also, there are plenty of colorful charts and graphs to illustrate everything :)

Yaaaaaay! I've been wanting to write more about strategies for each country, but as nice as the stats on old articles are, they're... well... old. I'm really interested to see this.

There's one thing I'd like to see included that I think isn't too difficult that I also didn't see listed (though I might have just missed it) -- how frequent certain openings are. Since you have the success rate of various common opening moves, the data is already there, and knowing how likely it is that Austria goes to Galicia, for example, would be very useful in planning opening tactics.

I can list them in order, but I looked at the first 100 games using the various openings for each country. Galicia is tied for the most common with moving to Trieste.

I'll be emailing you for a copy, but in opinion it's who plays and not the country that matters. New players vs experienced players. How about a break down based on pot sizes? Ex. New players hate Austria, I don't mind playing Austria as I know A/I have to work together or they are both dead. In probably 90% of the games. A game where everyone understands the Eco system that is the map & how 1 affects he other, this is what makes a good game. I bet most are not played with this understanding. My view on diplomacy, I could be wrong. Stats will not show this & I'm sure a France or Germany will look to be the best.

this sounds awesome. i live for stats like this.

+1 omgwhathappened

kinda disagree with IKE, but I'll wait for the stats before starting an argument about this. :P

Statistically, your statement is rather inaccurate. It is assumed that since each game is independent of the others and that since there is a simple random sample taken of games, you'll see games that contain experienced vs experienced, experienced vs inexperienced and inexperienced vs inexperienced types of games. This being the fact, the statistics will apply (in general) regardless of who is playing. I have stated that there is some minor difference for some countries but the overall information applies regardless of who is playing.

If you begin to separate games between experienced and inexperienced players then you get a skewed version of what really goes on with the game.

To cite an example: if a group of inexperienced players are playing, they are more likely to do strange things but given that it's easier to win with certain countries than others, the trend will remain that x country wins y percent of games.

In the report, I'm examining over 5,000 games and the odds that skewing is occurring based on who is playing the game is reduced to essentially nothing. Where there is a discrepancy between high caliber games and low caliber games, it has been noted,even still, that discrepancy is rather small.

^^ And even then, how would you define an experienced player? Just because someones been *here* a long time doesn't mean they've been playing longer than their opponent, who may be new to here but experienced in the real world.

That said, whilst I agree with your statement, it only holds if we assume new-new and old-old games will have similar properties. It could be that a game between new players is heavily weighted towards France and biased against Austria, with these reversed in the old-old stats, thus in the overall figures you'd see nothing.

Ninja - its easy enough to test IKE's theory. if assuming you measure the game based on just points (though GR would be much more accurate) you could have a mean for all games, then do stats for example on games that are 707pts+ also for games from 35-350 pts for example... then see if either of those two smaller samples have a wide divergence from your mean or from each other.

you dont have to do it for all the questions, but a representative enough sample to get an answer to IKEs theory. its extra work, so I'm not suggesting it, but just saying it CAN be done (assuming that our sample size of ~24,000 finished games is enough.

although, to counter IKE's theory... keep in mind that League games, Masters games, and other tournaments which usually have much better players are almost all in the 35 point range. so using points would have way too much noise to make the stats worth doing I think. GR really is the only option IF that data is even easily available.

two important things though..

1) I presume that you are separating out WTA vs PPSC vs GB vs PP games right? how about anon? that is less relevant but might also produce differences in statistical outcomes none of us expect.

2) I am hoping you've separated out all the games created before the important updates... as in, you are not including the early era games where draw was not an option etc.

I would think anon/non-anon wouldn't matter, because the analyses are all about the states whereas anon/non-anon is all about the players. And I'd imagine the varying levels of press have been accounted for, because he explicitly mentioned message ratio (which would require some looking at the message setting, naturally). But in any case those two should definitely be sorted.

And wait, there used not to be a draw option? That really sucks...

also, unsure about IKE's example, but that's for another topic

yeah, anon/public does matter. i play different in public games because i can check that person's history and track record. if i look back and see that a player tends to open a certain way as a particular country, i will weight my moves in favor of it happening again. i will also try and get folks to gang up on those with the most experience/win %, while i tend to cherry pick the serial losers as my own early stabs.

can't do that in private games.

ah, i didn't finish reading your post eden. we are basically saying the same thing 'public is about the players', which is really the point that i elaborated.

Well, right. I too tend to steer people toward attacking the better-record players, set up serial losers as stabs, and disregard the above if I find I have a high-CD player as a neighbor (-> gang up on him -> he quits). I imagine most do.

