Estimating NCAA D1 Nationals Qualifying Times

With NCAA conference championship season approaching, many swimmer’s focus is on swimming a time fast enough to get them invited to the national championship meet. Making their lives complicated is the fact that no one knows what time will be required to qualify for the meet. If you want the full details on the NCAA qualifying system (or a reminder), you can check out our detailed explanation.

There is good news for aspiring nationals swimmers. While the cut lines aren’t set in stone, they are more consistent from one year to the next than you might expect from the NCAA’s arcane qualifying system. Here is where the cut line has fallen each of the past 5 years:

Men

Women

The goal is to try and figure out where those times will fall this year. The times generally get faster from one year to the next (about .17% per event, standard deviation .2%). Generally the majority of qualifying times get faster each year. The main exception was the men in 2016 when 8 of 13 times were slower than the year before. Before we blame that on an Olympic year, it’s worth noting that the women’s qualifying times were slower in only 2 of 13 events that year. More likely, this minor slow down can be explained by simple random noise in the data.

The consistency of the year to year change in the data is enough to provide a decent estimate of what the qualifying times will be this year. However, we can do better because we know more information than just last year’s qualifying times. We also know how swimmers have been performing so far this year. A comparison of the current year’s results as of January 31st to the previous year’s results as of January 31st has some correlation with how qualifying times change one year to the next.

To model qualifying time change, I only looked at swimmers currently ranked 25th-70th. For the most part, this represents the swimmers on the bubble to make the meet.  A typical men’s cut line is around 30 and the women’s cut line typically is around 40. I tried different ranges for the men and women because of the different cap numbers, but this range seemed to work best for both. I also tried adding in variables for stroke and distance but they did not improve the estimate at all. This historic data reduces the error of our estimates by 17% for the women and 35% for the men. The trend lines are below. Negative is faster, positive is slower. The red lines represent a 95% confidence interval of estimation.

Men

Women

With that model in place, we can now use it to guess the cut times this year. The best guess of the model is reported as the ‘Estimated Cut’ below. However, it’s unlikely that the cut will fall exactly on any of the estimates. Instead, it’s likely to fall within a range of times around the estimate. The model believes there is a 95% chance the real cut line will fall between the ‘Estimate Lower Bound’ and the ‘Estimate Upper Bound’ listed below.

For the most part the model is pretty conservative predicting small time changes in each event. That is no guarantee that the qualifying times will be faster or slower where the estimate is faster or slower. Even if every assumption of this model is correct, I would expect at least 1 time to fall outside the predicted ranges. That being said, I also expect the majority of the times will fall within the estimate ranges and will fall closer to the center of the estimate range than the outer bounds.

Men

Women

Update: I discovered a slight bug in the code that caused some errors in the estimates. They have been updated to the correct estimates.