Fitness tests and archery talent: Taha 2018 study

Fifty young Malaysian archers were split into two groups on the basis of fitness tests and score, and the fitness tests alone then recovered those groups with an accuracy of up to 97.5%. The archers, 37 boys and 13 girls aged 13 to 20, came from several development programmes in Malaysia. Each did six tests (grip strength, vertical jump, standing broad jump, one-leg balance, and push-ups and sit-ups in a minute) and shot one end of six arrows at 50 metres.

A cluster analysis of these seven measures separated a "high potential" group from a "low potential" one, and the first had better mean values on all of them, score included. Six versions of a support vector machine (SVM) classifier then tried to recognise the two groups from the fitness measures. Four of them (linear, quadratic, cubic and medium RBF) reached 97.5% accuracy, an error rate of 2.5%; the coarse RBF reached 87.5%; the fine RBF reached 67.5%, because it recognised nobody in the high group. On ten archers held back for a final check, the first three made no mistakes, the medium RBF one, and the other two several.

With grip strength removed from the model, accuracy rose to 100% for the linear, quadratic, cubic and medium RBF versions. The authors conclude that grip does not help to tell the groups apart, while the other measures contribute roughly equally.

The tests follow standard protocols. Sit-ups and push-ups were counted over one minute, with a single attempt to avoid tiring the archers; balance was the standing stork test, on the ball of one foot with hands on hips, timed and averaged over both legs; grip was measured with a dynamometer; the standing broad jump and the vertical jump (with a Vertec) were the best of three attempts. Shooting took place on a range set up like a competition: ten arrows at 50 metres, the first four as a warm-up and the last six scored. The paper does not say what bows the archers used or what target face they shot at.

Before the analysis, outliers were removed with Grubbs' test, without saying how many. The groups were formed by hierarchical cluster analysis (XLSTAT) on the six fitness measures plus the score. The data, rescaled from 0 to 1, were used to train the classifiers in MATLAB with five-fold cross-validation on 40 archers; the remaining 10 served as a final check. The models were assessed on accuracy, sensitivity, specificity, precision, error rate and the Matthews coefficient, and each measure's contribution was estimated by leaving it out in turn.

The main limitation lies in the design. The two groups are not archers who went on to succeed and archers who did not: they were built from the same fitness measures that the classifier then uses to recognise them, and in the cluster analysis the score is only one variable out of seven. Recovering, from those measures, groups made from those measures is close to a foregone conclusion, so the 97.5% says little about predicting talent. The "high potential" label is not tied to any later outcome: nobody was followed over time, and the score is a single end of six arrows.

The numbers are small, 40 archers for training and 10 for the check, after an unstated number of outliers had been removed. There are inconsistencies too: an age range of 13 to 20 with a standard deviation of 0.6 years is hard to reconcile, and the formula for the Matthews coefficient printed in the paper is not the standard one. The same 50 archers appear in a conference paper by the same authors using a neural network, and probably also in a book chapter that adds a psychological questionnaire: these are not independent confirmations. The authors themselves acknowledge that the sample and its origin do not allow the findings to be generalised to other levels.

For a coach deciding who joins a youth squad, the study does not offer a ready-made battery. It says that in this group the youngsters who jumped further, did more push-ups and sit-ups and held their balance longer also tended to score a few more points over six arrows, and that these measures hung together well enough to form two distinct groups. It does not say how much fitness weighs on score, or whether a fifteen-year-old with better test results will become a better archer.

The finding on grip strength deserves a note, because it is a test widely used in selection: here it did not help to separate the groups. That too applies only to these 50 archers and this method. In the registry the study gives weak support to the claim about fitness test batteries for young archers, with the caveat that its design is circular.