@Chris: The current system does not hide players that CDed in anon games.
Once a country is taken it reveals the identety in the gamechat (with link to the players profile).

Oli, thanks.

Maybe I could add a column in the country-overview of finished games with the CD-players...

@kaner: First I will add gamecreation-options for the NMR and CDratio. But finally i would like to add these 2 (and only these 2, not anything else) to one rating. Obviously a noCD of 85% is really bad, but the 85% look much too good (think about a win-ratio of 85%).

Oli,

Thank you VERY MUCH for adding the NoNMR and NoCD statistics on the profiles !

1° May I please suggest that you rephrase the :
"0 out of 1079 phases missed"

Indeed, it does not sound clear at all.
May I please suggest instead :
"0 missed phases out of 1079"

And same with "0 out of 14 games left" that could please become :
"0 left games out of 14"

2° NO ! 85% of NoCD ratio does not look too good at all !
Please think about this example : a man's wife tells him that she has been faithful to him 85% of the nights. Do you really think that many people would find that that percentage is too good ? People are not that stupid.

And if they are, if some players think that a NoCD ratio of 85% is actually good, well that's really too bad for them but they won't understand better the question if you use a rather complicated equation or formula that is intended to help them understand.

@Retillion: Yeah. 85% does not look good to you and every expert player. But for quite a lot of people 85% look ok, because in most other context 85% is a solid rating. That's why I want to add some more general calculation that put these 85% in a more common context.

That would not remove your ability to set filters for noCD and noNMR separately.
You will be able to decide if you want to set filters based on noCD and noNMR status or on a more general RR. But this RR should directly use these 2 ratings combined. Nothing more. IMHO it should not be possible to ballance this by time or by CDtakeovers.

But it should be clear for every player that 15% CDs are really bad. Also you are limiting your possibilities by using these noCD and noNMR settings on their own (but I will not hinder you to use them), because basically you just can work with a very limited variety (between 90% and 100%).

Oli, if you think that 85% noCD ratio "looks much too good", you might turn it around into a CDratio of 15% (obviously, you can do the same with the noNMR and NMR ratios).

That would measure UNreliability rather than reliability, but might actually be more immediately useful for (very) unsophisticated players: for example, if you compare two players with CD ratios of 15% and 5%, it is immediately clear that the 15% guy is 3 times more likely to go into CD than the 5% guy; this is less immediate if with 85% vs. 95%.

We all know that CDratio and noCDratio are exactly equivalent, but the CDratio might give a more immediate "perception". The only drawback I can think of is that "lower is better" is somewhat less intuitive than "higher is better".

that makes a lot of sense to me diatarn

@diatarn: That makes it better, but not optimal.
The % for noCD and noNMR are ok. If you change this, a 1%-rating look bad, but these are the best players....

I would like a more linear spread for a combined rating. Best with 50% = average. This is a more obvious system.

And I would like to stress: This should not and will not replace these 2 separate ratings. I just want to merge these 2 in one catchy rating with 0% total fail, 50%=average, 100% top.

I guess we need to access the database to see what the mean is for CDs. Something like:
CDs recorded / number of players

Oli, what about using the rank of the player in the reliability ratings?
I mean, suppose there are n players on the site, and m of them have a worse noCD ratio than player P. Then, the reliability score of P might be simply 100*m/n
In this way, the most reliable player has a reliability score of 100, the median-reliability player has a reliability score of 50%, and terrible-reliability players are close to 0%.

If you want to account also for noNMR ratio, you might either mix the two ratios (e.g. with kaner's/mapu's formula) before calculating the rank; or calculate the score about noNMR separately, and then take an average.

Other possibility: use something like
[(number of games played) - (number of CDs)] / [(number of CDs) + constant]
which (apart from the constant, which is needed to prevent divisions by 0 for players who never went CD - btw, I suggest constant=0.5, but the choice is not unique), with some normalization (e.g., you might calculate the score above for all the active players on the site, then choose where "100%" should be).

ooops, i erased part of a sentence in the previous post:
which (apart from the constant, which is needed to prevent divisions by 0 for players who never went CD - btw, I suggest constant=0.5, but the choice is not unique), with
-->
which (apart from the constant, which is needed to prevent divisions by 0 for players who never went CD - btw, I suggest constant=0.5, but the choice is not unique) tells you how many games were finished normally for each CDed game. This should should be used with

Normalizing about all players here is not a good idea, because most have max. 1 CD and you can't set 1 CD=average=50%. I think using common sense is a better aproach...

