Reliability of Teams’ Game-Related Statistics in Basketball: Number of Games Required and Minimal Detectable Change

ABSTRACT

In basketball, game-related statistics are the most common measure of performance. However, the literature assessing their reliability is scarce. Purpose: Analyze the number of games required to obtain a good relative and absolute reliability of teams’ game-related statistics. Method: A total of 884 games from the 2015–2016 to 2017–2018 seasons of the Spanish men’s professional league were analyzed using all games and clustered by scoring difference. Intra-class correlation coefficient (ICC) was calculated for each variable. The number of games required to detect a change and to achieve good relative reliability was calculated using minimal detectable change and Spearman-Brown prophecy formula respectively. Results: Using all games, the results showed that the minimal number of games required in each group was 30 to detect a medium change (d > .5), 187 for a small change (d > .2), and 100 for good relative reliability (ICC ≥ .75). Using balanced and unbalanced games, the minimal number of games required in each group was respectively 31 and 30 to detect a medium change (d > .5), 190 and 188 for a small change (d > .2), and 191 and 121 for good relative reliability (ICC ≥ .75). Conclusions: The sample needs to consist of at least 30 games in each group to detect a medium size change, and at least 190 games to detect a small size change. To be able to rank teams with good reliability, at least 100 games are required when including both balanced and unbalanced games.

KEYWORDS:

• Spearman-Brown prophecy
• absolute reliability
• relative reliability
• two-step cluster

Supplementary material

Supplemental data for this article can be accessed here.