I have switched from Forerunner 245 to Forerunner 965 recently (yes, Christmas present :)). On the day that I switched, the race predictions dropped enormously. Now, ten days later, the predictions are still way off, they are worse than all my PR's.
Does the Forerunner 965 use a different algorithm than the 245? And why does a newer model show worse predictions?
In my view - and recollection - the calculations were changed on the FW a while back (More than a year ago at least on the 955 if I recall - can't see it on the graphs now as they only go out a year) and…
On the day that I switched, the race predictions dropped enormously
The first suspect is your HR Max value. Your new watch can estimate your HR Max based on your training history. The accuracy…
I changed the manual value to be 50 beats lower and saw no immediate change in my VO2 Max or race predictions.
Obviously, you have to get a new VO2 Max value: just changing the HR Max value…
So, if Garmin primarily uses speed and HR, it can quite easily be manipulated
Let's back up a bit. Garmin says about VO2 Max estimates before getting into the details about white papers:
"Over the course of your activity, the Firstbeat Analytics engine built into your device examines the relationship between how fast you’re moving and how hard your body is working to maintain that pace."[...}
"Intelligence built into the analysis ensures that only the best, most meaningful portions of your performance data are used."
"Because the analysis is designed to learn about your own performance parameters and patterns over time, recording more walks and runs with your device improves the reliability of your VO2 max estimate."
https://www.garmin.com/en-US/garmin-technology/health-science/fitness-tracking/
What does this mean? The algo is retaining your best, most qualified (data filtering and qualification) running data during each run (sub-maximal or maximal efforts) and compounds this data with previous estimates (learning aspect) to give you a new prediction. Your HR/HRV and ability to sustain a certain pace for a certain time already reflects your shoes, the hills, the temperature, your overall health, your motivation, etc.: it is linked to the physiological cost of the effort.
So, if Garmin primarily uses speed and HR, it can quite easily be manipulated.
I am not sure what you mean here, but clearly if you introduce some bias in the data the watch collects, it will have an effect on your estimates.
In a lab environment, the average error rate of the algorithm is 5%.This is an average. For some people, it will be higher, but over time, with a large number of estimates, it should be around 5%.
HR Max estimations errors change the average error rate of the VO2 Max algorithm and tend to amplify it:
https://assets.firstbeat.com/firstbeat/uploads/2017/06/white_paper_VO2max_30.6.2017.pdf
To be fair, the validation of the algo has been done in a lab, so you are right that real life changing running condiitions, despite the "smart filtering" is more challenging.
However, since Garmin is using machine learning, notwithstaning the risk of chronic bias, the accuracy should be close to 5% which is excellent.
What is even more important is that the process is repeatable and predictable. So accuracy is not the most important thing here. We all use things that are inaccurate but very useful because they are predictable and repeatable (I am thinking about the fuel gauge in my car, for example).
So this was for VO2 Max estimates.
If you take a number of people, all with very close finishing times in a race, their VO2 max's will be very different.
Exactly right. In fact research has shown that LT data is much better predictor of endurance race performance than VO2 Max alone. If you take 2 people with the same VO2 Max, the one with the highest VO2 at LTHR as a % of VO2 max will win the race.
But again, for the race predictor Garmin is using all your running data history to maintain the relationships between ventilation (aka HR/HRV), pace and duration. VO2 Max is a key data point, but only one.
I have noticed that my LT has changed similarly depending on the type of runs I do - intervals with fresh legs and I get a good pace at LT, long progression run and Garmin tells me I'm slower at LT.
Yes, if you have LT auto-detection on, you will see variations depending on the intensity and duration of your intervals. You should rely only on the LT built-in test. It is set up as a calibrated ramp and is going to be more accurate.
As noted above, remember that HR Max also influences the estimation of your LTHR, so whenever you see a change in HR Max, you should perform another LT test to realign the expectations of zones, training effect, etc.
Garmin puts it at 168bpm. However I ran a HM at an average HR of 170, recorded on a chest strap HRM, which I use for every run
Commonly the speed and HR of a 10k is said to be very close to the LT values, while the same data for a 5k would in the 105-110% range of them. So, yeah your LT values seem off.
Isn't LT supposed to be what you can hold for an hour?
Technically speaking LT is a lactate blood concentration inflexion point.
Other thresholds (O2 or VCO2 exchange VT2, or maximum steady metabolic state) tend to happen close to that lactate blood concentration and are often assimilated with it.
The most interesting thing about LT is that improving your lactate clearance rate is better done by training in Zone2 than by training in or around the lactate threshold itself. Hence all the "rage" about 80/20, polarized training etc.
