Two archery techniques at 70 metres: Loh 2018 study

A Malaysian university recurve archer, aged 24, shot at 70 metres with two techniques the authors call "angular", with an open stance and a draw that brings in a rotation of the torso, and "linear", with the string drawn straight back to the chin. There were nine conditions: morning, early afternoon, night, heat up to 45 degrees, rain, calm, wind, overcast and six arrows in two minutes. For each technique and condition he shot 6 ends of 6 arrows, 648 arrows in all, over two months.

With the angular technique the archer put more arrows in the 10: 40 against 21, that is 55.6% against 29.2%. The authors then used a classification algorithm (Random Tree, in the WEKA software) to see whether the technique could be worked out from the scores: classification was correct in at least 83.3% of cases when conditions were considered, in 97.9% for the extreme conditions and in 98.1% when the first three ends were compared with the last three. In the six-arrows-in-two-minutes condition almost every case was misclassified. The authors conclude that the angular technique performs better and copes better with outdoor conditions.

Testing took place on the archery range of Universiti Sains Malaysia, with separate sessions for each condition. After each set the target was photographed with an iPhone app (Archery Score Pro) and each arrow converted into a score from 1 to 10; an arrow on a line took the higher value, and arrows off the target were discarded. The conditions were defined in words: morning from 7:30 to midday, early afternoon up to 16:00, night from 20:00 to midnight, wind between 1 and 5 m/s, overcast with a breeze below 1 m/s, heat in direct sun up to 45 degrees.

The main analysis does not compare scores with conventional statistics. Technique, condition and the score of each arrow were fed to the Random Tree classifier, assessed in "full training" mode, that is on the same data it was built from, with 80% as the target accuracy. The authors take the first three ends to represent the archer while still fresh and the last three the archer when tired.

A classifier assessed on the same data it was built from produces high percentages almost by design: without validation on separate data, the 83.3% and 98.1% say little. The authors themselves note that with only two classes the task is easy. There is also no direct comparison of the two techniques' mean scores with any measure of uncertainty: the only figure on performance is the share of arrows in the 10.

That share does not square with the design: 40 out of 72 gives 55.6% and 21 out of 72 gives 29.2%, but each technique accounts for 324 arrows, and the text does not say which 72 these are. This is one archer, who may simply have been more practised in one technique than the other, and the conditions are described in words rather than measured with instruments, with nothing to show they were repeated on different days: each condition carries with it everything else a given day brings. The two techniques are described in two lines and a figure, with no measurement of the movement.

For anyone who shoots, the study does not help choose between two ways of drawing the bow. It shows that one archer, in his sessions, put more arrows in the 10 with the draw the authors call angular, and nothing more: we do not know which of the two was his usual technique, and an archer shooting better with the movement he has practised for years is what one would expect.

The nine conditions are a sensible idea for anyone wanting to understand their own shooting outdoors: night, rain, heat and reduced time all come up in competition. In the six-arrows-in-two-minutes condition the classifier misassigned almost every case, and the authors think the time pressure put the archer under stress; that is their reading, not a measurement.