4 ms·
The solution is coming. Life-streaming. Don't rely on self reporting, just track everything and pick out the data that's relevant. Software to identify speci
by kbenson 10y ago
The solution is coming. Life-streaming. Don't rely on self reporting, just track everything and pick out the data that's relevant. Software to identify specific markers in the video streams, such as the presence of food, eating, drinking, exercise, etc will make this much more feasible.
A multi-year study with thousands of participants is possible. If you aren't worried about perfect data, wait a decade and start consuming what I'm sure will be many public life-streaming feeds.
- hyperpape 10y agoThat seems...somewhat subject to selection bias.
- zardo 10y agoAll population studies are, it needs to be accounted for and minimized, but it doesn't have to be eliminated.
- hyperpape 10y agoWith regard to the status quo, you could say: "All population studies are subject to incomplete or inaccurate data. It needs to be accounted for and minimized, but it doesn't have to be eliminated". I guess livestreamers are another data point, so in a perfectly pure framework, they can't actually make your epistemic situation worse, no matter how bad the selection bias is, but I don't see how they're going to make the situation that much better.
- zardo 10y agoI think they do help. For constant dollars, each study has to balance the detail of the data captured against the size of the population included. Anything that allows more detailed data to be gathered from a larger population is helpful. While that population is going to skew wealthy if the study involves the participants making a purchase, that can be counteracted by seeking out under-represented populations and asking and/or incentivising them to participate.
- bluGill 10y ago>I think they do help. For constant dollars, each study has to balance the detail of the data captured against the size of the population included. Anything that allows more detailed data to be gathered from a larger population is helpful. Not exactly because we don't know all the factors to control for. You end up with a bunch of correlations but often it isn't clear why. Are people who eat X foods Y because they eat that, or because the common genetic background of people who eat X are Y? Or maybe it is because before they were born their parents did Z?
- kbenson 10y agoFor the public feeds? Possibly. That's not the entirety of the point though, just a note about a large corpus of public data that will likely be available in the near future.
- nradov 10y agoI think that will really help some people who are in denial about their eating habits. They complain that they can't lose weight even though they "count calories" and "hardly eat". Yet in reality they underestimate portions and don't count snacks, condiments, or beverages.
- Spellman 10y agoThen you also start to run into spurious correlations. Track enough variables in a population and you're bound to get correlations.
- nitrogen 10y agoSo iterate on the correlations and see which ones seem to have a sound mechanism of action and survive further study.
- ASpring 10y agoThis is a big topic in HCI called 'Total Capture'. The name implies that all audio and video of your daily life will be captured and enable many interesting studies.