@ugh said in For the Nth time - LOWER TV Diff - Even optionaly:

The metric here would be the number/regularity of posts that seem to see an issue with it, i.e. the feelsies-rate seems to be very high

Based on that metric we'd better focus all our attention on fixing the RNG instead, regardless of what the data says about it being fine.

@ugh said in For the Nth time - LOWER TV Diff - Even optionaly:

That's why I asked the question whether the TV-difference has the same predictive power for low TV of the underdog (i.e. fresh teams) than for higher TV. Fresh teams in general have fewer dice-tempering abilities which can change dramatically once they get a few level-ups.

Well, lets take a look at CCL season 3's data (since I was working on that recently and already have relevant fields created). We'll compare "fresh" teams (new teams with less than 5 games played) with teams that have 5 or more games played and see how things are:

## New Teams (less than 5 games played)

Correlation between TV advantage (their TV minus the other guy's TV) and match outcome: r = 0.133, p < 0.01

Correlation between cTVP advantage (their cTVPlus rating - other guys) and match outcome: r = 0.187, p < 0.01

TV advantage predicted 46.3% of outcomes for non-draw matches. cTVP advantage predicted 48.9% (ctvpf 86.6%) of outcomes for non-draw matches.

## Other Teams (5 or more games played)

Correlation between TV advantage (their TV minus the other guy's TV) and match outcome: r = 0.166, p < 0.01

Correlation between cTVP advantage (their cTVPlus rating - other guys) and match outcome: r = 0.332, p < 0.01

TV advantage predicted 56.5% of outcomes for non-draw matches. cTVP advantage predicted 63.5 (ctvpf 77.4%) of outcomes for non-draw matches.

So, the relationship between TV advantage and match outcome is weak for new teams - weaker than it later becomes, but weak all around. In fact, based on TV advantage's ability to predict the winner in games that have a winner, TV difference is literally *worse than flipping a coin* at predicting the winner accurately for new teams.

But lets look at those new teams at the suggested breakpoint of 300 TV difference and see how things are 0-290 vs 300-500:

## Below 300 TV difference

Correlation between TV advantage and match outcome: r = 0.075, p < 0.01

Correlation between cTVP advatnage and match outcome: r = 0.164, p < 0.01

TV advantage predicted 44.5% of outcomes for non-draw matches. cTVP advantage predicted 47.3% (ctvpf 81.1%) of outcomes for non-draw matches.

## Beyond the 300 TV difference

Correlation between TV advantage and match outcome: r = 0.139, p < 0.01

Correlation between cTVP advatnage and match outcome: r = 0.288, p < 0.01

TV advantage predicted 68.4% of outcomes for non-draw matches. cTVP advantage predicted 69.3% (ctvpf 76.9%) of outcomes for non-draw matches.

We find that the overall relationship between TV differences and outcomes is still quite weak, especially in comparison to our cTVPlus rating differences, but that the ability of TV difference to predict the winner in non-draw games goes up quite a bit. Is that a case for TV protection? Well, we'll need one more comparison to decide that, unless we're trying to guarantee wins for new teams (which is where the pressure ends up being given the < 50% prediction rate for TV advantage, which suggests that for new teams playing less than 300 TV differences, a TV DISadvantage seems to be a better predictor of winning the game). We need to look at non-new (5 or more games) teams and how playing games over 300 TV difference looks:

## Teams with 5 or more games, 300 or more TV difference

Correlation between TV advantage and match outcome: r = 0.287, p < 0.01

Correlation between cTVP advatnage and match outcome: r = 0.435, p < 0.01

TV advantage predicted 66.9% of outcomes for non-draw matches. cTVP advantage predicted 69.8% (ctvpf 78.1%) of outcomes for non-draw matches.

What that's showing us is that there's not much difference between 300+ TV difference outcomes for teams with less than 5 games played, or teams with more than 5 games played. If anything, the relationship between TV difference and outcome of matches gets stronger the more games a team has played, not weaker, which is what the whole "new team protection" concept is predicated on.

Given that, for these "new teams" the TV advantage had a mean of -42.6 and an SD of 136.4, we're seeing 95% or more of the games they play be in that under 300 TV range anyway, and they have a win rate of 56.3% in matches where they don't decide to concede, I think things are just fine the way they are.