@ diatarn_iv :

Your formula does not give 100% to the most reliable player ! It doesn't give either 50% to the median-reliability player.

Indeed, suppose there are 100 players on the site (n=100), player P is the most reliable one. As a consequence there are 99 players who have a worse noCD ratio than player P (m=99).
And so, according to your formula, player P would have a reliablity score of 100*99/100 = 99.


Also, what is the point of knowing the median-reliablity player ?
For example, let's suppose this extreme but easy to understand example :
There are 101 players, 50 players have missed 0 Phases, 1 of them has missed 1% of his played Phases and 50 players have missed 5% of their played Phases.
Then, according to your formula, the player who has missed 1% of his played Phases has a "Reliablity Rating" of 49,50%. Does that make sense ?

In other words, there is no point in knowing if a player is relatively reliable compared to the other players of this site : what we want to know is if a player is reliable.


Gentlemen, your are trying to complicate uselessly things.

When people make a statistics, it is always in order to show something : either to oneself in order to understand something or, most of the time, to show something to other people. So the question that one should ask himself when he calculates a statistics is "Is my statistic going to show to people what I want them to see ?"

Oli, if you really believe that an 85% NoCD rating looks too good, then what we need is to *educate* people. Maybe a clear notice could be written somewhere on this site saying that it is expected that players (almost) never NMR or CD.


Finally, if you still really think that people are so stupid that they do not understand such an easy concept as not NMRing or not CDing, which is actually just a matter of simple respect, and if you really think that they need some statistics that shows them how bad is their Reliablility Rating, what do you think about this most easy formula :

Reliablity about not NMRing is equal to :
NoNMR ratio ^ 2.

For example, a NoNMR ratio of 85% gives a result of 72,25%.
If you think that this result still looks too good for those who don't understand that they should (almost) never NMR, then replace 2 by 3 and the result becomes 61,41%.
If you still think that it looks too good, you can use any higher number after the ^ so that the result gives you the impression that you find correct to your taste.


But once again, wouldn't it be so much easier to *educate* players so that they *understand* that a NoNMR ratio of 85% is really terribly bad ?

NoNMR? That would be a good thing, since that means 'No No Moves Reveived', which most likely means Moves Received, right?

Yes, steephie22 : the "NoNMR ratio" is equal to the "Moves Received ratio".

@Retillion: That's exactly what I wanted to say. Comparing players with one another makes no sense.

Measuring a CD-count is easy, because it uses hard information. And I know a lot of people esp. people with a very math. background prefer this. And it's easy to value this right because you know the context. NoCD=80% (=bad), NoNMR=80% (=bad).

But if you have just one arbitrary information like reliability (without the context of CD and NMR) people have no clue how to interpret the given value and will apply already known value-distribution (In Germanys school system we use 50% = 4 = average, where 1 ist best and 6 is totally unacceptable). In preschool here kids get a grade for reliability too, and 80% would be a really good grade.

That's why I would like to set a basic non-fail value at 50%. But first I would like to gather some oppinions what to define as non-fail.

Using percentage values in the ratings is confusing the issue. A percentage implies part of a whole. Submitting 100% of your orders and completing 100% of you games is a perfect performance from a reliability standpoint. The problem begins when you attach a rating of 90% or 85% or even 80%, it looks pretty good because of the context it's taken in. The reality is that 85% reliability in Diplomacy is terribly dysfunctional. Once again the problem seems to be that no one is willing to call out the dirt bags for what they are. No one wants a cookie for turning in all their moves. We just want to be able to weed out perpetually problem players.

Also the more appropriate term for leaving a game is abandoning. Left doesn't really adequate explain the damage done by players who CD.

@Ruffhaus: +1
This is what I was trying to say... :-)

Maybe instead of a % we can agree on school grades (A-F).

