Skip to content
Kind and Quiet
9 min read research deep dive

Research Deep Dive: Kelley's Covariation Theory and the Science of Attribution

The Academic Framework Behind the Attribution Decoder

This research deep dive accompanies the Attribution Decoder tool. If you want to understand the academic foundations behind the framework, or you're curious why we simplified Kelley's original model, this is for you. If you just want to use the tool, you don't need this. This is for you if you want to go deeper.

What Was Simplified in the Newsletter (And Why)

Kelley and other attribution scholars distinguish between two aspects of the environment. This allows for a richer analysis of attributions that helps develop theory, but may be of limited use to apply this in everyday life.

The newsletter version of Kelley's Attribution Decoder uses a simplified framework that collapses "the situation" into a single category. This makes the tool practical and accessible for busy leaders who need quick attribution analysis.

The simplification:

“Stimulus” describes stable properties of a task or situation. For example, in a maths exam you are probably always confronted with the reality of solving math problems. This is a stable aspect of that situation.

“Circumstances” describe temporary aspects of a situation. Maybe in your last math exam, you were tired and distracted because neighbours had a party and you couldn’t sleep all night.

Why we simplified: The distinction between "stimulus" (stable properties of the task/environment) and "circumstances" (temporary, one-off factors) is academically important but adds cognitive load that makes the framework harder to use in real-time.

This deep dive presents Kelley's original framework, shows how it was tested empirically, and explains why people systematically fail at attribution. This kind of research also laid the groundwork for what became known as the Fundamental Attribution Error.

Kelley's Covariation Principle: The Foundation

In 1973, Harold Kelley articulated the core principle of covariation theory:

"An effect is attributed to the one of its possible causes with which, over time, it covaries."[1]

What this means: You attribute a behavior to whatever factor it consistently appears with. If someone always performs poorly in high-stress situations but performs well otherwise, you should attribute the poor performance to stress (situation), not to lack of ability (person). You might then even distinguish more by specifying that it’s stress from time pressure, not stress from high expectations, or a micro-managing boss, etcetera.

The normative ideal: Kelley's model is normative—it describes how people should think, not how they actually think. He never believed people naturally calculate covariation patterns in everyday life. The model provides a systematic framework for avoiding the attribution errors we make intuitively.

Kelley was inspired by statistical models such as ANOVA (Analysis of Variance). These models try to systematically distinguish between different sources of variance. If enough of the variance can be attributed to a specific source, the model allows the analyst to conclude that that source had an effect. It may sound complicated, but it’s actually quite simple and Kelley translated this logic into a model to help us navigate the social worlds we live in.

The Four Attribution Types

Kelley's original framework distinguished among four possible causes for any behavior:

1. Person Attribution
Something about the individual's personality, ability, or character caused the behavior.
Example: "Alex lacks forecasting skill."

2. Stimulus Attribution
Something about the task, object, or entity caused the behavior.
Example: "The project data/system is flawed."

3. Circumstance Attribution
Something about the particular temporary situation caused the behavior.
Example: "It's an off day" or "Unusual market conditions today."

4. Combination Attribution
Multiple factors (person + stimulus + circumstances) jointly caused the behavior.
Example: "Both Alex and the project contribute to the inaccuracy."

Why the newsletter simplifies this: For practical use, we collapse Stimulus + Circumstances into "Situation." This maintains the core insight (is it the person or not?) while reducing cognitive load.

The initial key is to distinguish between the person and the situation. You can then go deeper and get a sense for whether the situational factors are more likely to be stable features of the situation (stimulus), or whether they are some temporary aspect (circumstance).

The Three Information Variables

Kelley identified three types of information that, when analyzed together, reveal the correct attribution.

