ATP Tennis/Non-India Davis Cup

[prasen9]

In this decade, the decade of Djokovic and Nadal, Federer has won 4 titles, and Murray and Wawrinka 3 each. And as we learned from Rafa and Fed, we cannot write off Djokovic, Murray, and Wawrinka either.

The above comments seemed to be under the assumption that Fed will not increase his tally by many more. It could well be that Fed may win the Aussie and Wimbledon next year. Or the USO.

[Sin Hombre]

Nadal didn't win anything from FO 14 to FO 17 so hard to call it his decade; 2008-13 was his period.

It is commonly forgotten that he turned pro in 2001, just 3 years after Federer turned pro.

[Prashant]

True, Nadal is a huge outlier in terms of how young he got very good. This weekend he became only the third man to win GS titles as a teen, in his twenties and in his thirties. Sampras and Rosewall were the others and Rosewall's first was before the Open era.

[prasen9]

Nadal didn't win anything from FO 14 to FO 17 so hard to call it his decade; 2008-13 was his period.

In the 2010s, Djokovic has won 11 grand slam titles and Nadal 10 grand slam titles. The next best is Federer at 4. I'd say pretty much a Djokovic-Nadal decade. I am not saying that his best period has been this decade. But, in this decade, by far, the most successful men's tennis players have been Djokovic and Nadal. At least going by number of GS titles.

The 2000s was Federer's and 1990s was Sampras'.

[PKBasu]

The pro era obscured this fact, but the greatest tennis player of all time most likely was Ken Rosewall. He never won Wimbledon, mainly because he didn't play there in his prime. In those days, 3 of the four Slams were played on grass, and he could certainly play on grass.
Rosewall won 23 Slam titles in all -- 8 official Slams, and 15 Pro Slams (there were just 3 Pro Slams a year, not four). The Pro Slams were much higher quality than the official (amateur-only) Slams in those days. The year after Rod Laver won the Grand Slam as an amateur (1962), he played Rosewall 13 times on the Pro circuit -- and Rosewall won 11 of those matches in 1963!

[Sin Hombre]

Tennis was accessible to very few people in the 60s even in the western world.

Funnily enough, I was talking to someone who grew up as a girl in the greater Chicago area in the 60s and they played a lot of field hockey.

Of course, the same was true in India and that has gone away as well for the most part.

[prasen9]

Hockey (field) is very popular in our town among girls in the middle and high school. The U.S. always had a lot of girls playing hockey (field).

On an aside, "grew up as a girl", while grammatically not wrong, brought a smile on my face I know, I know, often I write more sloppily.

[Sin Hombre]

^Haha. I wasn't sure where to include her sex.


Anyways going back to scheduled programming, lots of discussion on MTF about silly "statistical" methods to get their guy higher up on the "GOAT list" but it did get me thinking on how I would do it.

I was thinking that of taking a base set of every player ever ranked in the top 5 (10?) and come up with 3 somewhat independent stats which do not penalize players from any generation and reward consistency, longevity and peak performance.

Consistency would be # of YE #1 (though it is easy to come up with others which make sense as well)
Longevity would be # of ATP titles (I think this one is a very good proxy)
Peak performance would be # of GS titles.

For each category, fit a half normal distribution and then calculate the size of the variable for each individual player which would be an indication of how difficult the achieved number was even amongst elite players, and then sum up over the 3 categories.

I am guessing the distribution may well be an exponential decay and that will require a slightly different approach.

Curious how you all would approach it.

[prasen9]

Computer scientists are debuggers by nature. So, I will question a few things.

Why use YE #1? That is a ranking at one point in time. Why not use # of weeks at #1? That would not bring that bias.

# of YE #1s, the # of ATP titles, and the # of GS titles will all correlate highly. All three of these are dictated primarily how long you were great. So, if someone is #1 for 10 years and then retires and the other one is #1 for 15 years, it is near impossible for the #1 for 10 years to have more ATP titles or GS titles even though theoretically it is feasible. Once we do that, we are going to pretty much come up with modern players who have much longer careers.

If we want to figure out who was great for most of their career, we can take all these and normalize by the number of years. That would give us how dominant the player was. That would prefer players who did not have much opposition during their tenure.

As you know, I have sent this one to the category of undecidable problems unless clearly there is someone who is #1 using all reasonable metrics and thus is the indisputable #1 and does not depend upon the metric chosen to be #1.

Going back to first principles, what does it mean to be the greatest player of all time? I think if we put it in words, we will see that we are using different definitions, which results us in using different metrics, which results us in getting different answers.

In case of contemporary ranking systems, we can say whether a ranking system is good or bad by seeing how well it predicts the outcomes of matches. But, when you cannot have inter-generational-players playing at their peaks at the same time, we cannot really use any real data to validate our model.

Which is a long-winded way of not answering your actual question: what ranking function would I use? Who knows!

[Prashant]

How about, start at the beginning of the Open Era, and use ELO (everyone begins with the same baseline) and see who achieved the highest ranking ever. Gut feeling says that would give you either Djokovic or Connors. Just to throw another method into the mix...

[prasen9]

For the current players, you can find the peak ELOs here. And for all players, you can find it here Peak ELO Ratings Djokovic followed by Bjorg and McEnroe. The dominance showed by Djokovic was indeed quite breathtaking. The non-calendar year grand slam was something I think we have never seen in the open era. And, I don't really see a reason why a non-calendar year slam is any worse than a calendar year slam. From the fitness perspective, perhaps a calendar year slam is harder, but, from a form perspective, the winter break actually possibly hinders the momentum.

[arjun2761]

One flaw in the ELO ratings is that it doesn't seem to account for the stage. Clearly, a GS win is worth more than a random tour win as the top players prepare for and "peak" for the grand slams.

However, these ratings do provide interesting data. The Indian top 13 for Peak ELO is
1. Vijay Amritraj (2077)
2. Ramesh Krishnan (1987)
3. Leander Paes (1885)
4. Somdev Devvarman (1841)
5. Anand Amritraj (1760)
6. Sashi Menon (1741)
7. Yuki Bhambri (1723)
8. Premjit Lall (1677)
9. Ramanathan Krishnan (1663)
10. Prakash Amritraj (1624)
11. Jaidip Mukerjea (1619)
12 Ramkumar Ramanathan (1619)
13. Rohan Bopanna (1616).

Meets my gut instinct test with the exception of the pre-Open era stalwarts such as Ramanathan Krishnan who should be much higher (perhaps at the top).

[Sin Hombre]

No Saketh there?

I guess you need to win a minimum of 1 match on the main tour.

[arjun2761]

Looks like they filter for 20 matches - main tour and Davis Cup.

[Prashant]

Good info Arjun - I agree, that rating is certainly a respectable one.