How can we train skills effectively? 🧠

That was the topic when I joined Olympiatoppen’s podcast yesterday, where I discussed this with Caper Ruud’s technical coach, Øivind Sørvald. #tennis #sport #motorlearning #skillearning

There's lots of work behind this study. Here are some images showing how we prepared the hill and course the night before testing. (n/n)

Thanks! 🇸🇯

We have analyzed the ranking and performance of each strategy. More on this in the paper. A short comment: We found big race time differences between the strategies. This finding is important for ski coaches by itself! (7/n)

What information guided the strategy decisions in the two learning groups? Wefound that both learning groups’ decisions could be accounted for by a ’win-stay, lose-shift’ heuristic. Their difference was not statistically significant. (6/n)

We predicted strategy choice could explain our findings, but
reinforcement learning did not show a greater probability of selecting the theoretically best strategy or each skier's estimated best strategy. (5/n)

We found that the reinforcement learning group improved more during the acquisition sessions and performed better during retention than the standard coaching instruction group (supervised free choice) (4/n)

We compared performance and strategy choices of three groups. The supervised (target skill) learning group was instructed to use the (theoretically) best strategy and
therefore served as a benchmark for the upper limit of performance achievable through optimal strategy choices. (3/n)

To test this hypothesis, we delineated four strategies, each carefully selected to enhance skiers’ performance in flat sections in slalom. We defined the “extend with rock skis forward” as the best based on principles from mechanics and quantitative evidence from elite skiers (2/n)

So in principle I could do it like this. Now I just tested difference between gates (named test).

Must be more efficient code, but I just wanted to make sure I get you right.

Thanks. I will have a look. My approach with the gates solution would be something like this. Than I evaluate the expected difference in means at the gates at least. I will look at the stackexchange link now :)

@mjskay.com - With the #tidybayes package, is it possible to extract the expected difference between lowest and highest points during in the waves I have marked with red? Or do I need to switch approach?

This is from a GAM model fitted with brms

Hi again, Daniel. In the method section, do you recommend that I highlight that I used a shorter slalom course and had to skip one training session, or would you recommend to only discuss that in the discussion section?

This is the model. I have tried to use the marginaleffects package, but if I do something like this: avg_slopes(mod) I get Error: Unable to compute predicted values with this model. Do you have some tips?

Hi,

I have a multilevel logistic regression, and I want to calculate the average marginal effect for the two sessions (left and right). It is the difference between the two the treatment groups that is of interest @solomonkurz.bsky.social @vincentab.bsky.social @stephenjwild.bsky.social (1/2)

I feel the sampling average method is too hard and unfair because the strategies have very overlapping distributions. See the example below. Is there a more flexible method that I can use instead of the sampling averaging method to operate with ranges (or something else) instead?