What the Walking Evidence Actually Shows
The question here is not whether walking feels refreshing, but what the research can support about walking and two measurable forms of creative thinking. A systematic review and meta-analysis separates divergent idea generation from convergent solution finding, then asks how large, consistent, and transferable the observed effects are. Reading those layers together lets us keep the positive result for generating ideas without turning it into a promise that walking improves every form of creative judgment.
How the review built the comparison
The review began with a broad search rather than a handful of familiar walking studies. The authors searched PsycINFO, MEDLINE, Scopus, and ProQuest from the beginning of each database through January 8, 2025. They also checked the references of earlier reviews, searched Google Scholar using terms related to creativity and walking, and examined the references of studies that appeared eligible. This matters because a result can look impressive when it is built from a convenient subset of the literature. A systematic search does not guarantee an unbiased answer, but it gives the answer a more defensible starting point.
The search produced 5,634 unique citations after 767 duplicates were removed. Eighty-six records went through full-text review. The final synthesis contained 16 articles and 23 individual studies, with 1,036 participants reported in the abstract and study-characteristics section. One later sentence in the discussion says 1,039 participants. That is an internal inconsistency in the published article, not a reason to invent a third number. The repeated report of 1,036 is the safer figure to use, while the discrepancy is worth keeping visible when exact totals matter.
The inclusion rule was also narrower than the word “creativity” suggests. The participants were adults, or students in post-secondary education, and the studies had to examine walking in relation to creative thinking represented by divergent or convergent outcomes. Studies of stair climbing were excluded. Verbal fluency alone was not accepted as a creativity measure because fluency captures output volume but not the novelty required for a creative response. Common divergent measures included the Alternate Uses Task and the Torrance Test of Creative Thinking. Convergent thinking was often measured with a Remote Associates Task, in which several cue words point toward one linking answer.
The review treated different experiments as separate only when their samples were independent. If the same participants appeared in multiple experiments, the authors selected the experiment with the larger sample or the measure with more established reliability and validity. Two reviewers independently extracted study design, participants, walking exposure, control condition, creativity measure, and results. Disagreements were resolved through discussion, and the lead author cross-checked the consensus data. Screening and quality assessment were also performed independently and in duplicate.
Those procedures answer one important question: was the comparison assembled systematically? They do not answer every later question. The review used random-effects meta-analysis, which allows the true effect to vary across studies rather than assuming that every experiment estimates one identical value. It assessed heterogeneity with I² and tau-squared, checked for publication bias in the larger divergent-thinking analysis, and used a creativity-specific eight-item risk-of-bias tool. It then applied GRADE across risk of bias, inconsistency, indirectness, imprecision, and publication bias. In other words, the review did not treat a statistically significant average as the end of the analysis. It asked how much confidence that average deserves.
The denominator is uneven in ways that matter. Sixteen of the 23 studies used the Alternate Uses Task, while only seven assessed convergent thinking, mostly with some version of the Remote Associates Task. That means the positive result is supported by a broader and more varied set of divergent-thinking observations than the selection result. The two pooled estimates are not mirror images with equal statistical weight. A reader who sees one large effect and one small effect should first ask how much information stands behind each number.
The comparator also matters. Walking was not compared with one universal “non-walking” condition that means the same thing in every experiment. The authors had to make decisions when studies included several walking conditions, choosing the intervention that most closely represented natural walking; they also had to account for how a sitting control was instructed. A control that permits small movements, a control that restricts movement, and a control that differs in attention or expectation can produce different contrasts even when each is labeled sitting. This is why the review's heterogeneity is not a footnote. It affects what the pooled difference represents.
Finally, the review's quality judgment has several dimensions. Randomization can reduce selection bias, but it does not make a creativity test a measure of long-term creative value. A statistically stable direction can still be indirect if the participants are mostly students, the test is brief, or the outcome is a laboratory score. Conversely, a nonsignificant result can remain uninformative when there are few studies and a wide interval. The evidence becomes decision-relevant only when magnitude, comparator, design, uncertainty, and population are read together.
The structure of the review is therefore part of its result. It does not pool “creative thinking” into one number and ask readers to decide what that number means. It keeps the generation and selection outcomes apart, because combining them could hide the very asymmetry that matters.
What the estimates mean—and where they stop
For divergent thinking, 20 studies entered the meta-analysis: 11 used a between-subjects design and nine compared conditions within the same participants. The random-effects pooled effect was Cohen’s d = 0.93, with a 95% confidence interval from 0.44 to 1.42 and p < 0.001. Cohen’s d expresses the difference between conditions in standard-deviation units, which makes results from different tests comparable. A value around 0.93 is large by the interpretive rule used in the review. In practical terms, the average pattern is not a trivial fluctuation: after the walking condition, participants generally performed better on the included measures of generating novel ideas.
