In the aftermath of last year's Boston Marathon attacks, the Boston Athletic Association (BAA) was faced with—among other, more serious problems—a pair of logistical issues. It had to consider how to provide the nearly 6,000 runners who were unable to finish the 2013 race with an opportunity to qualify for this year's race; it also had to decide how to treat their missing finish times, which were of real significance to the people who had an opportunity to fulfill a dream taken from them. Ultimately, all runners who completed the first half of the race but did not finish (DNF) received automatic entries. The BAA also pledged to provide official finish times for every runner who was still on the course when the race was halted, and made this a priority. I was a part of team of experts assembled to provide the BAA with guidance on how best to predict the finish times of the runners who DNF. Our proposed solution was to find a set of comparison runners who finished the race in 2010 and 2011 and whose running patterns were as close to each DNF runner as possible. These 2010 and 2011 runners were then used to predict the finish times. We did not consider the 2012 marathon due to a heat wave. With the help of the BAA, we created a dataset consisting of all the runners in the 2013 race who reached the halfway point but failed to finish, and all the runners from the 2010 and 2011 Boston Marathons. The two explosions occurred at 2:49 p.m.; based on the latest recorded start time, anyone running faster than 3 hours, 56 minutes would have finished the race. Therefore, we restricted our analysis to runners projected to finish the race in more than four hours. The data consist of "split times" from each of the five-kilometer sections of the course and the final 2.2-kilometer section. Our task was to predict the missing split times for the runners who failed to finish the 2013 marathon. We considered several different prediction algorithms, two of which I will discuss momentarily. First, it was necessary for us to objectively validate the quality of the different prediction algorithms. We wanted to know how close our predictions were to the true finish times. Of course, for runners in 2013 who failed to finish, this is impossible. However, we were able to create an independent validation dataset using the data from 2010 and 2011 by setting aside the finish time for some of the runners. We then applied all of the proposed algorithms to predict the missing finish times in the validation dataset. We compared our projected finish times with the true finish times of these runners. In this way, we were able to assess the accuracy of the different approaches. The most common and straightforward method to predict finish times of a race is to employ a constant pace rule—that is, assume all runners have and will continue to run at a constant pace. For elite marathoners, this might be a viable solution; however, for the nearly 6,000 runners who did not finish the 2013 Boston Marathon, this is not ideal. To understand why, let us consider a hypothetical runner who has just crossed the 35-kilometer mark. Will this runner continue to run at a constant pace for the remaining 7.2 kilometers? Or will this runner hit the "wall" and slow down considerably? A less obvious problem is that the runner may have started too slow, as is common with first-time marathoners, and might be expected to speed up over the last 7.2 kilometers. These different running patterns are the fundamental problem in predicting finish times of a race, and we need to be able to identify them to be able to accurately predict finish times. For runners who were stopped after 35 kilometers but before 40, the constant pace rule predicted only 26 percent of runners in our validation dataset within three minutes of their true finish times. This emphasizes the importance of accounting for the speeding up or slowing down of the runner. To achieve better predictions, we proposed five different prediction algorithms that, in one way or another, account for the different running patterns. While each of these methods outperformed the constant pace rule, a method based on a k-nearest-neighbors (KNN) algorithm worked best. The KNN algorithm is a method that, as a runner, I find very intuitive. The idea behind KNN is to find a set of comparison runners who finished the race in 2010 and 2011 whose split times were as close to each DNF runner as possible. These runners are called the "nearest neighbors." We chose to consider the 200 nearest neighbors, and performed a local linear regression among these nearest neighbors to produce our final predictions. (Oversimplifying, the local linear regression fits a line through the 200 nearest neighbors, and then uses this line to predict the finish time.) Under the KNN approach, we are able to predict 69 percent of runners who were stopped after 35 kilometers but before 40 within three minutes of their finish time. This highlights the usefulness of incorporating all available information on the running pattern into the prediction algorithm and not naively assuming a constant pace. Another measure of accuracy is mean absolute error. We found that on average across all