(Re)examining the effects of athlete brand image (ABI) on psychological commitment: an empirical investigation using structural equation modelling (SEM) and fuzzy set qualitative comparative analysis (fsQCA)

ABSTRACT

Research question: How does athlete brand image (ABI) affect psychological commitment (PC) when operationalised at the dimension- (attribute-) level, and measured using reflective indicators? Previous studies operationalise ABI at a higher-order construct level, and/or measure ABI using formative measures. Such operationalisations obscure potentially different relationships between ABI’s image attributes and PC.

Research methods: A questionnaire was used to collect data from 197 UK respondents over a six-day period within two weeks of the Rio 2016 Olympics concluding. Data were analysed through structural equation modelling (SEM) and fuzzy set Qualitative Comparative Analysis (fsQCA) techniques.

Results and findings: Through SEM, the ABI attributes, athletic expertise, life story, role model, and competition style are positively related to PC, sportsmanship and symbol are negatively related, and rivalry, physical attractiveness, body fit, and relationship effort are nonsignificantly related. Many structural paths between ABI’s attributes and PC are also significantly different. Through fsQCA, high PC exists under three complex ABI attribute configurations, while it is absent under four complex configurations.

Implications: Theoretically, finding different relationships between ABI’s attributes and PC highlights why operationalising ABI at the dimension-level provides a more in-depth understanding of athlete brand image’s effects on PC. Managerially, the findings suggest athletes need only a subset of ABI attributes for high PC. Subsequently, managers can more-efficiently and effectively direct resources towards those attributes that best-capitalise on athletes’ images.

KEYWORDS:

• Athlete brand image (ABI)
• operationalisation
• structural equation modelling (SEM)
• fuzzy set qualitative comparative analysis (fsQCA)
• reflective measurement

Acknowledgement

The researchers would like to thank Professor Ian Hodgkinson (Loughborough University) for his constructive comments on an earlier version of this manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Peter Dickenson http://orcid.org/0000-0002-3516-0341

Notes

1. Arai and colleagues name the ten attributes, ‘subdimensions’, and the three higher order factors, ‘dimensions’.

2. Multi-value QCA (mvQCA) also exists. However, there is debate surrounding its set-theoretic status. csQCA and fsQCA are also regarded as the main QCA approaches (see e.g. Schneider & Wagemann, 2013).

3. Following the logic of Wagemann et al. (2016) and Ragin (2000) we would have removed truth table rows (i.e. before the final analysis stage) contradicting this sufficiency condition for high PC’s absence if they had existed empirically. However, whenever empirical cases of competition style’s absence occurred, empirical cases of high PC’s absence also occurred.

4. To demonstrate this we reanalysed our data using (a) binary logistic regression (bLR) to mimic conditions’ presence/absence, and (b) multinomial logistic regression (mLR) to – conceptually at least – mimic fsQCA’s gradation (with our ‘crossover point’ acting as the baseline category). The bLR analysis suggests significantly more variance is explained than when no predictors are included [Δχ²(10) = 41.234, p < .001; R² = .189 (Cox & Snell), .264 (Nagelkerke)]. Subsequently, the individual parameter estimates suggest role model (b = .478, p = .018, Odds Ratio = 1.613) and life story (b = .385, p = .008, Odds Ratio = 1.469) significantly predict high PC. Meanwhile, including the ten ABI dimensions in the mLR model resulted in significantly more variance explained than in the baseline model [Δχ²(20) = 78.209, p < .001; R²= .328 (Cox & Snell), .369 (Nagelkerke)], and the Pearson (p = .728) and Deviance (p = .754) statistics were nonsignificant. Subsequently, the individual parameter estimates suggest sportsmanship (b = .874, p = .014, Odds Ratio = 2.396) significantly predicts high PC, while role model (b = .441, p = .054, Odds Ratio = 1.555) is marginally significant. Conversely, expertise (b = −.684, p = .048, Odds Ratio = .505), sportsmanship (b = .926, p = .003, Odds Ratio = 2.524), relationship effort (b = −.658, p = .009, Odds Ratio = .518), and life story (b = −.641, p = .001, Odds Ratio = .527) are significant predictors of high PC’s absence.

5. We also reanalysed the data through bLR and mLR using only the ABI dimensions and the associated second-order interactions retained in the final fsQCA stage. For bLR, few significant predictors were found and some odds ratios appeared implausibly high. Similarly, few significant relationships were found in mLR and some odds ratios were questionable. This demonstrates the high demands placed on data when so many interactions are considered (Schneider & Wagemann, 2013). We do not report these results because we would only really know which ABI dimensions to include in the logistics regression models after fsQCA was undertaken (even with painstakingly exploring parsimony). In turn, this strengthens our view that fsQCA and SEM are complementary.