Heterogeneous sports participation and labour market outcomes in England

ABSTRACT

Based on a unique composite dataset measuring heterogeneous sports participation, labour market outcomes and local facilities provision, this article examines for the first time the association between different types of sports participation and employment and earnings in England. Clear associations between labour market outcomes and sports participation are established through matching estimation while controlling for some important confounding factors. The results, which are supplemented and supported by a formal sensitivity analysis, suggest a link between different types of sports participation to initial access to employment and then higher income opportunities with ageing. However, these vary between the genders and across sports. Specifically, the results suggest that team sports contribute most to employability, but that this varies by age across genders and that outdoor activities contribute most towards higher incomes.

KEYWORDS:

• Sports participation
• human capital
• labour market
• matching estimation
• sensitivity analysis

JEL CLASSIFICATION:

• Z2
• I12
• I18
• J24
• L83
• C21

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

¹ See also the 2007 White Paper on Sports by the European Commission (http://ec.europa.eu/sport/white-paper/white-paper_en.htm; accessed 12/11/2014).

² Appendix 1 provides further descriptive statistics. Appendix 2 also provides the details and results of the extensive robustness analysis, which includes a variation in the sample definitions, the already mentioned formal sensitivity analysis, as well as an alternative instrumental variable identification strategy. Both Appendices are provided on the internet.

³ In this way, Rooth (2009) finds that physical attractiveness, which in part is an outcome of physical activity, might improve the chances of employability, such that females might be judged more harshly when connected with obesity.

⁴ Stevenson (2010) also discusses this possibility.

⁵ The legal change involved Title IX of the Educational Amendments to the 1964 Civil Rights Act. This banned gender discrimination in federally funded educational institutions.

⁶ The robustness of the analysis is checked by identifying if the 1990 data for income measurements are a source of bias because it reflects the employment status of the physically active at the time. Earlier income before physical activity is used to check for reverse causality. Differences in the health status of twins are also used to check for any subsequent specific differences in health to those that are measured at the time that physical activity is measured.

⁷ This survey is due to be replaced by the Active Lives Survey because of the most recent UK government policy (DCMS 2015).

⁸ http://www.ons.gov.uk/ons/datasets-and-tables/index.html. (accessed 12/11/14) APopS data are generally available at the level of Government Office Region but can be accessed at local authority level by special license, which is the case here.

⁹ From July 2012, which covers the end of APS 6 and onwards, the sample covers respondents aged 14 years.

¹⁰ Activities with participation rates below 1% for all subsamples are omitted.

¹¹ The variables used in this section as part of the so-called propensity score used in the estimator include the number of children in the household, ethnicity, age and age-squared, education, region, local authority characteristics and numbers of different types of sports facilities. Appendix 1 presents the variables and results from the probit analysis.

¹² See also Huber, Lechner and Steinmayr (2015) for operational details of this estimator. The particular version of this estimator used is the RAD_MATCH Gauss package version 3.1.1. It has the feature that sampling weights are accounted for in general. Furthermore, bootstrap inference as described in Huber, Lechner and Steinmayr (2015) is based on weights that are combinations of sampling weights, matching weights as well as regression weights. Furthermore, the improved bootstrap smoother as proposed by Racine and MacKinnon (2007) is used to economize on the required bootstrap replications. In addition, the variable degree and the sample weight are used as additional variables in the Mahalanobis step, in which the propensity score is overweighed by a factor of 5. The distance measure is set to 150%.

¹³ To relate these effect estimates to the appropriate levels of the outcome variables, refer to Table A1 in the Appendix.

¹⁴ Note however that the sensitivity analysis in Appendix B2 suggests that it might be somewhat smaller than 10%.