DNFs, the KNN approach predicted the finish time within 1.57 minutes of the true finish time, compared with 3.25 minutes using the constant pace rule. We recommended that the BAA adopt this approach, but they decided to use a constant pace rule that assumes all runners would continue to run at a constant pace. Was this a bit surprising? Absolutely, but I still believe that our approach is much better at predicting finish times than the method employed by the BAA, and the data support this claim. There is one silver lining, though. In most cases, the BAA finish times based on the constant pace rule are lower than our projections. This means the BAA projected finish times are more favorable to the runners than our predictions. Looking beyond the specific issues posed by the 2013 Boston marathon, there are other applications for this approach. Many large road races provide real-time information on competitors' performances via split times and predicted finish times. We adapted our KNN approach to provide real-time predictions as to when each runner will arrive at various checkpoints along the course (e.g. five-kilometer sections of the Boston Marathon) and more realistic estimates of the finish times than provided by the constant pace rule. This information would be valuable to spectators by providing real-time updates estimating when specific runners will arrive at various locations along the course. This is an example of a problem for which the data sources are freely available and the problem is easily stated in language that does not require advanced scientific expertise. I encourage you to download the data from our website and produce your own predictions. This article summarizes the work that Dorit Hammerling, Jessi Cisewski, Francesca Dominici, Giovanni Parmigiani, Charles Paulson, Richard Smith, and I did in conjunction with the BAA to provide official finish times for all of the runners of the 2013 Boston Marathon. Of our team of analysts, Francesca, Giovanni, and Dorit all participated in the 2013 race. For a complete description of the work, see our paper titled " Completing the Results of the 2013 Boston Marathon" appearing in the April 11 edition of PLOS ONE.
Related Posts
Christopher Nkunku s first-half penalty was enought to send Chelsea through to Friday s Europa Conference League draw despite a 2-1 second-leg defeat against Servette, going through 3-2 on aggregate.
Nkunku s 14th minute opener was cancelled out with goals either side of half-time from Jeremy Guillemenot and Enzo Crivelli, but the Blues were able to hold on to progress.
After a nervy start in Switzerland, Chelsea were awarded a spot-kick when Keigo Tsunemoto fouled Mykhailo Mudryk inside the box, with Nkunku stepping up to score from 12 yards as he did in the first leg.
But the hosts would draw level in the 32nd minute when Guillemenot was picked out by Dereck Kutesa before firing past Filip Jorgensen to level the game on the night.
Servette, though, won…
Brighton and Hove Albion manager Roberto De Zerbi has been linked with a number of top clubs across Europe, even with the post of long-term successor to Carlo Ancelotti at Real Madrid. But speaking to the press ahead of his team s Premier League clash with Brentford on Wednesday, the Italian tactician has revealed that there are ongoing talks ongoing about a new contract to keep him at the Amex.
I feel very well in Brighton , De Zerbi said.
I have a great relationship with the players, with the club, with Tony [Bloom], with Paul Barber, David Weir and everybody who works in Brighton.
Yes, we are speaking about the new contract, but at the moment we are speaking, it s not done yet.
In September 2022, Brighton appointed De Zerbi to replace Graham Potte…
Watford have appointed Valerien Ismael as their new head coach.
Ex-Barnsley and West Brom boss Ismael replaces Chris Wilder, who has left Vicarage Road at the end of his short-term deal.
Ismael’s contract length has not been disclosed by Watford but he is the 19th full-time appointment made by the Pozzo family during the past 11 years.
Watford technical director Ben Manga told the club’s official website: “Being able to appoint a new coach so soon in May means we can prepare very well together for next season.
“To be able to welcome someone of Valerien’s experience to Watford is good news for our club. We are all looking forward to working with him.”
It continues the extraordinary churn of managers at Watford under the ownership of the…
Jose Mourinho was on Wednesday named Mauricio Pochettino s successor at Tottenham.
It s a remarkable return to English football for one of the game s most successful, and controversial, managers.
A quick look at the fixture list shows the Portuguese will have to hit the ground running if he is to turn Spurs season around the north London side are 14th heading into the weekend clash at West Ham, 11 points adrift of a Champions League spot.
After the cross-capital clash on Saturday, Mourinho will have to plot a route through a congested fixture list that also sees clashes with some old friends and foes.
Jose s back.
After Porto, Chelsea, Inter, Real Madrid and Manchester United, the Portuguese has now rocked up at Tottenham.
