Evaluating The Potential Impact Of Coronavirus On Game Predictions

NFL injury rate coronavirus player absences

NFL Players have a high injury rate, and we compare that to the coronavirus projected infection rate (Photo by Leslie Plaza Johnson/Icon Sportswire)

Summary:

  1. As long as things are going well enough that sports are being played, the overall impact of teams losing players to positive coronavirus tests isn’t likely to be big.
  2. NBA players should benefit from the extra protections of the Disney World bubble, where infection rates should be lower than in the community at large.
  3. NFL players won’t benefit from bubble protection, but still face a much greater risk of getting a typical football injury than contracting coronavirus.
  4. As a result, we are not planning to make any changes to our team ratings or predictive models to try to account for players being quarantined after positive tests.

Player Absences For Coronavirus: The New Sports Risk

Positive coronavirus tests for athletes will continue to happen as we move toward sports restarts.

Take the MLS Is Back restart tournament in Orlando. Several teams had to deal with the impact of players testing positive shortly before tournament games were scheduled to begin, and two teams withdrew from the tournament entirely due to team outbreaks.

Across various sports, for teams that continue to play, the chance of temporarily losing one or more players to a positive coronavirus test is a factor that we initially thought we would need to address in our team predictive ratings — at least until the pandemic situation in the U.S. improves.

After looking into it more, though, we have decided not to make any significant adjustments to our power ratings, or the amount of uncertainty we build into our projection models, for this possibility.

What We’re Investigating: Game Predictions

To be clear, we are not minimizing the overall risk of the pandemic in the US. Escalating infection rates could still result in more sports suspensions, shortened seasons, or cancelled seasons.

We are simply considering the best way to predict the result of a future scheduled game between two teams, assuming the game happens, but the potential exists that one or both teams might be missing players who are quarantined on game day.

In other words, we are talking about modeling the potential impact of an uncertainty, which is an important part of predictive systems. (For example, our projection models currently factor in uncertainty based on the chance that, say, an NFL team’s star quarterback gets injured in a future game.)

For the reasons outlined below, as of mid-July, we believe that player absences due to coronavirus testing should be a relative non-factor from a future game prediction standpoint, assuming games are being played.

The NBA: Risk of Players Missing Games Due to Coronavirus

Over the course of the resumed NBA regular season and playoffs in 2020, we estimate that roughly 5% of the US public is expected to become infected with COVID-19:

  • As of early July, the website Covid19-projections.com projects that 2.7% of the US population will become infected with COVID-19 in the month of August.
  • The projected infection rate goes down slightly in September, to 1.8% of the US population expected to become infected that month.
  • Given that the NBA season will now extend into some of October as well, we think that 5% is a reasonable overall estimate of the US infection rate over the course of the entire remaining 2020 NBA season, assuming all of it gets played.

The NBA is taking lots of precautions, though, and NBA players in the Disney World bubble should be much safer than the typical American.

They may not be safer than people on islands like Hawaii or Guam, where virtually no new infections are predicted. After all, Disney employees will be entering and exiting the park, and the broader community of Orlando is a relative coronavirus hotspot.

However, it seems reasonable to assume that NBA players should be at least as safe as residents of some of the safer continental US states. Wyoming, for example, is expected to see only about 1% of its residents become infected over the remaining NBA season, which is only about 1/5 of the expected national infection rate.

Early results show the impact of isolation in the bubble is working to reduce the infection rate. In early testing, 7% of NBA players tested positive when arriving at Orlando from the outside world. On July 20th, the NBA announced that zero players tested positive among those who had already been in “bubble.”

Individual Player Risk Assessment

Assuming the Disney World bubble is about as safe as living in Wyoming, over the course of the remaining NBA season, one would expect only three or four NBA players, on average, to become infected:

  • Assume that the Disney World Bubble will be as safe as Wyoming, with a projected 1% infection rate over remaining NBA season
  • There will be 330 NBA players in Orlando (22 teams x 15 players per team = 330)
  • 1% infection rate x 330 players = 3.3 expected total infections, or roughly 3 players expected to contract coronavirus

Remember, too, that all the players in Orlando aren’t going to be at Disney the entire time. Around the start of September, the playoffs will be down to the conference semifinals, with only eight teams remaining. By October, the NBA Finals will be underway with only two teams remaining. So that estimate of around three infected players assumes the NBA player population will all be present for the full time period.

Assuming A Worse Case

What if we theorize that NBA players in Orlando are no safer than the public at large, though, and in a much riskier environment than Wyoming?

In that case, one would expect around eight NBA players to get infected during the month of August, on average, then a smaller number players should get infected in September and October once the playoffs start:

  • Assume that the Disney World Bubble will be as safe as the US in general, with a 2.4% projected infection rate in August and 1.9% rate in September
  • There will be 330 NBA players in Orlando
  • 2.4% infection rate x 330 players = 7.9 expected total infections in August
  • Expected infections then decrease in September/October as infection rate drops and teams lose in playoffs and depart Orlando

That’s still a small enough expected number that it’s not worth trying to adjust our team ratings algorithms to try to somehow try to account for the risk.

NBA Infection Cluster Risk

Of course, if one NBA player becomes infected, then it’s quite possible that a cluster of others players do too.

However, if an coronavirus outbreak becomes more widespread, one would assume that the NBA would again suspend or even cancel the season entirely, rather than push forward with the Orlando experiment with a large number of players missing.

(Commissioner Adam Silver recently said as much himself.)

Again, our goal here is to best predict future games assuming they happen. Given that condition, we believe the ongoing risk of ad-hoc player absences due to coronavirus testing is relatively low.

