Re: 2014-2015 Minnesota Women's Hockey: The Maroon & Gold Strike Back!
eeyore is confusing "momentum" with improving over the course of the year
the "data" eeyore refers to doesn't look at improvement over the year, only a selected portion at the end of the year, "momentum" and the article specifically states that is what it is looking at
Are you actually innumerate? Comparing a team's record over the latter stages, which can be defined in different percentages of the season, of the season to its record over the season as a whole is measuring a team's improvement over the course of the season. I could buy an argument that it isn't measuring the absolute amount of improvement if we are in an environment in which all teams improve over the course of the season but it is most definitely measuring a team's improvement relative to everyone else.
Functionally this is the same as measuring absolute improvement for all of the purposes which you are trying to achieve; an improving record reflecting the absolute level of improvement is just a special case in which the average improvement of all teams is zero. If we are trying to measure the quality of the coaching relative to other teams or the likelihood of winning future games it doesn't matter whether or how that baseline slopes. Regardless, any valuable effects will show up as a team's record improving over the course of the season.
There will be a lot of noise in the data and a fair amount of the variation in winning percentage over the course of a season won't be meaningful for predictive purposes. This is the main reason why looking at the performance over the whole season is more useful than looking at performance of a smaller portion. I could believe that there actually is something to the idea that a team could be playing better at the end of the season and that a team that is has a better chance to win the playoffs. The problem is that this effect, if it exists, is sufficiently minor that it gets swamped by the noise of the smaller sample size. If it's really there, no one has found a way to identify those teams for which it is real and for those for which it's just a mirage.
It's not just playoff performance for which this is true; it applies to hot and cold streaks of all sorts. It's a useful way to try to explain the difference between descriptive statistics and predictive ones. If we look at the past we absolutely can find times when teams or players were performing better or worse than their expected baseline. It's not even just random chance and an outlier within the established distribution of performance. Teams really do appear to go through periods when they have a greater or lesser chance of winning and individuals getting a hit or whatever than they normally do; the contortions you have to go through to make all outcomes come out of the same distribution make it really unlikely.
However, while hot streaks clearly exist in retrospect as descriptions, they have almost no predictive power. Just because a team has won a bunch of games recently doesn't mean that they are more likely to win the next one than would be expected from using the larger sample size of a whole season (or, in the case of individual statistics, more than one season). Whatever conditions lead to a hot streak will eventually come to an end, almost always completely unpredictably without any discernible warning. So, while a team may have a sixteen game winning streak going, the probability that the 16th game marked the end of the streak is sufficiently large that, along with the noise inherent in smaller sample sizes, it means that you shouldn't be more confident that they will win that 17th in a row than you would be if they weren't on the streak.
) that was made in a recent discussion in another thread:
