Social identification, exercise participation, and positive exercise experiences: Evidence from parkrun

ABSTRACT

Growing evidence suggests that social identities may have profound implications for physical activity participation. Real-world examinations of this relationship have, however, been lacking, with research predominantly examining intentions and hypothetical scenarios. To address this shortcoming and further advance understanding in this area, the present study tested relationships between group identification, participation, two exercise-specific outcomes (exercise-specific satisfaction and group cohesion), and a broad health indicator (life satisfaction) among individuals recruited from parkrun. Participants (N = 289) completed questionnaires measuring all variables except participants’ parkrun participation, which was objectively assessed. Structural equation modeling demonstrated that group identification was significantly associated with greater participation, exercise-specific satisfaction, group cohesion, and life satisfaction. Findings provide real-world evidence of the health-related benefits associated with forming strong social identities in exercise settings.

KEYWORDS:

• Social identity
• exercise
• structural equation modeling
• parkrun

Acknowledgments

We would like to thank parkrun UK and all the individual parkrun event teams we contacted for their support.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ We also measured enjoyment, via Raedeke’s (2007) 8-item version of Kendeirski and DeCarlo’s (1991) 16-item Physical Activity Enjoyment scale, hypothesising a positive association between this variable and group identification. However, confirmatory factor analysis (see Analytic Procedures and Preliminary Analyses section) demonstrated poor fit for this scale: χ²[20] = 253.829, p < 0.001, B-S p = 0.004, CFI = 0.870, SRMR = 0.072, RMSEA = 0.201 (90% CI .180; .224), PCLOSE < 0.001. Alternative models (e.g., covarying error terms for which large modification indices were observed), while significantly improving model fit (e.g., by ∆χ²) – by way of example, after covarying the two pairs of items with the largest modification indices, the model improved to: χ²[18] = 145.706, p < 0.001, B-S p = 0.008, CFI = 0.929, SRMR = 0.051, RMSEA = 0.157 (90% CI .134; .181), PCLOSE < 0.001), ECVI = .631 (90% CI .510, .778), BIC = 247.701, CAIC = 265.701 – still resulted in models with some unacceptably poor fit indices. Furthermore, neither models with items removed, nor two- and three-factor models demonstrated acceptable fit. Thus, given warnings about including poorly fitting measurement models in structural models (e.g., see Bowen & Guo, 2011), we removed enjoyment from all subsequent analyses.

² A subsequent model in which the path from participation to life satisfaction was removed in the interest of model parsimony produced a very similar fit: χ²[51] = 59.121, p = 0.203, B-S p = 0.323, CFI = 0.997, SRMR = 0.038, RMSEA = 0.024 (90% CI .000; .046), PCLOSE = 0.976, ECVI = .393 (90% CI .365, .474), BIC = 212.114, CAIC = 239.114.

³ Given our primary interest in participation as an outcome variable in our hypothesised model, we considered it most appropriate to use participation data for the six months following questionnaire completion throughout these analyses. A test of Model 2 with participation data for the six months prior to questionnaire completion also produced a good fit (χ²[50] = 66.403, p = 0.060, B-S p = 0.199, CFI = 0.993, SRMR = 0.039, RMSEA = 0.034 (90% CI .000; .054), PCLOSE = 0.903, ECVI = .425 (90% CI .368, .513), BIC = 225.063, CAIC = 253.063), while the same paths were significant in both instances.

⁴ Given our primary interest in testing the potential for participation to act as a predictor variable in this model, we considered it most appropriate to use participation data for the six months preceding questionnaire completion for these analyses. A test of Model 3 with participation data for the six months following questionnaire completion produced a similar fit to that observed when it was tested with the participation data for the six months preceding questionnaire completion (χ²[50] = 82.851, p = 0.002, B-S p = 0.064, CFI = 0.987, SRMR = 0.079, RMSEA = 0.048 (90% CI .028; .066), PCLOSE = 0.559, ECVI = .482 (90% CI .409, .583), BIC = 241.511, CAIC = 269.511), while the same paths were significant in both instances.