Exercise Identity

What it is

The difference is between "I am trying to exercise more" and "I train". The first is a project with an end date and a success condition, so a bad month ends it. The second is a description, and descriptions survive bad months, because missing a week does not stop somebody being the kind of person who trains.

The research construct comes in two forms. A role identity is the sense of occupying a position, a runner, a swimmer, a member of a club. A self-schema is a cognitive generalisation about oneself built from experience of the activity, and it matters whether the person treats that part of themselves as important. Both measure whether movement has been absorbed into the self-concept, and the meta-analysis pools across them.

The originator is Deborah Kendzierski, not Rhodes and not any trade book. Kendzierski originated the exercise self-schema in 1988 and 1990, and showed that people holding one were more likely to adopt an exercise programme than people without one. Anderson and Cychosz built the Exercise Identity Scale in 1994, which is the instrument nearly everything downstream uses. Rhodes and colleagues meta-analysed the result in 2016. Popular writing credits none of the three and attributes the whole idea to trade books published twenty-five years later, which is the single most common attribution failure in this subject.

In effect

Research: the meta-analysis

Rhodes, R. E., Kaushal, N., & Quinlan, A. (2016). "Is physical activity a part of who I am? A review and meta-analysis of identity, schema and physical activity", Health Psychology Review, 10(2), 204 to 225. Across role identity and self-schema measures, the relationship with physical activity participation is r = .44, 95 per cent CI .39 to .48, a medium-sized effect. That is the random-effects estimate, and it is the figure in the paper's own abstract. The fixed-effects estimate on the same data is r = .40.

Corrected 11 September 2026. This page previously printed .40 as the headline and described .44 as a secondary-source figure, which was the wrong way round. Both numbers are in the paper: .44 random effects, .40 fixed effects, .48 after trim and fill for publication bias, and .45 in the longitudinal subset against .43 cross-sectional.

The chain of credit, stated once

Kendzierski originated the construct in the late 1980s. Anderson and Cychosz built the scale in 1994. Rhodes, Kaushal and Quinlan meta-analysed it in 2016. That sequence is worth keeping because it is the sequence popular writing collapses.

Everyday life: which way round it arrives

The trade-book version of this finding is an instruction: become the kind of person who moves and the behaviour will follow. The correlational evidence does not support the direction of that arrow, and there is a more defensible reading available.

In the argument this publication makes, identity is what a routine produces. Somebody who has trained on Tuesdays and Thursdays for eight months does not have to decide whether they are a person who trains, because the calendar has already answered. The identity arrived as a consequence, late, and the instruction to install it first asks a beginner to do the hardest part before they have any evidence.

That reading is an argument rather than a finding, and it belongs to whoever makes it. A 2024 meta-analysis of intervention effects on physical activity identity exists and would need reading in full before anybody makes a causal claim in either direction.

What it does not say

It does not say identity causes exercise. The meta-analytic evidence is overwhelmingly correlational and cross-sectional. Identity and behaviour reinforce each other, and the design cannot separate them.

It does not say the identity effect is small. Corrected 11 September 2026: this entry previously read "it does not say r = .44" and instructed the site to print .40. That was wrong. The paper's own abstract gives the pooled random-effects estimate as r = .44, with .40 as the fixed-effects estimate on the same data, so the two are models rather than rivals. What the paper does not say is that any of these estimates is precise: heterogeneity is severe, I squared 88.42.

It does not credit Rhodes with the idea. He wrote the meta-analysis. The construct is Kendzierski's and the measurement instrument is Anderson and Cychosz's.

It does not make identity an instruction. "Become the kind of person who moves and the behaviour follows" is a claim about direction that the evidence does not settle, and this site marks it as an argument wherever it appears.

It does not tell you how identity gets built. The measurement literature is about whether somebody holds one, not about what puts it there.

---