Comment by schneems | Hacker News Reader

Comment by schneems | Hacker News Reader

Parent Full thread

> it does seem like negative reinforcement works better than positive reinforcement

It 100% seems that way, but perhaps not for the reason you might think.

In “thinking fast and slow” they highlight that fighter pilots seemed to do better after being criticized rather than praised. They eventually figured out that it’s due to regression to the mean rather than due to the praise/criticism.

Effectively: on average you’ll perform at your baseline. If you do something worth of praise the next likely thing to happen is that you’ll perform at baseline again (so it looks like praise didn’t work). Versus if someone is critiquing a poor performance your next thing will likely be better (but still baseline). This gives the impression that criticisms works better than praise. But it’s really just regression to the mean/baseline.

It’s over generalizing to say that’s the cause 100% of the time, but I bring it up because not intuitive yet makes total sense. I choose to take this and believe that in the long term praise has more upside than punishment (assuming the recipient has high standards and wants to perform well).

Hmm was this for very experienced pilots?

For newer pilots going through training, it seems like the baseline should move progressively higher with training, and so you’d want good data to know what kind of feedback accelerates that baseline improvement.

For experienced pilots who’ve sort of maxed out their inherent talent/capability, it would make sense to me that over and under performance would be somewhat random.

Anecdotally, I’ve never found feedback to be a one size fits all kind of thing. Everyone responds differently and needs something a little different from you.

Really what people need is an accurate sense of their performance (where they are at), an alignment with a goal (where they want to be), and a concrete action plan on how to make up that delta (how to get better).

I think instead of asking “does criticism or praise produce better outcomes on average over the duration of training” (my interpretation of your comment) I would ask “how can we ensure our assessments of student performance are accurate and how can we make sure it is translated to the most effective feedback per individual student.”

If you HAD to generalize to only positive/negative feedback, my guess would be that people need both, so an ideal value would be a ratio rather than binary. Say: 2 positives to every 1 negative (for example). But as a sometimes educator it’s really hard to keep track of that ratio for each student (unless it’s written or formalized in some kind of way).

I would agree with you if there was a way to ensure feedback isn't interpreted as a personal attack or a shortcoming.

So, this assumes you can separate feedback from motivation too. Authentic praise (as opposed to generic praise) is more effective in the long run because it reinforces motivation and allows time, experience, and in ideal situations, achievable high standards to take care of negative feedback.

The thing I don't get about regression-to-the-mean is whether there is an implication that your baseline is fixed.

That is, can a fighter pilot move up from being a level 4 in a dogfight to a level 7 with training, or will they forever be a level 4 because it's their baseline? Did Michael Jordan have the same ability as a basketball player after 2 seasons as he did after 8?

Saying “great job” after MJ sinks a shot won’t make him awful and saying “try harder” after he misses won’t make him amazing.

Baseline moves with sustained practice and accurate, actionable feedback.

The comment is speaking. To the feedback of the instructor. If an instructor is shortsighted and only looks at the immediate next effect it will always (incorrectly) seem that criticism produces better results. If an instructor has a longer lens, they will use the best combination to quite the individual’s learning style. They should focus on setting and aligning on goals and making accurate assessments and not worrying about whether they should be more or shouty.

Reply on news.ycombinator.com