Disclaimer: The views on this site are entirely my own and do not represent those of any employer, past or present. This post reflects on graduate coursework. Where I describe published research I am reporting what those papers say, including where the field disputes them.
Week 2 of a course on the psychology of competition. Two short prompts, answered in the way you answer something you already know about yourself. Then the class, the paper, and a six-question quiz I finished in one minute with full marks — on a model my own reflection had contradicted a few hours earlier.
Everyone has a theory about whether competition brings out their best, and almost everyone’s theory is built from two or three vivid memories. A deadline that produced the sharpest work of your year. A comparison with a colleague that made you smaller for a month. Both memories are real, which is why the argument never resolves — and why I answered the prompt without hesitating.
What I wrote first
The prompt was blunt: Do you think that people perform better or worse when they are competing? Why?
“The goal for most people is to be the best. Generally, people thrive under competition, though some might struggle with the pressure and freeze up. Task familiarity likely plays a big role in how competition is perceived.”
I reached for childhood — racing to a tree, solving puzzles — and then for television, because the example that came to mind was The Great British Baking Show, where contestants help each other at cost to themselves.
Read it back as a claim rather than a reflection and the structure is: a confident main effect, a hedge, and an unexamined moderator. Generally people thrive — direction stated. Some might freeze up — exception noted and immediately dropped. Task familiarity likely plays a big role — a moderator proposed and never used again.
That is a specification with a known edge case in it that nobody costed. I have reviewed hundreds of those.
Three things I was calling one thing
The class opened by splitting the word I had been using loosely. There are three separate questions hiding inside “competition”:
- Structural competition — is there an actual, externally organised contest? A game, a tournament, a ranked review cycle.
- Trait competitiveness — how competitive is this particular person, as a disposition?
- Perceived competitiveness — how competitive does someone believe the people around them are being?
Trying to outrun the person beside you at the gym is trait competitiveness. Noticing that they are trying to outrun you is perceived competitiveness. Neither requires a race to exist.
My baking-show example is structural. My childhood examples are trait. My release-night intuitions are almost entirely perceived. I had been treating one word as one variable, which is the measurement error I write about constantly and had just committed in a graded document.
What the paper reported
The assigned reading was Murayama and Elliot’s meta-analytic review in Psychological Bulletin — the paper proposing the opposing processes model. The abstract does not ease you in:
“The first meta-analysis revealed that there is no noteworthy relation between competition and performance.”
Then the finding that matters:
“The absence of a direct effect is the result of inconsistent mediation via achievement goals: Competition prompts performance-approach goals which, in turn, facilitate performance; and competition also prompts performance-avoidance goals which, in turn, undermine performance.”
Inconsistent mediation is the technical name, and it describes a situation most people meet regularly and misread: two indirect paths running from the same cause to the same outcome with opposite signs. Each is real. Each is measurable. Their sum is approximately nothing.
The dashed line is the finding everyone quotes. The two solid paths are what actually happens. The zero is not an absence of effect. It is a cancellation, and it describes nobody in the sample.
The mechanism was given in class as a formula I have not stopped using since: motivation = energization × direction. Competition reliably supplies the energization. It has no opinion whatsoever about the direction. That is the whole model in five words, and it is why the average comes out flat — you are adding force to a vector whose sign you never checked.
My own answer contained both paths
The second prompt asked us to define the four goals in Elliot and McGregor’s 2×2 framework, and state which we hold. Two dimensions: the standard of comparison (mastery — against yourself or the task; performance — against other people), and the valence (approach a good outcome, or avoid a bad one).
I answered honestly, and named both halves of the model I had not yet read.
The 2×2 · and what I claimed for myself
| Goal | Definition | What I claimed, pre-reading |
|---|---|---|
| Mastery-approach | Learn and improve against yourself or the task. | Claimed |
| Mastery-avoidance | Fear of falling short of your own standard. | Described, not claimed |
| Performance-approach | Strive to do better than other people. | Claimed — the + path |
| Performance-avoidance | Strive to avoid doing worse than other people. | Claimed — the − path |
My words for performance-approach were straightforward: aiming to excel, wanting the highest grade. My words for the fourth were quieter, and I did not label them as a problem:
“However, I also leave tasks to the last minute, which reflects Performance-Avoidance Goals.”
