Barry Revzin contributed to this report.
Swimming Canada has announced it would not be holding national competition in the Windsor International Aquatic and Training Centre due to a current found in the pool during a recent event.
The Windsor pool hosted the 2018 Eastern Canadian Championships back in April. Per a Swimming Canada memo published by the CBC, the pool passed a “FINA appreciable current test” on April 3, but as of the first day of the meet, coaches began to complain about a current. Organizers ran a second current test, but also found no surface current. However, Swimming Canada says it was still concerned about a current beneath the surface and requested that the facility lower the flow rate of its pumps. Athletes who swam 50 meter races on day 1 were given a chance to time trial those events later in the meet.
This Windsor pool was not the one used to host the 2016 Short Course World Championships; that meet was held at the Windsor Family Credit Union Centre.
This pool – the Windsor International Aquatic and Training Centre – is not set to host a national-level meet until 2020 according to the CBC report, so the city believes it can fix the problem before then. The Windsor Star reports that the city is bringing in technicians from Myrtha Pools, the company that built the pool.
SwimSwam’s own statistical expert Barry Revzin ran the numbers from the meet, finding evidence of a current even more dramatic than his high-profile analysis of the Rio Olympic pool in 2016. (You can find more follow-up analysis on that pool here).
Here are some graphs Barry put together showing just how sharp the disparity was between the outside lanes (lanes 0 and 9 are shown in blue and yellow, respectively) and the middle lanes (shown in red), where a circular current would have less impact.
The men’s 1500 took place on day 1, when coaches first complained about the current. Splits show the athletes in the outside lanes trading off – like clockwork – faster and slower splits based on which direction they were swimming.
The bottom of the graph shows the running distance, the left side of the graph the split time in seconds.
The women’s 1500 took place on day 4, after the pump levels had been sharply reduced:
Here’s a few more data visualizations Barry put together tracking the 50 meter events (which would either be swimming with or against the current the entire race, depending on which side of the pool one swam on), including Barry’s own introduction of the data:
Here’s the effect of lane change (prelims à finals) on speed change in the 50s, broken out by day. The 50 Back and Fly were on Day 2, 50 Breast was Day 3, and 50 Free was Day 4. The statement from Swimming Canada says they were running the pumps at 100% during Day 1 prelims, 60% during Day 1 finals and Day 2 prelims, and 20% for the rest of the sessions. The data suggests there is an effect of the pump – we see a large impact in lane change for day 2 (which had the pumps running during prelims – 0.423 cm per second per lane with a pvalue of 1e-7) but we cannot say that there is an effect on either day 3 or day 4 (pvalues of 0.14 and 0.84).
The slope on day 2 is -0.423cm per second per lane. So in the most extreme case, if you were in the most disfavorable lane (lane 9) and you moved into the most favorable lane (lane 0), that would be a shift of -9 lanes, and you would be expected (from this model) to go 3.8cm/s faster. That is, if you swam your 50 free swum in lane 9 in a 24.00, you would be expected to have swum it in 23.57 in lane 0. Note also that the effect is this dramatic despite the pumps being at (a) not even max capacity for (b) only one of the two sessions.
And Barry also analyzed the change in speed over consecutive 50s in the distance races. Again, his analysis:
Another thing to look at is directionally-adjusted speed-change in the distance events. If you drop the first and last 100m off of the 800 and 1500m events, you’d expect that splits would be roughly comparable in both directions. So I looked at the change in speed between consecutive 50s – accounting for direction (that is, positive change on the even 50s and negative change on the odd 50s). For example, Guillaume Lord’s first few 1500 splits were 31.40, 32.03, 34.60, 32.84, 35.21, 33.46, 35.09, 33.62, … I’m skipping the opening 100. After that, the pairwise speed differences are +7.74, -10.25, +7.43, -6.94, +6.23, … I’m flipping every other sign so that they’re in the same direction, giving me the sequence [7.74, 10.25, 7.43, 6.94, 6.23, …]
As before, I’m splitting this into the different sessions: prelims on day 1 one run at 100% (this is “1A”), finals on day 1 was run at 60% (this is “1B”), and day 4 was run at 20%. The data show an enormous effect for Day 1A (2.28cm per second per lane with a pvalue of 0), a large effect for 1B (1.29cm per second per lane with a pvalue of 8e-103), and a very small but still statistically significant effect for day 4 (0.06cm per second per lane with a pvalue of 5e-7).
As a concrete example, the impact we see here is about 10cm/s in one direction for lane 0 during the morning session on day 1. A change of that magnitude means that if you expect to perfectly even split a race at 34s per 50, you would actually swim a 31.84 in one direction and a 36.48 in the other.