Garmin Predictor – Still Wrong
A recent update to Garmin’s race pace calculator promised to introduce improvements when you had completed near-distance personal best performances.
For example, if your pace in a recent 12-mile race was the best you’d ever done, your half-marathon prediction would be closely matched.
This is all very obvious, and the calculations have been known for a very, very, very long time.
The same applies to cyclists, where the same type of calculation (with a slightly different scaling factor) can be used to determine your FTP from a 19-minute effort rather than a 20-minute one.
What I experienced a few days ago
I’ve not run a fast 5k for several months now, and my predicted 5k time was about 30 seconds slower than I thought. Fair enough.
I ran a parkrun almost a minute faster than Garmin’s prediction, coming in with a low 80s age-graded time but with a bit more in the tank.
Garmin’s 5k prediction the following day?
A: 40 seconds slower than the time I’d just done.
Yikes
I investigated further. Perhaps the distance was not correctly recorded. It wasn’t. It was recorded as 4880m, so I manually corrected it to 5k on the web interface. The next day, the prediction on my watch was unchanged and wrong. It’s still wrong today by the same amount.
If Garmin can’t get this very, very, very, very basic stuff correct, what hope does it have for the FAR more complicated algorithms it uses for VO2max, lactate thresholds, recovery and the like? You kinda start to get the feeling it’s all nonsense.
Maybe it factors in the fact that you just did that record and you won’t be able to re-do it today, due to fatigue. Tracking it over time might be more useful; what time will show you tomorrow for example?
nope, same today
I stopped keeping up to date with the various watch brands and the metrics they are adding to the watches. Are there other brands that also do this Race Predictor, LTHR, FTP auto calculation stuff like Garmin, but are more accurate?
I find the same with the Polar race predictor.
Ran last 5km, faster than the predictor and the prediction didn’t change, although the Polar one seems to be designed to take more data points over a longer period of time. It also seems to respond too much to long slow runs, eg, if I just run 60 mins at an easy pace, it will predict I am faster, but if I run intervals or a fast race, it will lower it’s estimate.
I think the closest prediction is from Stryd, which has mostly been spot on, they give a +- time which I guess helps them out somewhat! Last 2 HM races (split over 6 months) the target time was achievable and pretty much at my limit on hilly courses.
Stryd predictions are pretty close to real life achievements. For me, at least.
yep
keep in mind, Garmin who “measures” you resting HR during sleep all thought any medical paper will tell you HRRest is determined in still, non motion but conscious state 😴
rhat Garmin HR Rest is typical 5 to 10bpm lower as Suunto,Polar, Fitbit doesn’t bother anybody there since years…
For a long time, the race predictor simply took your estimated VO2max (really more of a VDOT), and spit out the equivalent performances at various distances. This was inaccurate for most because very, very few recreational athletes are capable of running their VDOT performances at longer distances. It was still useful information, though, because it told you roughly what your potential was at longer distances, if you were properly trained.
Because people so rarely hit those times, Garmin revamped the system to account for the kind of training that you’ve actually done. This has proven to be a massive overcorrection, and many people have had experiences like yours. E.g., I once did a 16 mile fartlek where I hit 13 miles in 1:14, and afterwards Garmin said I could run a 1:18 half. This seems needlessly inaccurate. That said, I very much doubt that it’s feasible to come up with a truly accurate race predictor that holistically takes all of your data into account. I’d much prefer that it just went back to the old system where it was just “this is your aerobically determined ceiling, but it may take a lot of work to actually run that time”
yes I agree broadly.
the (cycling) power models are pretty good and only usually fall over at predicting performances beyond your normal training durations.
running adds the complication of technique/form degradation and other factors
other points might be
1. if i’ve just run something i expect the race predictor for the same thing to be correct.
2. I would expect race predictor to precisely predict similar distances with the same energy systems (roughly) equally involved
> That said, I very much doubt that it’s feasible to come up with a truly accurate race predictor that holistically takes all of your data into account.
