June 19, 2014

Is the USATT Rating System Inflationary, Deflationary, or Stable?

I don't have exact numbers on this, but it's fairly obvious that, over the years, the ratings have inflated. When I started out in 1976 there were only three players rated over 2400 (Danny Seemiller, D-J Lee, and Gil Joon Park, with the latter two from South Korea); now there are 116, and this is only among USA players. There are more foreign players now listed as USA players than before, so this is part of the reason, but the bulk of these 2400+ players are just as much USA players as those back in the late 1970s. Dan Seemiller had just reached top 30 in the world with a rating just over 2500. Insook Bhushan (then Insook Na) had just come to the U.S. from South Korea, and was top ten in the world among women, but was rated only about 2250. These days top ten in the world among women would be about 2650. At one point I was 18th in the country among U.S. citizens with a 2292 rating; these days it wouldn't make the top 100. So yes, the ratings have inflated. (My impression, however, is that any inflation has decreased or stopped in recent years. For one thing, the highest rated USA players now are actually a bit lower than some from the previous generations, but that's offset by the fact that the previous generations had players with higher world rankings and deserved the higher ratings.)

But wait, some of you are thinking, hasn't the level of play improved, and that's why there are so many more higher-rated players these days? That modern players have improved is absolutely true - but that has no bearing on the ratings. As players on average improve, so do their opponents. Think of it this way. If everyone were to suddenly improve 100 rating points in level, there would be no effect on the ratings themselves since opponents would also be 100 points better. And so even though everyone's about 100 points better, the ratings themselves would stay the same. 

The level of play has improved because of more training centers, more coaches, better equipment, and more advanced techniques. For example, backhand play these days is far stronger than it was when I started out. Players attack from closer to the table, making it harder to keep a rally going. And if I could have had some modern sponges back in the early 1980s, I (and most top players) would have caused some serious havoc.

The interesting question here is what has inflated faster, the rating system or the level of play? It's a tough call. I would say a 2000 player from the 1970s is more skilled than a 2000 player of today, but that doesn't mean he'd beat the 2000 player of today, who makes up for his lesser skill with more modern techniques and better equipment. (For this, I'm not going to worry about details like the larger ball, different serving rules, etc.) To use a simple example, I'm fairly certain that any modern 2300 player could go back in time to the 1940s with a sponge racket and be World Champion. The very best players from the 1940s were more skilled than a modern 2300 player, but the 2300 player would have modern sponge, looping, serves, etc. (To put it another way, at my peak, and with my sponge racket, I could have beaten the best players in the 1940s, but I don't think I was a more skilled player than the best hardbatters of that era. An interesting question is how long it would take the best players of that era to adjust?)

So why has the system inflated? Actually, the system would be a deflationary system except the adjustment factor is too high. The inflation comes from all the points pumped into the system from the adjustment factor, where any player who gains 51 or more points in a tournament is adjusted upwards. (There are no downward adjustments.)

If there were no adjustment factor, the system would be deflationary, and the average rating would be dropping. Why? Because the average player improves after his initial rating. Assuming no adjustment factor, let's say that the average first rating is 1200, and that the average player then improves to 1500. That means the player takes 300 rating points from others in the system. Result? Assuming the same number of players in the system, there are now 300 less points distributed among them, and so the average rating goes down - even though the average level of those players has stayed the same. This should be true of any rating system where there's a direct or indirect exchange of rating points.

Let's assume that the average player instead got worse on average. Then they'd be giving the system points, and so the system would be inflationary

One distinction to make here is the difference between the ratings going down on average while the average level stays the same (a deflationary system), and one that goes down because there is a large influx of new players with lower levels. The addition of all these lower-rated players would lower the average rating, but deservedly so since the average level will have gone down. But among the established players, where the level has stayed the same, the ratings wouldn't change, and so the system isn't really deflationary, though the average rating has dropped. 

"Can You Predict the Odds in a Match from their Ratings?" Revisited

Yesterday I blogged about the above. In it I showed why a rating system will always have more upsets at the lower levels than at the higher levels, even if statistically it appears that the odds should be the same at all levels. Here's an easy way of explaining this, using 100-point upsets as an example.

The most accurate rating system in the world is still going to have more 100-point upsets at the lower levels (and upsets in general) for the simple reason that no matter how accurate the rating is at the time the player last played, players at lower levels are more likely to have major improvements than players at higher levels. In other words, the ratings might be accurate at the time the players played, but they become inaccurate at lower levels more quickly than at higher levels. 