Let’s say we’re observing an analyst Alex provide an inaccurate financial forecast that had devastating consequences for the company. Kelley’s covariation model would have us consider three factors:

1. Distinctiveness
Does Alex behave differently across different situations/stimuli?

2. Consistency
Does Alex always behave this way in this situation/with this stimulus?

3. Consensus
Do other people also behave this way in this situation/with this stimulus?

McArthur (1972): An Empirical Test

Leslie McArthur conducted the first major empirical test of Kelley's theory in 1972.[2] Her study provided systematic evidence for how people actually use (and misuse) these three information variables. It’s also a great read (if you’re into nerdy original research papers) because she’s intellectually meandering in a stream-of-consciousness way when reporting and interpreting results. Not the most fun you’ll ever have (hopefully), but quite entertaining for an original research paper.

Method

McArthur presented subjects with 16 different behavioral scenarios (emotions, accomplishments, opinions, and actions). For each scenario, subjects received:

Subjects then attributed each behavior to one of four causes:

  1. The person (e.g., "Something about John probably caused him to laugh at the comedian")
  2. The stimulus (e.g., "Something about the comedian probably caused John to laugh")
  3. The circumstances (e.g., "Something about the particular circumstances probably caused John to laugh")
  4. A combination of the above

Key Findings

McArthur's results revealed both how the three variables should work and how people actually use them.

Maybe most importantly, (and this is what she also wonders about) is the stable observation that people seem predisposed to attribute behaviour to the person, more so than to the situation.

For example, on pages 178-179 she writes:

“Only one information combination—high consensus, high distinctiveness, and high consistency—yielded a significantly higher proportion of stimulus attribution than did no information (P = .61 versus .18, p < .001; see Figure 2)."

Person Attribution

Person Attribution was Strongest when:

Example pattern: Alex gives inaccurate forecasts for many projects (low distinctiveness), always gives inaccurate forecasts for this project (high consistency), and hardly anyone else produces inaccurate forecasts on this project (low consensus).

Attribution: Alex lacks forecasting skill (PERSON)

Stimulus Attribution

Stimulus Attribution Was Strongest when:

Example pattern: Alex gives inaccurate forecasts only for this project (high distinctiveness), always gives inaccurate forecasts for this project (high consistency), and almost everyone else also produces inaccurate forecasts on this project (high consensus).

Attribution: The project data/system is flawed (STIMULUS)

Note: Only ONE information combination produced stimulus attribution more frequently than person attribution: high consensus + high distinctiveness + high consistency. This combination yielded 61% stimulus attribution vs. 18% when no information was given.

Circumstance Attribution

Strongest when:

Example pattern: Alex never gave inaccurate forecasts here before (low consistency), but this is the only project where he's had this issue (high distinctiveness).

Attribution: This is an off day for some reason; temporary circumstances (CIRCUMSTANCE)

The Complete Attribution Patterns

Here are the full 8 patterns that emerge from combining the three information variables at high and low levels:

High Distinctiveness Scenarios

"Alex gives inaccurate forecasts only for THIS project"

Consistency High Consensus
(Others also struggle)		 Low Consensus
(Others don't struggle)
High
(Always happens in this situation)		 🟦 STIMULUS
The project is flawed		 🟪 PERSON-STIMULUS
Alex and the project jointly cause it
Low
(First time / rare)		 🟨 CIRCUMSTANCE
Alex usually overcomes project difficulties		 🟨 CIRCUMSTANCE
It's an off day for Alex

Low Distinctiveness Scenarios

"Alex gives inaccurate forecasts for MANY projects"

Consistency High Consensus
(Others also struggle)		 Low Consensus
(Others don't struggle)
High
(Always happens in this situation)		 🟧 AMBIGUOUS
Both Alex and the project contribute		 🟩 PERSON
Alex lacks forecasting skill
Low
(First time / rare)		 🟧 AMBIGUOUS
Both normally manage, but not today		 🟧 AMBIGUOUS
Alex struggles, but others handle it

Key insight: The clearest attributions occur when all three variables align:

Ambiguous patterns occur when variables conflict or when distinctiveness is low.

