Sivodsi

Just watched the new video. Wow, those shootouts across the river were something. A question: what was the tiger doing in the river? Looked motionless at an angle for a long time, and then we saw it trucking on through. Was that a glitch?

For the RoW data you should really use this free software: Minifac rather than winsteps because it requires more than two levels of data (scenario, player, side). Edit to add: besides, if you do it in minifac first, it has a function that automatically produces a winsteps control file. The free version has all the features of the full version (which I have) but is limited to 2000 data points, and I'm pretty sure that it is sufficient to run RoW. If you're interested I'm happy to send you the control file I did for the last RoW.

Actually you can, and it forms one of the principles that underlies Rasch. Rasch works from probability, working from expected and unexpected results. If you get a string of good results from someone, then it is reasonable to assume that this is because the person has a high degree of skill. However, if one person gets a string of low results, and then one high one, then is it not reasonable to assume that this was a 'lucky' result? The person who gets the lucky result is still credited with it, but Rasch will signal it by calling it a 'misfitting' result, because it is outside expectation. As I said before, I am no mathematician, but I do understand how this extremely useful method of statistical analysis works.

As it was on the British side of operations, that means we can get the jagdpanther with the commonwealth module, right?

Well this is what Rasch statistics does. You enter the various facets of the data: in CMBN's case that would be: scenario, side, player score. You centre the facets that you are not interested in on 0. This lets the facet that you are interested in float free, in this case 'player'. In effect you are asking, "Given the side and the scenario being played, what is this players ability?" Thus, all players abilities are put on the same scale, irrespective of what side is being played, or the difficulty of the scenario. Yes, the math is complex, and not being a mathematician, I can't explain how it works in more than general terms. I am merely a user of the software, 'facets' and I use it as part of my job and the research that I do into my PhD, which is in the field of language assessment. Fortunately, all the math goes on under the hood of the software. What I can do, and have done, is encode the data, run it in the software and interpret the results. Yes, it is a card game, where each 'scenario' is based on a selection of cards that should be random. I'm not sure whether the human made scenarios in CMBN could be considered exactly analogous to that.

Thanks for the clear explanation. Hmm yes, and how do you do the fiddling, eh? That brings you back to the same problem inherent in the pools system: Who plays who with what side? The main problem is that this system doesn't take into account the skill involved in getting a certain score in a certain scenario. A person can show greater skill in losing an unbalanced scenario by a narrow margin than a person who pulls off a comfortable win when all the odds are in his favor. In this system it looks like all scenarios need to be of approximately equal difficulty. It seems also, that you must keep axis players separate from allied players. Still, it has the advantage of being an easy system to understand and put into action.

Sounds interesting. I'm unfamiliar with Bridge scoring, but a quick look at the wikipedia page leads me to believe that its not something I would pick up quickly! How would you apply this to a 72 person tournament? The basic problem as I see it, is that if you divide people into pools it is very difficult (impossible) to make sure that each group is 'fairly' composed. That is, even if you have 4 extremely skilled players out of 6 in the pool, only two will go on to the finals. Yet, another pool may have only 1 who is skillful enough to beat all the other members, but nowhere near as good as the 4 in the other group. Yet this person would make it into the finals while the two in the other group who are actually better than him would not. You might be able to get around the problem by seeding somewhat, but the problem is that so many people's CM skills are unknown relative to each other. So you see, in this kind of setup, it becomes more a matter of luck whether you make it into the next round or not. The proposed Rasch format at least beats the pool system in this respect. So, can your Bridge scoring system overcome such challenges?

It'll be interesting to get their take on it. Rasch is relatively recent development, and they might not know much about Rasch. I don't think many statistics courses at university level include anything on it. As I said before, Rasch is diametrically opposite in its approach to classical statistics. Where the statistician would be asking "how does our model best fit the data", Rasch looks at the data and says "how well does this data fit the model".

Not likely to find Rasch mentioned in a statistics book! Try this. Okay, I'll catch you at BoB in few months - when I have time to be a member again.

I'm not a statistics wallah either! In fact, Rasch works in a fundamentally different way to classical statistical methodology, but I won't bore you with the details... How useful the statistics at WeBoB are depend on how inter-related they are. If enough people have played enough different scenarios you could create a couple of different ladders, one showing how good the players are, the other showing how difficult the scenarios are. For a trial run for RoW though, I'd rather do it as a trial tournament, because its much clearer what the figures mean, and its better as a learning experience. - hey you were my sponsor at Bob, remember? You've sponsored so many that it might have slipped your mind, eh? I'd love to rejoin when CMBN comes out.