Here's a simple example. Suppose you have a highly accurate rating system that accurately rates 20 players. Ten are accurately rated at 1000, and ten are accurately rated at 2500. The next time these 20 players play, the ten who were rated 1000 are more likely to have improved to 1100 than the ten players rated 2500 are to have improved to 2600, and so it's more likely the 1000-rated players are going to be beating 1100 players than the 2500-rated players beating 2600 players. Therefore, it is more likely that these 1000 rated players are going to pull off 100-point upsets than the 2500 players. 

Here's still another way of looking at it. The odds of a 1000-level player beating an 1100-level player may be the same as the odds of a 2500-level player beating a 2600-level player, i.e. 1 in 6. The problem is that it's more likely that a player listed as 1000 is actually 1100 in level than a player listed as 2500 is actually 2600 in level. 

Playing the Middle

Here's a new coaching article from Samson Dubina, "Are You in a Jam?"

Help Wanted - USATT CEO

Here's the job description and application info for CEO of USA Table Tennis. I read over the listing - that's a LOT of requirements!!! I'll probably blog about this tomorrow.

Review of the Nittaku Poly Ball

My review of the ball in my blog on Monday is now an ITTF featured article. (I did a few minor updates to the blog yesterday when they asked if they could use it.)

Follow Your Favorite Players on Facebook

Here's the article, with links to these player pages.

Lily Yip's China Trip Photo Album

Here's the photo album of Coach Lily Yip in China with Lily Zhang and Krish Avvari.

2014 U.S. Open Foreign Players

Here's a chart of the number of players attending from each country. Here's the U.S. Open Home Page. Here's the where you can see who is entered and who is entered in each event. There are 713 total entries.

Ping Pong Summer Openings

Here's a list of scheduled openings for the movie around the country, including Ocean City; Omaha; San Francisco; Phoenix; Miami; Louisville; Grand Rapids; Athens, GA; Goshen, IN; and Winston-Salem.

Table Tennis Camps for Veterans & Members of the Armed Forces with Disabilities

Here's the listing.

Table Tennis Nemesis

Here's the article about author Geoff Dyer and table tennis.

Promotional Video for Waldner & Appelgren's Club Sparvagen in Sweden

Here's the video (1:57).

Table Table Tennis

Here's the video (11 sec) - they are playing with two tables set a distance apart.

Earthly Table Tennis

This is what I call an out-of-this-world ping-pong table. I want one!!!

***
Send us your own coaching news!

One point about comparing 1000 vs 1100 rating and 2500 vs 2600 rating. If we forget about rapid improvement (which usually works only for younger players, or for players who started to get some good coaching - a minority, probably) then there is a much better chance that 1000-1100 rating is more precise (better reflects actual strength) than 2500-2600. Why? because when 1000-1100 players go to a tournament they usually have quite a few matches vs players around their level and therefore adjusting their ratings better and faster to reflect their actual strength. 2500-2600 player at a tournament can easily have just one match (or two if he is lucky) vs players close to his strength. His rating will change (if at all) much slower and will not reflect the actual changes in skill unles he plays a lot of tournaments and meets a lot of high-level opponents.

Same happens with many players in regions where their peers are few - say, you are a 1900-2000 level player in Vermont and therefore you get to play a USATT-rated tournament only a couple of times a year. You go there and you play 5-7 matches, 3-5 of which are against players either 300 points below you, or 200-300 points above you, so you only have one or two real chances of making some progress in your rating. That's not a lot, and on average you don't end up being adjusted. So even if your real strength now is 2050 instead of 1950 nobody would know that by looking at your formal rating which still hovers around 1940-1960 for a few years in a row.

That is not fixable, of course. Just shows that there are many other factors that affect statistical validity of "rating vs strength" theories.

In reply to by JimT

There's always going to be other factors such as this, but overall it's so much easier to go from 1000 to 1100 than 2500 to 2600 (or 1800 to 1900, or any other higher ratings) that this easily overpowers the other factors. Regarding 2500 players, they tend to congregate in regions where there are other players in their range, and so can get ratings at their level pretty easily. Many of them are overseas players who tend to go to the big cities. Plus higher-rated players tend to play more tournaments anyway. Finally, at the higher levels players have long established themselves, and so are usually rated close to their level, until they improve or get worse from lack of practice.