causation when evaluating human behavior.
What is Kelley's Covariation Model?
Introduced in the late 1960s and early 1970s, Kelley's covariation model posits that people attribute behavior to specific causes by examining factors that are present when the behavior occurs and absent when it doesn't. Think of it as a mental ANOVA test run without software.
To determine the root cause of an action, observers rely on three distinct informational criteria: To understand the bigger picture, we recommend the excellent analysis by The New York Times.
- Consensus: How other people react to the same stimulus. If everyone yells at a malfunctioning vending machine, consensus is high. If only one person yells, consensus is low.
- Distinctiveness: How the actor responds to other, different stimuli. If a worker complains only about this specific software tool, distinctiveness is high. If they complain about every tool they touch, distinctiveness is low.
- Consistency: How the actor behaves toward the same stimulus across different times and contexts. If a student always falls asleep in statistics class, consistency is high. If they fell asleep just once, consistency is low.
The Three Core Attribution Categories
By combining high and low states of consensus, distinctiveness, and consistency, observers arrive at one of three primary causal explanations.
1. Internal (Dispositional) Attributions
When consensus is low, distinctiveness is low, and consistency is high, you point the finger inward. For example, if Sarah is the only person who skips team meetings (low consensus), skips all other company events (low distinctiveness), and has done this every single week for a year (high consistency), you conclude that Sarah's behavior is driven by her own internal traits or habits.
2. External (Stimulus) Attributions
When consensus is high, distinctiveness is high, and consistency is high, the cause lies in the external stimulus itself. If every student fails a specific midterm exam (high consensus), those same students pass all their other exams easily (high distinctiveness), and every cohort that takes this professor's test fails it year after year (high consistency), the test itself is the cause.
3. Circumstantial Attributions
When consensus is high, distinctiveness is low, and consistency is low, the environment or context dictates the outcome. If everyone bursts into laughter during a solemn lecture because a loud siren outside interrupts the room, consensus is high, but because people don't normally laugh during lectures (low consistency) and laugh at many different unexpected noises (low distinctiveness), the specific circumstance explains the reaction.
How Covariation Links to Modern Statistical Causal Inference
Although Kelley built his model for social psychology, the underlying logic mirrors modern statistical methods used to isolate variables in data science, such as propensity score matching and directed acyclic graphs (DAGs).
In modern quantitative research, establishing causality requires controlling for confounding variables—a mathematical equivalent of checking consistency and distinctiveness across datasets. Just as a psychologist looks for covariation across time and people, a data scientist uses regression discontinuity or instrumental variables to see how an outcome changes when a treatment variable varies while holding background conditions steady.
You can explore more about formal statistical techniques through resources provided by the American Statistical Association or read foundational discussions on research design via the National Institutes of Health.
Common Mistakes When Making Causal Inferences
People and analysts alike frequently stumble when trying to assign causes to observed patterns. Recognizing these errors prevents flawed conclusions.
- The Fundamental Attribution Error: Overestimating internal dispositional causes while ignoring external situational factors. You assume a driver who cuts you off is inherently reckless without considering they might be rushing to a hospital.
- Ignoring Base Rates: Failing to look at how often an event occurs naturally across a broader population, leading to false associations between rare variables.
- Confusing Correlation with Covariation: Spotting two metrics moving together in a dataset and immediately assuming one causes the other without testing for confounding third variables.
Frequently Asked Questions
What is the main difference between internal and external attributions?
Internal attributions assign the cause of an action to a person's inner traits, skills, or motivations. External attributions assign the cause to outside forces, objects, or environmental pressures.
Do people actually calculate consensus, distinctiveness, and consistency consciously?
Not usually. Psychologists find that people run these evaluations heuristically and automatically, especially when facing unexpected or negative events that demand an explanation.
How does Kelley's model relate to statistics?
Both fields rely on covariation. While statistics uses numerical variance and covariance formulas to test hypotheses, Kelley's model describes the intuitive cognitive framework humans use to process data in everyday social settings.
Can observational data prove causation on its own?
No. Observational data shows correlations and covariations, but proving true causation requires experimental controls, randomized trials, or rigorous causal inference models to rule out alternative explanations.
