[yk]

I would read that as "adhoc guess of people outperforms statistical model," so the model probably misses some factor that people just take into account. As an example, someone who knows that they have cancer and that therefore their health will rapidly decline in the near future will likely not report that they are leading a healthy and active live, even though data right know shows that they are trying to keep in good shape. Or a combination of many similar but less spectacular things.

Not to belittle the researchers, but trying to outperform a brain that gets data about the individual situation 24 h a day is a quite hard task.

[dcow]

Outperforming the brain is different from making poor assumptions regarding data collection, isn't it? It sounds like in your example the model is not winning because of inaccurate setup but at the end you're conceding that humans might just be innately better than the model.

[paultopia]

I was thinking something similar. Like some of the people who say they aren't active are people who experience pain or discomfort from activity, and hence consciously avoid it (and that's salient so they report inactivity); some deaths are resulting from whatever caused that pain. Or something like that.