Ahead of the first Tamworth Cycling Festival this weekend, our resident data nerd, Swanny has been making good use of our historical race performance data to perform some predictive analysis for Sunday’s Nemingha – Nundle handicap race. Irrespective of whether it’s accurate or not, it’s pretty impressive. Thanks Swanny and good luck to you and all the other Armidale Cycling Club competitors this weekend.
Armidale Cycling Club maintains comprehensive results for handicap races, with average speeds available for individual riders going back to 2014 here. Data were extracted for 9 riders entered in the Nemingha – Nundle 100km handicap race to be held Sunday 2nd September 2018. These data were used in statistical analyses to predict possible race times.
Note that the author has had no involvement in compiling these data.
Plots of the raw data with smoothed trendlines are as follows:
Some comments about the riders:
- Ayllie Allen has raced infrequently with the club.
- Three riders have shown a high degree of consistency over almost 5 years of data, Mick Hoult, Dave Munday, and Phil Thomas.
- Pete Harris has shown a high level of performance since 2017, after beginning to train seriously around that time.
- Jason Simmonds started racing in 2017 and quickly achieved a high level of performance.
- Tom Simmonds raced and trained infrequently during 2017.
- Two periods are evident for Andrew Swan, with performance pre-2016 similar to Mick Hoult and Dave Munday. After a serious accident in January 2016 he has not been able to return to his previous level.
- Paul Williams has shown more variable performance, with a dip in the trendline in 2017 co-inciding with a period of injury (not cycling related).
A linear model was fitted to the data to predict average speed for each rider. Average speed was modelled as a function of rider, time period coded as “year:season” (i.e. 2014:Summer, 2014:Autumn, and 2018:Winter), and course (Dangarsleigh, Long Swamp, Boorolong). Race lengths range from 30 to 45 km. To obtain predictions closer to current performance, data were removed for Pete Harris prior to 2017, and for Andrew Swan prior to 2016. The model provided a relatively accurate fit to the data, with an R-squared value of 68%.
The following table shows predicted average speeds for each rider from ACC club races, with lower (conf.low) and upper (conf.high) confidence limits. Projecting these speeds onto the 100.7km race distance for Nemingha-Nundle, we calculate predicted race times e.g.B 2 hours 32 minutes for Dave Munday.
We note that projecting speeds for race distances beyond ACC club races may not be quite as accurate for longer races/courses outside the data. However we, have suggested groups for each rider in the last column based on the predicted race times – the final group allocated depending on the time gaps decided on by the handicapper.
[table id=279 /]
As noted above, extrapolation of the model fitted to ACC club races to a prediction for the 100km Nemingha – Nundle will have reduced accuracy. However, we have two previous times for the race from Andrew Swan in 2014 and 2015, during the period he was performing at a similar level to Dave Munday and Mick Hoult. The times were recorded on a Garmin 510 computer: 2014 was 2h34m17s (39kmh), while 2015 was 2h31m26s (39.6kmh), close to the times predicted for Dave Munday and Mick Hoult.