The best use of this data isn't for finding problems, but for evaluating solutions:
- Do I sleep better if I don't listen to techno music before going to bed?
- Does not exercising late in the evening help?
- How about not drinking coffee in the afternoon?
- Lowering the room temperature?
Sometimes the effects are obvious, but often they are subtle, and it's easy to delude yourself (for a while at least) that a solution is helping when in fact it isn't.
Researchers do studies that show e.g. that drinking coffee in the afternoon has a significant effect on sleep (let's say it results in an average of 15 minutes less sleep, p < 0.05).
This is valuable information, but people are different, so the question is, to what extent does this apply to me?
Genetics might help predict this (rs762551 ?), but in practice I'll just need to test this for myself.
The notion that there aren't always one-size-fits-all solutions is well accepted (though often still not well implemented) in medicine (a.k.a. personalized medicine).
"That is researcher's job" - Appeal to authority fallacy. We all can collect stats, run experiments, and draw conclusions. Some of us will have a better grounding in theory than others, but that doesn't mean we shouldn't experiment anyways. It's like saying "nobody but professional chefs should cook"
"Need at least thousands ppl's stats to get a small conclusion". - Not necessarily. It depends on the design of your experiment, the distribution of the data under question, and the desired margin of error. (Amongst other things)
But even if you only roughly estimate those things, you can gain conclusions that are valuable to yourself.
To make up an overly simplistic example: I spend two weeks not drinking coffee, two weeks drinking coffee in the morning, and two weeks drinking coffee in the evening. If the amount and quality of my sleep do not change significantly between the three experiments, it's safe enough to assume that I can consume coffee without influencing my sleep.
It's not a sound experiment, because it doesn't control for any number of variables. It has a wide margin of error. I have not really looked at the sample distribution of sleep duration/quality. And yet, I can make the assumption that drinking coffee in the evening will not affect my sleep.
The point is, it doesn't matter if it's 100% correct. It's roughly correct, most of the time - which is good enough to plan your life around.
It's of course not good enough to make any statements about the general impact of coffee on sleep patterns. But that's not what self-measurement tries to achieve.
- Do I sleep better if I don't listen to techno music before going to bed?
- Does not exercising late in the evening help?
- How about not drinking coffee in the afternoon?
- Lowering the room temperature?
Sometimes the effects are obvious, but often they are subtle, and it's easy to delude yourself (for a while at least) that a solution is helping when in fact it isn't.