Leaving work to the last minute is not laziness in this framework. It is a hedge. If the work is late, the result is about the timing rather than about your ability, and the ability never gets tested cleanly. That is performance-avoidance functioning exactly as described — protecting the self-assessment by ensuring the assessment never happens.
So in one document I reported both mediators running in myself, and in the other half of the same document asserted that people “generally thrive.” I had both paths and reported only the sum I preferred. Not a memory reinterpreted afterwards — a submitted, dated artifact containing the contradiction before I had read anything that would let me see it.
Six out of six, in one minute
Here is the part that actually stings.
That same evening, after the class, I took the quiz on the model. Question four: What sort of goals does competition tend to elicit in people? The options were performance-approach, performance-avoidance, mastery-approach, or both A and B. I answered both A and B. Correct.
Score: 6 out of 6. Time taken: one minute.
So within a few hours I produced two documents. One asserted a confident positive main effect. The other demonstrated, at full marks and at speed, that I knew the effect splits in two. Nothing was learned in between that I did not already have — the 2×2 answer was in my own reflection.
Passing the test measured recall of the model. It did not measure whether I could apply the model to the last thing I wrote. Those are different skills, and only one of them was assessed. I have spent a career watching green test suites certify systems that were quietly broken, and this is the smallest, cleanest example of it I own: a perfect score sitting alongside my own unexamined claim, both timestamped the same day.
The cancellation, and where I have shipped it
The statistical point generalises past competition, and it is why this reading changed how I read dashboards.
A team metric that sits flat quarter over quarter gets read as “no change here.” It can equally be two populations moving apart at the same rate — a stable mean sitting on top of a widening variance. Flatness at the aggregate is one of the most efficient ways to hide a problem, and I have written status reports that did the hiding without knowing it.
So the diagnostic is not a leadership tip, it is a reporting rule.
One flat line · three different situations · how to tell them apart
| Reading | The test |
|---|---|
| Nothing happened | Variance unchanged, distribution keeps its shape. Report the mean and move on. |
| Inconsistent mediation | The mean holds while the variance grows. Split by any plausible moderator — role, tenure, location, shift — and look for two groups moving in opposite directions. If you can name the mediator, measure it directly rather than inferring it from the outcome. |
| The measure is insensitive | Check for a ceiling, a coarse scale, or a construct mismatch. If most respondents sit in one bucket, the instrument has no room left to register movement. |
The rule I took away: never report a null without reporting the dispersion beside it. A mean with no spread attached is an assertion that everyone is the same, usually made by accident.
The field does not consider this settled
One thing I would have missed if I had read only the assigned paper: the reading list for that class included a published response to it, and the response is not a footnote.
Johnson, Johnson and Roseth published “Competition and performance: More facts, more understanding?” in the same volume of Psychological Bulletin, immediately following the article. They credit the contribution and then raise a set of specific objections — chiefly with how the types of competition were defined, and with the clarity and choice of the control condition the competitive groups were compared against. They also question the nature of the performance-approach and performance-avoidance constructs, the comprehensiveness of the model itself, and whether performance-approach goals adequately capture constructive competition.
Two of those land squarely on what I had just learned in class. If “competition” splits into structural, trait and perceived — three different things — then a meta-analysis pooling studies that operationalised it differently is combining conditions that may not be the same condition. And a null result is always partly a statement about the comparison group: no effect relative to what? If the control condition is unclear, so is the null.
Murayama and Elliot published a reply in the same issue. So the sequence a reader gets is finding, critique, and response — three papers, one volume.
The honest summary is therefore not “competition doesn’t matter.” It is: one large meta-analysis reports no reliable direct effect and offers a mediation model explaining the null; that account has been challenged on definitional and control-condition grounds; and the authors have answered. Both halves belong in the sentence.
What I would answer now
Same prompt, shorter answer.
People do not perform better or worse when competing. Competition supplies energy and takes no position on direction. It amplifies whichever question a person is already carrying — can I show this is good? or can I avoid being the one who broke it? — and it does not set that question. The framing does, and the framing is decided long before the deadline.
The part I got right in Week 2 was the throwaway line: task familiarity likely plays a big role. I proposed a moderator and abandoned it, which is the same move as reporting a mean and abandoning the spread — noticing that the population is not uniform, then writing the conclusion as though it were.
Thank you for reading. Look at a flat metric you currently trust, and ask what it would look like if two halves of your team were moving in opposite directions at the same speed.
Comments