Hacker News Reader

Comment by lloeki | Hacker News Reader

From the paper:

> There are limitations to consider in the interpretation of these results. First, devices were worn for one night.

That is odd.

What I noticed is that my Garmin watch may be wrong at times, but overall provides a fairly good and very useful assessment, especially comparatively day vs day and overall for trends.

In addition it takes a bunch of time for the watch to work out enough data to tune the algorithms for a particular person: some UI bits explicitly call out that for accurate results the watch should be worn continuously for ~2 weeks.

I've now been wearing this particular one (Fenix 7) for 6 months, a serious upgrade from a Forerunner 735xt continuously worn for 5 years. 95% of the time the watch is stupidly accurate in ways it has no business being as a simple wrist-worn set of sensors.

View on HN

My Garmin (255s) is pretty inaccurate when it comes to sleep. It often mistakes REM for awake and vice versa. Honestly, nowadays I only wear it for my runs. Even if it was accurate, I can just tell by feel. All the metrics such as sleep, resting HR and HRV are pretty bad compared to your nervous system.

The metrics aren't really actionable either. I got a bad night of sleep, so what? I know it. I have low HRV because I drank alcohol/over-trained/am ill. I don’t need a watch to tell me that. I know it.

It’s just collecting data for the sake of collecting data.

> It’s just collecting data for the sake of collecting data.

No, sleep is one of the most important parameters for health!

If in 2024 you are not collecting data on your sleep to take action (and for ex, stop overtraining, reduce drinking etc) to reduce the risks of future brain problems, you are doing it wrong.

> Even if it was accurate, I can just tell by feel

I can't, and I fear it may be the same "overconfidence" issue that cause car accidents when drunk.

I collect logs with my sqlite based bash history collection that I've been running for 5 years: https://github.com/csdvrx/bash-timestamping-sqlite

Thanks to my garmin, I have identified a pattern where I feel ok, but my garmin reported less REM sleep than usual: in the day, I run more commands than usual but I have way more mistakes (as can be seen by the non-zero return values)

So now, when my watch reports low REM sleep, I try to be very careful to avoid negative-productivity days

Sleep isn't the metric to track here, there are much better things to analyze such as HRV. For reducing drinking you shouldn't be doing that because of your sleep, you should be doing it because of your overall health. If you care about future health, take blood tests, weigh yourself, evaluate your physical fitness.

Personally, I don't drink, and I've noticed sleep to be the best indicator of my coding performance but YMMV

I've changed my 250 to an Insight 2. It's an improvement and it helped me be more attentive with my sleep, exercise. I don't do runs but walks, hiking, cycling and several water sports during summer. It's also used as a HR sensor for my cycling Garmin device.

If the software compares behavior between nights and can then say "that was relatively good", then you're right and the limitation matters

If the software behaves no differently from one night to the next, that is, if you can take any random night's assessment and it would be as likely to be accurate as the first, then it does not matter that they had participants only wear it one night. You'd get more data (from multiple nights) without needing to recruit more participants, but you'd also be biasing towards whether the tracker works well for this set of people (probably a worthwhile trade-off, up to a point)

Your watch says to use it for a while. It could be that that's just because it's not flawless and they want people to see it's often correct even if they got unlucky the first night(s). Or maybe your watch works differently from the ones used in this study

I'm also surprised they only did one night, both because you get a lot more data per amount of effort by having participants keep it for a month, and because I'd not be surprised if the software is more comparative than absolute. But I don't know if we can draw real conclusions from that

Unless they’re spewing out random garbage (which I doubt they are) they’re good for tracking trends. My overnight RHR is about 55. If I’m starting to get sick, even before symptoms set in, it raises to about 60. By looking at trends I can tell when I’m about to start feeling bad.

Reply on news.ycombinator.com