The ACC women’s psych sheets have been released (our summary). This allows us to speculate about the outcome of the meet. The first step is to limit swimmers to their top 3 events (they are allowed to enter 5 and scratch down to 3 at the meet) and score the psych sheet out. This produces:
|Psych Sheet Points|
The biggest flaw is that the divers are currently unranked. Every diver has a score of “NP” so diving isn’t included in the scored psych sheet. This means that this ranking gives us a nice baseline, but it’s not very different from where the Swimulator top times ACC rankings have been for the last month (a more in depth breakdown).
We know for certain that swimmer’s times will change a lot at the meet. Some psych sheet ranks are rock solid in the face of large changes (ex. teams with lots of swimmers who’ll likely never miss finals or ever score) where as others are illusions disrupted by small time changes (ex. teams that are over seeded or have many middle of the pack swimmers with seeds vulnerable to small time swings). We can gain some insight into the stability of each team’s ranking by running Monte Carlo simulations of the meet with differing assumptions. How each of these will work:
- Remove swimmers from all events except their top 3 ranked events
- Modify each seed time by a random percentage based on their team’s performance history. This simulates a taper
- Re rank each event based on the modified times and score out the meet
- Repeat 10,000 times
The first scenario I tried was assuming each team’s taper will look like it did last year. I got the taper numbers from Swimulator’s taper tab which shows how each team’s times improved from this point in the season to the end of the year. Last year, ACC women’s teams improved by the following margins vs their top times as of this point in the season:
Using those tapers with a standard deviation of 1.8% (pretty standard for D1 teams) we get the following place probability chart (again this does not include diving). Each number represents percentage of simulations each team finished in each place. The 0%s represent that it happened in the sim, but less than .5% of the time. The blanks mean it never happened.:
This model picks Louisville as the favorite to win the swimming portion of the meet, but gives NC State a puncher’s chance of an upset. North Carolina and Virginia are pretty locked into the next two spots. However, this assumes that last year’s tapers were a reasonably good predictor of this year’s drops. Team drops are correlated one year to the next, but they aren’t exactly the same every year. Before we can trust these predictions, we need to know how sensitive the predictions are to changes in team’s taper patterns. First, let’s try a couple small changes.
NC State is typically a big taper team, but last year was especially big. Virginia is typically an average taper team, but their new coach used to be on NC State’s staff. It’s reasonable to assume NC State’s taper will be a bit smaller (regression to the mean) and Virginia’s will be a bit bigger this year (new coach, new philosophy). To see what this scenario looks like, I tweaked NC State’s taper down to a more reasonable 2.3% and tweaked Virginia’s up to 1%. This produced the probabilities in the table below.
Louisville becomes an even stronger favorite. Virginia and North Carolina get closer, but North Carolina remains favored for 3rd. Both North Carolina and Virgina have strong diving contingents. Last year North Carolina scored 115 diving points to Virginia’s 111. Neither team graduated any key divers, so diving doesn’t look like a decisive advantage for either team. This scenario still assumes a larger taper for North Carolina than for Virginia. Despite Virginia’s coaching change I think this is reasonable. UNC has a history of big tapers (for example: 2015). Virginia’s mid season taper, by all indications, was pretty substantial. They have two relays currently ranked in the top 3 in the country and several of their returning swimmers are ranked as well or better than they were ranked at the end of last season (ex. Marrkand projects to 15 points at nationals right around her mark from last year, Cooper projects to 14 after not scoring last year).
Next I tried a split the difference approach. I ran a model with each team’s taper number set at the half way point between last year’s taper and the conference average of 1.1%. This accounts for team specific taper patterns while throwing out some of the single year variability of only using one year as a data point by regressing to the mean. The downside to this method is that it might underrate team specific taper training patterns. Despite that, of all the methods in this article, I like this one the best.
Another approach is to throw out history and say everyone’s taper baseline is the same. With everyone’s taper set to the conference average, we get the below probabilities. Suddenly Virginia becomes favored for 2nd. They also have a diving advantage over Louisville which would give an outside, but not impossible chance to win. Another interesting note here is that, despite being ahead of Virginia Tech on the psych sheet, Duke is ranked behind them in this sim with a neutralized taper factor.
Based on your expectations of how teams will taper, some expectations about how the meet will play out should change, but some should remain constant. Louisville is always the favorite. It would require pretty unreasonable assumptions to say other wise. Virginia, NC State, and North Carolina are the next three in some order depending on how you expect times to evolve. Miami and Boston College are the bottom 2.