It’s a great day in sports analytics! Today I’m going to touch on Bland-Altman plot analysis and its application in the world of sports. Bland-Altman plot analysis is a graphical method that involves plotting the differences of two techniques against their mean or average. The differences can also be plotted against one of the two methods if the method has been recognized to be a reference or gold-standard method.
Bland and Altman, in their experiments, discovered two similar methods that can be used for assessing agreement without assuming that the differences between the recorded test and retest scores are normally distributed.
This method uses a plot, known as the Bland-Altman plot and it is constructed as a simple scatter diagram on an XY graph. The Y–axis bears the difference between the test and retest scores (test – retest) while the means of the test and retest scores ((test + retest)/2) are plotted on the X-axis. The non-parametric methods require that the performance analyst calculate the values outside which a certain proportion of the observations fall.
The values can be obtained by determining the differences between the test and retest scores, and then recording the range of values that is left after a percentage (2.5%) of the ‘sample’ data is removed from each end of the frequency distribution.
Performance assessment using the Bland-Altman plot method identifies critical events (called performance indicators) in individual or team sports. Such events are thought to be a key determining factor to success in those particular sports. Many of these performance indicators are discrete, categorical events (counts or frequencies), such as the number of winners and errors in tennis or badminton, number of shots on goal in soccer etc. Others may be ratios of variables that often represent the characteristics elements of efficiency of performance.
The Bland-Altman plot method requires that analysts treat individual sport performances as individual variables, and this is usually done by reporting the reliability coefficients for each of the performance indicators of interest. This process is important especially where the performance indicator used is in a relatively rare event in a given sport (e.g. the number of shots on goal in soccer).
Bland-Altman method also recommends that sports analysts use a minimum sample size of n = 50 during studies that are aimed at assessing agreement with the parametric 95% limits of agreement method (that is, those predicated on normal distribution theory). Also, generating larger sports samples aids sports analysts performing ‘test-retest’ reliability studies on the data.
Conversely, when the performance indicator is a relatively common event, such as the number of passes made in a game’s activity, the analyst can perform simple techniques like dividing a single match into two-minute periods and entering the number of passes performed in each time period into the analysis system. This way, a sample of at least 40 time periods for a rugby match (80 minutes duration) or 45 time periods in a soccer match can be generated, and the reliability of the data entry (between the test and the retest) accessed.
In conclusion, Bland Altman plot helps sports analysts access individual player performance as well as the reliability of sports data entry.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Learn about the Groundbreaking Sports Analytics Model coaches and sports analysts are talking about!
Learn all about Sports Analytics here.