An exploratory analysis of missing data from the Royal Bank of Canada (RBC) Learn to Play – Canadian Assessment of Physical Literacy (CAPL) project

BACKGROUND: Physical literacy comprises a range of tests over four domains (Physical Competence, Daily Behaviour, Motivation and Confidence, and Knowledge and Understanding). The patterns of missing data in large field test batteries such as those for physical literacy are largely unknown. Therefore, the aim of this paper was to explore the patterns and possible reasons for missing data in the Royal Bank of Canada Learn to Play–Canadian Assessment of Physical Literacy (RBC Learn to Play–CAPL) project. METHODS: A total of 10,034 Canadian children aged 8 to 12 years participated in the RBC Learn to Play–CAPL project. A 32-variable subset from the larger CAPL dataset was used for these analyses. Several R packages (“Hmisc”, “mice”, “VIM”) were used to generate matrices and plots of missing data, and to perform multiple imputations. RESULTS: Overall, the proportion of missing data for individual measures and domains ranged from 0.0 to 33.8%, with the average proportion of missing data being 4.0%. The largest proportion of missing data in CAPL was the pedometer step counts, followed by the components of the Physical Competence domain and the Children’s Self-Perception of Adequacy in and Predilection for Physical Activity subscales. When domain scores were regressed on five imputed subsets with the original subset as the reference, there were small and statistically detectable differences in the Daily Behaviour score (β = − 1.6 to − 1.7, p < 0.001). However, for the other domain scores the differences were negligible and statistically undetectable (β = − 0.01 to − 0.06, p > 0.05). CONCLUSIONS: This study has implications for other researchers or educators who are creating or using large field-based assessment measures in the areas of physical literacy, physical activity, or physical fitness, as this study demonstrates where problems in data collection can arise and how missing data can be avoided. When large proportions of missing data are present, imputation techniques, correction factors, or other treatment options may be required. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (10.1186/s12889-018-5901-z) contains supplementary material, which is available to authorized users.