That said, these don't affect the STATES involved at all, because states are randomly assigned to players. We're only discussing the motivations behind the initial policies. That is a completely different animal from the reasons behind a state's success.

And between random assignment AND a large pool of games observed, the effect from this cannot be significant because (1) not all games are non-anon and (2) not everyone does those policies [gang up on CD/good players, stab weak players later] even in non-anon games; in fact, I think the trend I've observed is that it's more likely for someone *not* to play by the players even in non-anon.

So ultimately, yes, public is about the players. Which is why there's no need to acknowledge it separately in the analysis: the analysis is about the states, not the players.

please tell me you wrote a script to parse the data and are not doing it manually!

Please tell me you asked for the database dump rather than data-mining the site?

It'd be too easy for kestas to see a data-mine eh? But the database dump would be much easier anyway!

@ omg and PE... and that is exactly why I try never to play non-anon games anymore =P it kidna sux to have to be stuck in the anon box as I'd like to build relationships with folks... but its too high a price to pay

I don't often end up getting the gang train in non-anon, though I'd like to think I have one of the better records on the site. I prefer non-anon because it seems like I cannot find an anon game without at least one cheating little bastard somewhere...

There are fewer cheaters than you think, especially at the level you should be playing.

I play a lot of live games, though. Not only not-guaranteed-to-be-high-quality, but highest concentration of cheaters.

@ Babak: To begin with, to fully answer IKE's question a little more clearly, you really don't want to separate out high calibur games because then you dont' get the full scope of the game. Anytime you sample a specific demographic you will indeed change the results (to some degree). Rather, you want an accurate reflection of the game as a whole. To cite two examples:

A) Let's assume that you wanted to test the health of the average American and you find that 2% of all Americans have heart-related problems. This accurately reflects the adult population (the number is merely a proposition and not intended to be accurate). If you instead test all adults of retirement age and higher, this number is bound to go up because you change the demographic.

B) for the game of Diplomacy, here is how the numbers would change: Some countries would indeed have slightly higher win percentages (Italy and Austria for example) and all of them would have higher draw percentages because most high-caliber games end in draws and the average draw type is no longer a 3-way as it is in the game under normal conditions. If you want the raw data for the average player in an average game of Diplomacy (which is what the report is a reflection of), then you need to account for ALL games.

To answer the other questions: I do not list Gunboat or Public Messaging games, I do account for anonymous games, PPSC and WTA because they are a part of the natural game. I do not count games where the server was in its early stages (back when you could end a game before a player reached the 18 SC's needed to win). I do count live games or other "classic" games so long as no one went into CD within the first full year.

All of the controls I have accounted for are listed in the introduction to the paper itself

@Geofram and Figle: To get my data, I filtered through the archives by the type I wanted, manually opened each and every game for the first 5-6,000, checked the first and second season of each and carefully recorded data the old-fashioned way. Using any type of program would pick games that I wanted to bar from my list. I don't want inappropriate outliers in my data since many of the statistics rely on normal data and the presence of outliers can corrupt the statistic. No offense to anyone who is offended by this (?) but it more accurately reflects the nature of the game.

"manually opened each game"
ouch.

Yes. It took a long time but for the results, it was way worth it. To be honest, this endeavor has spanned about a year in the making to get everything going into this.

I've actually never opened the sql file even when I had Java and C++ on my laptop.

A year? Oh my god ninja. Some of the ways you're looking at the data seem a bit complicated, but this could of been a week with the database dump.

I'm not belittling your efforts because I can't wait to see the results, but a year?!

I am very interested to see what the stats end up being and have emailed you asking for a copy. Do you mind my asking, what specific diplomacy games have you included in your analysis? Are they games from this site? If so, I'm assuming you take steps to exclude games from analysis in which a country went into civil disorder, especially early in the game.

Ill make sure you get a copy :)

Yes, I have a lot of controls going into the game selection (hence the manual collection of games as opposed to allowing a program to tally stats) and CD's in the first year are one of them. All games observed came from this site.

@Geofram: Yes, a year on and off (during semesters). I didn't have the capability to collect data with a program that has the ability to filter games the way I needed it to. If you can do it, feel free to do so and email me a copy of the numbers in a readable format and I'll perform the stats on em. Statistically, however, all you need are sample sizes of 30 or greater for ease of the formulas.

Ninja - thank you for the explanations.

but I guess what IKE was getting at (and I agree) is that including games with absolute newbies are not reflective of "regular" games... another words, those ARE outliers that should be excluded. ;) but I am very much looking forward to receiving the paper and reading it. can't wait. congrats on the paper, and props on the hard work you've done for this endeavor.