Nah Oli, school grades still are subject to misinterpretation. For some people, a "B" is real good, but here we might assign a "B" to 90% reliability...which is quite terrible. Education is the answer. Inform people more about what it means, and make it blatantly obvious when people are screwing up.

@ Retillion:

First of all, i suggested the "ranking" formula, because I was trying to find something that might satisfy Oli's requests ("one catchy rating with 0% total fail, 50%=average, 100% top"). It was the only "natural" idea that came to my mind.

Unfortunately (as you rightly noticed) I was very sloppy in describing it.
As you noticed, there are several bugs in my suggestion, but I believe that all of them can be easily fixed with a bit of common sense.
- A first correction is to use 100 * m / (n-1), which is justified because you don't want to compare P to himself; this would correct your first objection.
- For even better results, use 100 * m / (n- tP), where tP is the number of players that are tied with player P (I know there exist pathological situations, e.g. when all the players have the same noCDratio.. but is that so important?); in your example, this would give 100% to the players who missed no phases/never went into CD.
- A more debatable improvement is to introduce a tie-breaker for players with 0 missed phases/0 CDs: a player with 0 CDs over 10 games should rank lower than a player with 0 CDs over 100 games. This would mean that in your example the "worst" of the top 50 players (i.e. the one with the lowest number of phases played) would rank barely above 50%. This is not good.. but is that a realistic situation?

Finally I think that (apart from satisfying Oli's "catchiness" request) knowing relative reliabilities can be useful, because you can "cross-reference" them with the absolute reliabilities in order to find what reliability constraints you can realistically put on games you create.

@drano: Thats why I would assign B to 98% for example. Why do you want to use B=90%? This is what I always say in this thread we can't use these the same percentages for the pure stats(here this works) and the term reliability (here not).

I was just throwing it out there Oli. I definitely would never consider a 90% a "B". I guess the problem with assigning "grades" is going to be that most people see things differently (again). Some of us, like Ruffhaus and myself, would consider 93% absolutely terrible in reliability. That's almost 1 out of every 10 phases missed, meaning 1 or 2 per game most likely. Others might see that as totally acceptable. So how do you determine what grade it gets? Is it a "D" which implies real bad, or a "B" which implies pretty good? That's why I feel just strict percentages give the best option. People can see the % and decide for themselves, without having to figure out what "B", "C", and "D" mean.

Yea. That's why % are hard too. Some see 90% as totally acceptable, and some absolutely terrible. But even with grades. Most people see a B grade as good, but some see absolute fail on B or worse.

@ diatarn_iv :

Having a tie-break between some players could be possible in order to rank them.
The only point of such a ranking would be some recognition from the community towards those best ranked players.

HOWEVER, a tie-break should certainly NOT be used to "calcute" the reliability of players.

For example, your stats are PERFECT :
- 0 missed Phase out of 1203 played Phases.
- 0 CD out of 21 games. By the way how come you have 26 finished games and only 21 games in your stats ?

And for example, my stats are PERFECT too :
- 0 missed Phase out of 1079 played Phases.
- 0 CD out of 14 games.

This is nothing personal against you but do you really think that you are more reliable than I am ? If, because of some "calculation", this site would describe me today as less than 100% reliable, I would be MOST disappointed !

Another example : a player with 5 CDs out of 5 games is, from my point of view, as bad as a player with 20 CDs out of 20 finished games. And I would totally disagree with a calculation that would declare that first player more "reliable" than the second one.

Also, how would you decently calculate that ratio ? I mean, what is "1 game" ? Could we really compare a 1vs1 game (that can be paused most easily) or any small variant to a 36-player game ? Could we compare a game in which the player was eliminated after a few turns to a game that has lasted for 6 months ?

----------------------------------------

Finally, I would like to point out, once again, that I absolutely agree with players like drano019 or RUFFHAUS_8 : a NoNMR ratio of 93% (for example) is absolutely terrible !
What we really need is to educate people : they should know, understand and feel that they should (almost) never NMR nor CD.

I changed the Newgame settings.
You can now set a noCD or better and a noNMR or better instead of th e"old" RR.
The RR does still work for the ongoing games, but at the moment you need to use these 2 new settings.
Maybe we can agree on a RR after some time...

Now that is fantastic ! Thank you VERY much, Oli !