The strongest check on that conclusion is the sensitivity analysis restricted to randomized studies. It produced d = 0.82, with a 95% confidence interval from 0.35 to 1.28. The estimate is somewhat smaller, but it remains large and its interval stays on the positive side. Randomization helps balance known and unknown differences between groups, reducing the risk that people who choose to walk—or are assigned to walk under systematically different conditions—were already more likely to perform well. The similar direction and magnitude make it less plausible that the positive result is only a by-product of non-randomized studies.
That is strong evidence for a focused claim: walking can improve performance on the divergent-thinking measures represented in this literature. It is not the same as evidence that every walk has the same effect or that the resulting ideas will be adopted. The review rates the divergent evidence as moderate certainty, downrating it for inconsistency. The authors interpret the pooled effect as practically meaningful, but a standardized score still does not tell a manager, teacher, or researcher how many valuable ideas one particular person will produce in a particular setting. The estimate describes an average response under the studies' conditions.
The convergent result is much less decisive. Only six studies were available, all using within-subjects designs. Their pooled effect was d = 0.16, with a 95% confidence interval from −0.31 to 0.63. The review found no statistically significant effect, but the interval includes a modest benefit, essentially no effect, and a modest negative effect. The GRADE assessment rated the evidence very low because of study type, inconsistency, and imprecision. The most accurate reading is therefore that the review found no reliable evidence of improvement in convergent thinking, while the evidence is too limited to rule out a meaningful effect in some conditions.
This difference between “no reliable evidence” and “evidence of no effect” is not a technical quibble. A small, narrow evidence base can produce a near-zero average because genuine effects vary, because the tests are insensitive to the relevant change, or because the studies lack the design needed to separate alternatives. The review could not perform subgroup or sensitivity analyses for convergent thinking. It also notes that convergent performance is more closely associated with working memory and intelligence than divergent performance, and cites randomized work suggesting that walking has limited effects on working memory. That is a plausible explanation for the contrast, not a demonstrated mediator of the pooled walking result.
The moderator analyses make the positive result more useful without making it simpler. Indoor and outdoor walking did not differ significantly for divergent thinking; the location comparison produced p = 0.676. The discussion reports a somewhat larger estimate for outdoor walking, d = 1.19, than for indoor walking, d = 0.92, but the outdoor interval is wide and most of the evidence is indoor: 17 of the 23 studies. The result does not show that nature is irrelevant, and it does not show that nature is required. It says the available data did not establish a reliable location advantage.
Study design shows a similar pattern. The review found no statistically significant difference between between-subjects and within-subjects effects, with p = 0.229. But the observed estimates differed in size: between-subject studies showed a medium-sized effect, while within-subject studies showed a very large effect. Within-subject designs have greater statistical power because each person can serve as their own comparison. They can also be vulnerable to practice, demand, and learning effects. Between-subject designs may be more conservative, but they can be noisier because people differ from one another. Neither design is automatically trustworthy; the contrast tells us why the pooled number needs context.
The same caution applies to the creativity measure. The review found no significant difference between studies using the Alternate Uses Task and those using other divergent measures, with p = 0.857. That is reassuring, but the comparison had limited power because the Alternate Uses Task dominated the evidence. A test's familiarity does not make it a complete measure of creative work. It is useful that the effect does not appear confined to one scoring format, but it remains an effect on measured divergent thinking.
The authors found no evidence of publication bias for divergent thinking by funnel-plot inspection and Egger’s regression test, which gave p = 0.586. This lowers one concern, but “no evidence detected” is not the same as “bias cannot exist.” The number of studies is finite, and statistical tests for publication bias are themselves limited. More importantly, unexplained heterogeneity remains. In one example, three experiments by Murali and colleagues showed I² = 91%. The authors suggest that differences in the sitting control—particularly whether participants were explicitly told to restrict movement—may have contributed. If the control condition permits quiet movement in one experiment and suppresses it in another, the same label does not represent the same comparison.
Other unanswered moderators include walking duration and intensity, the timing of the creativity test, educational level, and baseline fitness. Positive studies also draw heavily from a cluster of work by one author, Oppezzo, and the review recommends independent replication. Most participants were undergraduates or otherwise in post-secondary education, so the result may not transfer unchanged to older adults, children, people with different fitness levels, or people working under high-stakes professional constraints. The risk-of-bias tool was designed for creativity research but has not been validated, another reason to treat the evidence as moderate rather than definitive.
Taken together, the review supports a clear but bounded conclusion. Walking has a large average association with better divergent-thinking performance, and that direction remains when the analysis is restricted to randomized studies. The review does not provide equally reliable support for convergent improvement. The near-zero convergent estimate is a warning against overclaiming, but its uncertainty is also a warning against declaring the question closed. The next issue is conceptual: why should the ability to produce more unusual possibilities and the ability to choose a solution be expected to respond differently in the first place?