His spell at Old…
Emil Forsberg is confident RB Leipzig will progress to a higher level when Julian Nagelsmann takes over as head coach next season.
Leipzig confirmed in June that Nagelsmann would take over at the Red Bull Arena at the end of his contract with Hoffenheim, who he steered clear of relegation after taking over from Huub Stevens in February 2016 before leading them to successive top-four Bundesliga finishes.
The 31-year-old suffered a 2-1 defeat against his future employers last Saturday, but Forsberg has no doubt the club have made the right choice in naming him their coach from the 2019-20 campaign.
I don t know him personally. From what I see, he is a great manager. He is very popular here in Germany. He did a great job at Hoffenheim, Forsberg told Omnisport.
…
Negotiations are continuing as Manchester United seek to secure out-of-favour Atletico Madrid forward Joao Felix on a loan deal this month.
Joao Felix became Atletico s club-record signing back in 2019 when they purchased him from Benfica for €126million, but he has failed to live up to his potential under Diego Simeone.
The Portuguese international with 28 senior caps has only started seven of the 15 LaLiga fixtures he has been available for this campaign, but has recently hit some form, scoring four goals in his past three appearances prior to Sunday s 1-0 loss to Barcelona.
Joao Felix s performances at the World Cup, coupled with his rediscovered nose for goal, have begun to rehabilitate a depressed transfer value – and he is said to be United s for the ta…
Pep Guardiola believes Manchester City, Liverpool and the other Premier League heavyweights are pushing one another on to higher levels in the manner of tennis greats Rafael Nadal, Roger Federer and Novak Djokovic.
City beat West Ham 4-0 on Saturday and are unbeaten after 13 games of their title defence, with 35 points on the board.
Normally, such a return would see them established as runaway leaders but Liverpool s hard-fought 3-0 win at Watford means Jurgen Klopp s side are only two points off the summit.
Chelsea were also undefeated heading into the weekend before being taken apart 3-1 by Tottenham at Wembley, who themselves have 30 points.
Guardiola overcame West Ham counterpart and City predecessor Manuel Pellegrini at London Stadium, a coach with w…
Newcastle manager Eddie Howe and sporting director Paul Mitchell had reportedly fallen out over the perceived failures of the summer transfer window. The club tried to sign Crystal Palace defender Marc Guehi and Nottingham Forest winger Anthony Elanga and failed on both fronts.
However, according to , Howe and Mitchell have resolved their differences after Howe publicly called for the club to stand united ahead of the international break. They ve apparently moved on, and are now working together towards identifying potential transfer targets for January.
The Magpies are believed to have made four attempts at bringing Guehi in from Selhurst Park in the summer, and they hope Crystal Palace will lower their price which, in August, exceeded £70 million. Come January, the E…
Newcastle midfielder Sandro Tonali and on-loan Aston Villa forward Nicolo Zaniolo are returning to their clubs from Italy’s training camp after it emerged they were being investigated by the Turin Public Prosecutor’s Office.
The Italian Football Federation said the decision was taken because the players “are not in the necessary condition” to be involved in the upcoming Euro 2024 qualifiers against Malta and England, as well as “for their protection”.
The statement from the FIGC did not specify the nature of the investigations which Tonali and Zaniolo, who is on loan at Villa from Galatasaray, were facing.
Reports in Italy claim it is relation to an investigation into illegal betting. On Wednesday, it was reported that Juventus midfielder Nicolo Fagio…
Anthony Knockaert s stunning goal secured a 2-1 Premier League win for Brighton and Hove Albion against rivals Crystal Palace at Selhurst Park on Saturday.
The game appeared to be petering out for a draw before Knockaert picked up the ball on the right in the 74th minute, waltzed inside and whipped a delightful finish in off Vicente Guaita s right-hand post.
Glenn Murray s 100th league goal for the Seagulls had put Chris Hughton s side ahead in the 19th minute before Luka Milivojevic s eighth penalty of the season pulled the hosts level five minutes after the restart.
Palace looked more likely to find a winner after that but Knockaert had other ideas, curling in his second league goal of the season to move Brighton level on points with Roy Hodgson s men.
…