I've had people that talk a lot but really had nothing to say. A lot of messages may not necessarily mean a good diplomat... it just means the guy is a talker.

Also, RD, for your win/draw pct, you need to remove the gunboat games from the calculation.

Your msg/gm and win/draw pct should be barring gunboats entirely.

@ FK, I somewhat agree with that. I disagree in that the game is about building relationships and if talking about things outside the game helps foster a trusting relationship, then it can have an effect. I'm sure much of the time, the comments are rather negligible but not always.

Win % is kinda meaningless if you play different maps, 5 v 7 v 17 players makes a big difference. I would like to see the winning percentages split by map type.

I calculated my formulas before the variants came out on this site. Dunno how they stack against playing variants.

@TruN - Thanks for the clarification. I don´t have time know to iterate on all my games to exclude gunboats and variants, but will do when I have some free time.
Just for the statistics sake, have you tested this fit against some actual data (don´t know if you have access to that, but if you have all the games database it can be extracted) to see if this correlation is still valid or if you can predict the ratio for new players?

lol clear moral of story - dont play gunboat :P

fk - well my point was just kind of that talkers tend to be better players. tend.

@Thucy:
"lol clear moral of story - dont play gunboat :P"

I couldn't have said it better.
Gunboat isn't diplomacy in my mind. It is a different game. It has its nuances and such, but it is the antithesis of diplomacy.

Actually, the gunboats I've played make me feel much better about myself. It justifies some of my losses and inflates my message rate, in my head.

Does anyone know if the messages sent counts cancelled games? Because it increases during games, not just when they finish, doesn't it?

linear prediction: 37.57
actual: >60

durr hurr gunboat

48.5!!! =D Hahaha, I don't play gunboats (mainly cuz I'm bad at them.)

Also since I suck at math can someone rephrase what TruNinja said, or explain how I'd find my linear prediction lol

wow, i have played some gunboats, but my ratio is 220 msg per game.. and my win/loss ratio is something... i don't know...

@ Thucy: All you have to do is take the 'total games' in your profile, subtract any games that aren't gunboat and divide your total messages by the number of non-gunboat games. This is your raw messages per game value (the x-value in the formulas).

After this, you can take time and subtract gunboat stats from your wins, draws, survivals and elims and recalculate the percentages and add the new win/draw percents together which gives you the ACTUAL percent. If you take the formulas above, plugging in your x-value, you'll get a number that's assumedly similar to your win/draw percent.

I'll take time to calculate them for you, though and post the results in a min.

Thucy:

For the linear fit, your *predicted* win/draw percent is 48.7
Your *actual* win/draw percent is 47

Therefore, your results are pretty comparable to what is expected.

@RD: not sure what you mean by "actual data". All of the data that I obtained was actual data taken from 100 sample profiles that didn't have any gunboat games or variants, simply the classic version and it was a simple random sample to incorporate a wide variety of msg/gm and win/draw pcts.

If by "actual" data, you mean the data collected by the site programs, then no. But all that is is simply a larger sample size and is by no means the population parameter, which is simply what the 100 that I've tabulated are.

I've thought about increasing the sample size to see how much the trends hold true over time, because I first calculated the formulas for a sample size of 60 people and the linear fit was a bit stronger (r = 0.76) and it was roughly equal to the logarithmic fit, but as the sample size increased, the correlation coefficient (r-value) decreased while the logarithmic fit shuffled around a bit but remained pretty steady. I hold the view that given a large enough sample size, the trends would show that it's less and less linear and more logarithmic.

I also surmise that, while there's a strong positive correlation between messages per game and win/draw percentage, that the r-value will still hang around the 0.7 - 0.8 mark because there are other factors that go into a person's win/draw percentage.

I have seven gunboats, and two 1901 draws (that were cancels/restarts before we had the cancel feature), so when I subtract those nine games from my total I'm at 317 messages per game, take that djbent/TruNinja! :-P

djbent is closer to about 400 messages per game which is both a testament to her dedication to the art of diplomacy as a symptom of mental illness.

typo... that should read... as much a testament to her dedication to diplomacy as a symptom of mental illness
Where is djbent? She has been quite lately.

Why do you assume a linear or logarithmic regression?

Could it not be exponential? Polynomial? Sinusoidal?

Or even normal distribution?

First time i read the thread title i read "Diplomacy Jokes" and immediately thought of this old gem..

http://webdiplomacy.net/forum.php?threadID=498819&page-thread=1#threadPager

It's only....a couple years old.....

@MM: wow, that's quite a bit of talking. It's also about three times my own :) I guess you're outside the bounds of the function. Think your amount of press per game has increased over the years?

@Gob: I assume linear and logarithmic because the r-values indicate a strong positive correlation. Exponential is essentially the same as logarithmic because the natural log function is simply the inverse of the exponential function, and so the data points technically are exponential but simply an exponent of a power less than 1.

It can't possibly be exponential of degree 1 or higher (which is what your polynomial idea is) because it's not at all supported by the data. Also that would suggest that if someone sends at least 100 messages per game then their win/draw percentage should be at least 100 if not higher. Last, it's not any trigonometric function because, again, that's not supported by the scatterplot since it's not cyclic.

Statistics have a value called the r-value (called the correlation coefficient) and the r^2 value (coefficient of determination) that are used to explain best-fit lines. An r-value (which ranges from -1 to 1) greater than 0.7 or less than -0.7 (|r| > 0.7) indicates a strong positive correlation to the given function. The r^2 value is the measure of what percentage of points that can be explained by the function.

@Gobbledigook: exponential? polynomial? sinusidal? suicidal perhaps even? I'm definitely proposing the gamma function, which seems to have the proper boundary conditions: no negative amount of messages is possible, an infinite number of messages ensures defeat, and a beautiful asymmetric peak somewhere in the middle. I'm sure the r² value of that, when rounded up to the nearest integer, would most appropriately be: +1.

gamma distribution. sorry. And gobbledydook of course. Again sorry.

Summary of the thread:
http://memegenerator.net/instance/10796757

I got over 300 per game an I mostly play gunboat. But i suck. Theory disproved

hehe what a great thread that was

@Santa, not sure how you're saying that the fact that you
a) have over 300 messages per game
b) play mostly gunboat
c) suck
disproves something that's not a theory...