I like runalyze’s system which also uses VDOT (what it calls “effective VO2 Max”), but it has a correction factor based on actual race results (you have to tag races manually). So if runalyze predicts a 19:56 5K based on an estimated VDOT of 50, but you actually run 21:48 for a 5k race (which corresponds to a VDOT of 45, according to the tables it uses), then it will use a correction factor of 45/50 = 0.9 for future VDOT calculations. As you tag more races, the correction factor is further refined.
It’s not perfect, but I think it’s a good idea. It’s definitely been a lot more realistic than Garmin’s old scheme of simply plugging VDOT into a lookup table for race predictions. For a while (until I had a really bad race), my Garmin FR955 predictions (based on the new scheme which accounts for training volume) and my runalyze predictions were pretty close. (Now the runalyze predictions are much slower lol)
Not only does it get very basic stuff wrong but it’s often contradictory. Eg one part of the app saying you’ve had too much exercise and another saying you aren’t doing enough. I’ve switched to Samsung and whilst it doesn’t have all the running features, it has what I need and doesn’t make stuff up. It also seems more accurate on measuring distances, although not much in it.
That’s why I still like what wahoo does (or does not do) with the Rival. No BS prediction, vo2Max estimation that changes how you feel about your last WO.
Paired with a Stryd it’s still a no frills, efficient training tool.
it would have been interesting to see someone (anyone) make a sports-only watch. as you say, with no BS. and then develop that over time.
rather than moving into all the smart features
I suspect it would have failed! even though i would have worn it for sport and then put my apple watch on afterwards.
the likes of garmin, though, do require the watch to be worn for recovery, sleep and other metrics that ARE important to athletes.
my ideal combo: sport watch + smart ring (or smart bed) + properly integrated smart watch
kinda what i have now with hardware but the data doesn’t link together properly
The last Garmin watch that did that was the 920xt for me.
An excellent watch …… RIP
yes that was a great watch (apart from the colour scheme)
i think if we looked back on it now tho it would be lacking in quite a few sports focussed abiliites, would be interesting to think where it would have gone if the format and ethos was kept for a 9x0xt series
I just clicked on it today after seeing the article, I never normally look at because it’s quite a way out usually.
For me at least the inaccuracy seems worse for longer distances.
Predication is faster for every distance.
It’s 1 minute off my 5km PB.
It’s 2 minutes off my 10km PB.
It’s 5 minutes off my Half PB.
It’s about 40 minutes off my marathon PB. I can categorically state I’ll never in my life run anywhere close to a 3.00 dead marathon as it predicts. I wish.
Interesting, for me it’s spot-on.
I just ran a 3:00:05 marathon and the watch prediction was 2:59:10 (which was very doable if i didnt have to throw up once, yikes!)
other times are slightly on the ambitious side but if I were to focus my training on that distance I’d say it’s also pretty accurate.
some stuff which might help why:
FR 955, sleeping with watch and training with chest strap everytime, also updating my weight twice a week in garmin, using custom zones that I got from lab test
ty
yep i do the same/similar.
my point specifically is that RP ignores very recent achievements above its predicted level. This was specifically highlighted by Garmin recently as addressed.
Interesting, it’s pretty out for me as well but I’d took it to be down to me being fairly new to running regularly and still being in the “big gains” phase. It hasn’t had much historic data so I can probably forgive it. I ran my first half last weekend on a relatively hilly course. It’s prediction for the course was 2:10, my general theoretical best time was 2:05, and I finished in 2:01.
I reported several similar issues, bugs regarding race prediction and daily suggested workouts. They just don’t make sense at all even though Garmin knows me as I wear their watch 24/7 for years. They don’t take the reports seriously. I’m not sure if they don’t understand basic maths and logic in the customer support or they really believe this is how it’s supposed to work. The only way I think it’s not totally random is because the predictions are always too optimistic, and the DSW tells me to do basic run with pace that is faster than my threshold and tempo runs. And all this is based on all the data they auto detect (LTHR, maxHR, zones, etc)