The literature search covered theses and dissertations published in Korea until June 2023. The academic databases used for the literature search were DBpia, Korean Studies Information Service System, National Assembly Library, and Research Information Sharing Service. We conducted a literature search on the effects of walking exercise on depression, querying ‘walking’ within each academic database and performing a search within the result for ‘depression.’

The inclusion criteria for the analyzed articles were selected according to PICOS, and the literature was searched [21]. First, all the participants were people. Secondly, the treatment method and program (intervention) included walking exercises. Third, the control group did not perform walking exercises. Fourth, depression was the outcome measure. Fifth, the study was designed as a two-group post-test. Sixth, the dissertation was prioritized in case of duplication of theses and dissertations.

For the meta-analysis investigating the effects of walking exercise on depression, we searched various academic databases comprehensively. We found 99 dissertations and 91 journal articles in DBpia, 45 journal articles in the Korean Studies Information Service System, 14 dissertations and 28 journal articles in the National Assembly Library, and 187 dissertations and 110 journal articles in the Research Information Sharing Service. We excluded 112 duplicate dissertations and 150 journal articles, categorizing them based on title and abstract. This categorization led to the exclusion of 177 dissertations and 105 journal articles. Subsequently, we conducted a comprehensive review of the remaining literature, excluding four duplicate publications, five with inconsistencies in measurement variables, four displaying irregularities in analysis methods, and two unavailable fulltext articles. Finally, we selected 3 dissertations and 12 journal articles. The literature selection process is visually shown in <Figure 1> through a PRISMA flow diagram [22].

Author, year of publication, publication type, participant, sex, age of experimental and control groups, and depression instrument were coded as shown in <Table 1>.

The depression assessment instruments used in the reviewed literature were the BDI, CES-D, GDS, GDSSF-K, SCL-90-R-K, and SDS, with adaptations and occasional utilization of abbreviated versions. The Beck Depression Inventory (BDI) consists of 21 questions regarding emotional, physiological, cognitive, and motivational symptoms, with higher scores indicating more severe depression. It can be administered in a short time, is inexpensive, easy to administer, can be scored, can be used in groups, and has high reliability, validity, and sensitivity. Scores ranged from 0–63, with 0–9 categorized as not depressed, 10–15 as mildly depressed, 16–23 as moderately depressed, and 24 or more as severely depressed.

The CES consists of 20 questions with a maximum score of 60, with higher scores indicating more severe depression. A score of 15 or less indicated no depression, 16–20 indicated mildly depressed, 21–24 indicated moderate depression, and 25 showed severe depression. The Geriatric Depression Scale (GDS) is a 30-item questionnaire with “yes/no” responses. Higher scores indicate more severe depression, with 0–9 indicating no depression, 10–19 indicating mild depression, and 20 or more indicating severe depression. The shortened version of the GDS consists of 15 questions, with scores less than 5 indicating no depression, 5–9 indicating mild depression, and 10 or more indicating severe depression.

The SCL-90-R-K is a short-form psychiatric diagnostic test consisting of 13 questions on depression that are scored on a 5-point scale. The measured score, which varied from 34–92 and represented a conversion of the T-score from the normative table for the general population, indicated higher levels of depression. The Self-Rating Depression Scale (SDS) is a 20-item, 4-point scale. The maximum score was 80, with a score of 49 or less categorized as not depressed, 50–59 as mildly depressed, 60–69 as moderately depressed, and 70 or more as severely depressed.

Specific criteria for coding must be provided, and inter-rater agreement and reliability must be verified before conducting a meta-analysis [23]. To this end, data coding of the selected articles was conducted by one sports psychology doctor and a sports psychology professor with experience in meta-analysis research. A sports psychology doctor was consulted in case of data coding discrepancies.

Meta-analysis was performed using the R meta-analysis package (ver.4.3.1). Effect sizes were calculated as standardized mean differences (SMD) using the post-hoc means and standard deviations of the experimental group with walking exercise, the control group without walking exercise, and the number of participants. The standardized mean difference allows different measures to be standardized to a single unit and evaluated using the same outcome variables. The operational definition of the standardized mean difference used in Cochrane Reviews is the effect size known in the social sciences as Hedges’ g [24]. Cohen’s d is a measure of the sample size. Because Cohen’s d tends to overestimate the effect sizes when the sample is small, it must be converted to Hedges’ g, which is correct for studies with large and small samples [25]. The confidence interval (CI) for the effect size is 95%, and an effect size of 0.2–0.5 was interpreted as a small effect size, 0.5–0.8 as a moderate effect size, and greater than 0.8 as a large effect size [26].

The I2 value was used to test the heterogeneity of studies. An I2 of 25–50% was considered low heterogeneity, 50–75% was considered moderate heterogeneity, and 75% was considered high heterogeneity [23]. In general, heterogeneity was judged to be significant if the ratio of true variance to total variance (I2) was greater than or equal to 50% and the significance probability of the homogeneity test was less than .10 [23]. We chose a random-effects model if the analysis showed high heterogeneity among the studies and a fixed-effects model if the analysis showed low heterogeneity.

To verify the validity of each study, we constructed a funnel plot and Egger regression analysis to verify publication bias [27].

Funnel plots and Egger regression tests assessed publication bias. The funnel plot visually examines, while Egger’s regression test offers a statistical evaluation. The funnel plot of selected literature for the meta-analysis of walking exercise effects on depression is shown in <Figure 2>. Regression analysis showed a bias of-3.0048 (t=-1.26, df=13, p=0.2307), indicating no publication bias.

The meta-analysis result on the effect of walking exercise on depression is shown in <Figure 3>. A random-effects model was employed due to the high heterogeneity among the studies (I2=78%, p<.01). The effect size was -1.00 (95% CI=-1.35, -0.65), indicating a large effect size with a significant difference (z=-5.56, p=.0001). Walking was also found to affect depression.