From a game prediction standpoint, it doesn’t appear to be a major additional risk factor beyond the already-present general injury risk of playing NBA basketball.

The NFL: Risk of Players Missing Games Due to Coronavirus

As of mid-July, the NFL plans on proceeding with the traditional scheduling of games in home markets.

NFL players may have a slightly lower risk of contracting coronavirus than the public at large, because teams will presumably do everything they can to educate and monitor their players.

Still, compared to playing in a bubble like the NBA, the risk for NFL players will likely be much closer to public/community averages.

The NFL Is Already Bad For Your Health

Compared to the NBA, serious injuries are pervasive in the NFL. Players missing significant time due to injury is an ongoing, serious risk in such a physical, contact sport.

Granted, knee injuries are not contagious like COVID-19 is. Still, after running some numbers, we believe the baseline injury risk in the NFL dwarfs the incremental risk of players missing games due to coronavirus.

Quarterback Risk Assessment

Let’s focus initially on the quarterback position, since it is the position most capable of significantly changing a game prediction.

Our rough analysis shows that the risk of a starting quarterback missing a game due to a positive coronavirus test is probably only about one-eighth (12.5%) of the risk of that same QB missing a game on account of a “general” football injury.

To start with, one can expect roughly four QBs per month to miss at least one game on account of injury:

  • In 2019, 13 starting quarterbacks played every game and three others only missed Week 17 to rest for the playoffs.
  • Two teams benched starters but not because of injury.
  • Three teams had multiple injuries to starting quarterbacks.
  • On average, the risk of a starting quarterback missing the next game was one per week (17 injuries, 17 weeks).
  • Those injuries kept quarterbacks out longer on average than the expected quarantine period for a coronavirus positive test.

In comparison, based on reasonable assumptions, one would expect only 0.56 QB’s per month missing at least one game due to coronavirus:

  • As shown above, the risk of a QB missing the next week to any injury is roughly 3.1% (1 quarterback per week divided by 32 teams).
  • With four weekends of games on average per month, that puts about four starting quarterbacks missing games for injury each month.
  • The expected US rate of infection on Covid19-projections.com is 1.8% in September and 1.7% and in October.
  • Applying those percentages to 32 starting quarterbacks equals 0.58 infections in September and 0.54 infections in October, for an average of 0.56 expected QB infections over the two months.
  • Those percentages also assume NFL starting quarterbacks have the same risk level as the public, when they probably have a lower risk and will take more precautions than the average American.

That expected frequency implies that over the course of the upcoming 2020 season, only about 1-2 starting quarterbacks on average would be expected to miss games due to coronavirus:

  • As shown above, we would expect about 0.56 quarterbacks per month to get infected.
  • The NFL season is 17 weeks (roughly four months) long.
  • That works out to 2.24 quarterbacks expected to get infected based on the expected infection rate for the general public.
  • Quarterbacks will more likely have a lower risk and take more precautions than the average American, so 2.24 is likely a higher-end estimate.

Would one or two starting quarterbacks contracting coronavirus be a big story? Yes.

Should we alter our projections to try to account for it? We don’t think so. Not when 16+ starting QBs already miss games — and often multiple games in a row — to regular old football injuries each year.

Other NFL player risk assessment

If we assume an NFL roster of 65 players (the 53-man roster, plus potentially an expansion of the practice squad because of the risk of some increased absences), one would expect roughly one player per team to test positive for coronavirus every four weeks:

  • The expected US rate of infection on Covid19-projections.com is 1.8% in September and 1.7% and in October.
  • 65 players per team times an average 1.75% infection rate over these two months equals an expected 1.14 players per month getting infected.

In comparison, one study found that there were 5.9 injuries per NFL game. Not all of those injuries resulted in missed future games, but given that the NFL has a higher injury rate than many other sports, general injury risk seems to be a much greater concern than a player missing games because of coronavirus.

(As one example, of the 34 running backs to carry the ball at least 10 times in Week 1 last year, 23 of them missed at least one game later in the season.)

If the overall situation in the NFL is stable enough that games are being played, players designated as out due to coronavirus ought to make up a small percentage of total absences. Of course, the NFL has stated that anyone that has “close contact” with a person who had a positive test must have two negative tests before participating. This could seemingly apply to a lot of players on a team in the immediate aftermath of one testing positive. However, our assumption here is that if large swaths of a team are ruled out from participating for a few days because of the close contact rule, the game would more likely be delayed or postponed.

Shouldn’t We Try To Model Everything?

Reading through this article, you may be wondering why we talk about whether it’s “worth” adjusting our predictive models for this new risk, when it’s almost certain that some players are going to test positive in the future — even if it’s only a few players.

Why shouldn’t we still try to account for that in some way in our team ratings and game predictions?

The primary reason is because it’s always risky to make changes to a model driven by years of trustworthy historical data, especially a potential change like this one where relevant historical data doesn’t exist.

Any adjustment we make to our models would rely on some pretty big assumptions about future COVID-19 infection rates in the US and the effectiveness of the NBA and NFL’s protective measures. We don’t have a very high degree of confidence in our (or anyone else’s) ability to predict the future in those areas, and if we make bad assumptions, the predictive accuracy of our models will likely get worse.

If there’s a huge potential benefit to taking that risk, we’d consider doing it. But in this case, there’s a decent chance that the small number of players we would expect to be ruled out due to coronavirus aren’t even players that would make a significant difference in team performance levels. If a backup kicker gets ruled out, it almost certainly won’t make any difference in a team’s odds to win a game.

Based on the analysis we’ve done so far, the rewards of adjusting our models to guess at what will happen with coronavirus testing don’t outweigh the risks that we will make the models worse in the process. So, we think doing nothing is the better option.