Legend:
🟦 Stimulus attribution (external cause)
🟪 Person-Stimulus interaction
🟨 Circumstance attribution (temporary)
🟩 Person attribution (internal cause)
🟧 Ambiguous (multiple factors)

The "Black Box" Bias: Foreshadowing the Fundamental Attribution Error

McArthur made a critical observation in her 1972 paper that foreshadowed what would later be formalized as the Fundamental Attribution Error:

"But, why are mundane events such as 'Sue is afraid of the dog,' 'George translates the sentence incorrectly,' 'Ralph trips over Joan's feet while dancing,' and 'Steve puts a bumper sticker advocating improved auto safety on his car' overwhelmingly attributed to characteristics of Sue, George, Ralph, and Steve (a total of 35 person attributions) rather than to the ferocity of the dog, the difficulty of the sentence, the clumsiness of Joan, or the attractiveness of the bumper sticker (a total of three stimulus attributions)?

One is hard pressed to come up with any logical explanation of this proclivity for person attribution. Certainly the real-world incidence of fearful people, dumb people, clumsy Ralphs, and bumper sticker buffs does not exceed the incidence of ferocious dogs, difficult sentences, clumsy Joans, and beautiful bumper stickers!

One can only conclude that there exists a bias in favor of attributing behavior to characteristics of the person rather than to the stimulus properties of his environment.

In other words, people are not naive stimulus-response theorists, but, rather, they are naive 'black-box theorists,' looking inside the organism for the causes of behavior rather than to external stimuli."[2:1] (p. 177)

McArthur even catches herself making the same error:

“Perhaps I am too, for I have just attributed a high incidence of person attribution to the characteristics of my subjects rather than to characteristics of my items."

This systematic bias toward person attribution—even when situational factors are clearly present, has become central to attribution research.

From Covariation to the Fundamental Attribution Error

Five years after McArthur's study, Lee Ross formally articulated what he called the "Fundamental Attribution Error" (FAE):[3]

The Fundamental Attribution Error: People systematically underestimate the power of situational factors and overestimate the power of dispositional factors when explaining others' behavior.

The connection to covariation theory:

Why consensus information gets ignored:
McArthur found that consensus accounted for only 6.25% of variance in person attribution (compared to 21.72% for distinctiveness). This is in line with the false consensus effect: people assume others would react the same way they would, so they don't seek out actual consensus data[4].

Practical implication for quiet leaders:
Your boss likely isn't systematically checking:

Instead, they're seeing: You underperformed (ignoring the situational circumstances), and ‘so’ You're the problem.

That's why making your situational constraints visible matters so much.

Why This Matters for Attribution Intelligence

Understanding the full framework helps you recognize the nuances of attribution errors:

1. Stimulus vs. Circumstance matters for intervention

2. Combination attributions are common
Most real-world situations involve multiple causes. The person AND the situation both contribute. Recognizing this prevents black-and-white thinking.

This goes back to Kurt Lewin’s original observation about The Psychology of Everything[5].

3. Low consistency is your defense
When defending yourself from misattribution, a powerful variable is showing: "I don't usually perform this way" (low consistency). This shifts attribution away from stable person factors toward temporary circumstances.

4. High consensus is your ally
When you can show "others also struggled in this situation” (high consensus), you also shift attribution from person to situation. But you have to actively gather and present this information. Your boss won't seek it out on their own.

References

  1. Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28(2), 107–128. https://doi.org/10.1037/h0034225 ↩︎
  2. McArthur, L. A. (1972). The how and what of why: Some determinants and consequences of causal attribution. Journal of Personality and Social Psychology, 22(2), 171–193. https://doi.org/10.1037/h0032602 ↩︎ ↩︎
  3. Ross, L. (1977). The intuitive psychologist and his shortcomings: Distortions in the attribution process. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 10, pp. 173–220). Academic Press. ↩︎
  4. Ross, L., Greene, D., & House, P. (1977). The “false consensus effect”: An egocentric bias in social perception and attribution processes. Journal of Experimental Social Psychology, 13(3), 279–301. https://doi.org/10.1016/0022-1031(77)90049-X ↩︎
  5. Lewin, K. (1936). Principles of topological psychology. McGraw-Hill. ↩︎
Share Share Share Share Share Email

Read next

Research Deep Dive Day 6: Which Kind of Power Do You Want?