It sounds like the movie represents the book well. Try 'War and Peace', now that is a book to read before CM comes out.

Thanks for posting the figures, WineCape. If you use the same numbers, one good way of running such a tournament to get solid measures of ability for Rasch would be this: Some points: 1) This is ideal, because everybody is linked (eg player 72 is linked to player 1, because he played 32, who played 68 who played 28 etc etc who played 1). 2) Rasch would still work under the previous RoW's small group setup because the players are linked through playing the same five scenarios, but this way gives more certain measures of players ability compared to each other. 3) Using Rasch you are not limited to a set number of players or scenarios. You could have 100 players playing 4 scenarios if the demand was there. 4) It is robust to having some players drop out. Rasch just uses the data that it has, so no need to worry about finding replacements. 5) From this tournament, you would get a measure of all player's abilities on the same scale. This measure is not a mere ranking as you would get by using means and medians etc: the difference in scores would represent the difference in ability. So if person A gets 87.6, person B gets 77.6 and C gets 76.7, this means that person B and C are very close in ability, but person A is quite a bit better (such statements cannot be made using raw scores, where the only thing you can say is that 'A is best, followed by B then C'). Rasch takes all the data and uses probability to create an interval scale. 6) This means that you wouldn't necessarily need a knockout round. However, you could still have one if you use Rasch to select say, the top 16 players for a mirrored knockout contest or something like that. (edit to add) 7) Oh yes, and it bears repeating that Scenario balance does not matter in a Rasch tournament because you can set level of difficulty of the scenario at '0' and let the players scores float: so maybe the average score for the Germans in one scenario is 20, in such a scenario a score of 50 for Germans would be brilliant. Rasch will recognize the skill inherent in such a result and adjust that players measure upwards. I'm also pretty sure that if that score of 50 was against a spectacularly incompetent opponent, that Rasch would also take that into account - however, I need to get some real world data to play with before I can be certain of how well Rasch deals with this. Hmmm, that's all I can think of for now. I suppose a drawback might be that you would lose that comradeship of being in a little group of six, but there are ways around this. For example you could set up groups in which say a couple of your games are with players from other groups. Interconnectivity would not be so good as with the above setup but it would be fine for the purposes of the tournament. I'm very confident that this system would produce accurate results, but as I say, I'd like to do a trial run with it first, say 16 people with 3 scenarios or something like that.

Out of curiosity how many scenarios do you see being used in the first round?

Sure, but we can already do this by comparing the last RoW figures with the Rasch measures. PM me and I'll e-mail them to you. As I pointed out in an earlier post, having independent groups as RoW has been run in the past does not give the ideal connectivity for Rasch, so the estimates of ability are not as good as they could be. You can still run the analysis, but the measures will have a larger error associated with them. This is the table from my previous post for the best format for Rasch. Please scroll back for an explanation of it. Could you make Nabla work with a format like this? Strikes me that you're asking for trouble if you use two different scoring systems! :eek:

Indeed, a rating for a scenario is not necessary, since the difficulty of the scenario is set by the scores that people get in it. However, I believe that you could set the difficulty level if you want to - the software has many functions that I have never had a need to use. As a user I have experience with processing speaking tests and multiple choice tests, and of course, for practice I used it on the RoW scores. I can set things up, run them and interpret the results, even explain how it works in a general way, but I'm certainly not a mathematician. Not quite sure what you mean by 'fly in the ointment'. What you allow your opponent to do is the thing you are trying measure, right? When I first ran RoW figures through 'facets' the software was not calibrated to run on figures like 76 - 24 (how CMX1 delivered the results), but the writer of the software fixed the problem and promptly issued a new version with the fix. Nice guy. Anyway, CMX2 delivers its results differently, but it would not be hard to reduce them to a % so that Facets can compute them, and if there is a problem the support of the program is brilliant. As for its application to a CM tournament, I definitely recommend a smaller scale tournament to see how it goes.

The beauty of Rasch statistics is that it doesn't work by means, medians etc, it works by probability. It has a function in which you can fix the scenarios' level, and let the 'ability' of the players float. In effect you are saying, given the difficulty of playing this side of the scenario, how good was this player?

I can see your point, but here the case is different. The bit that you quoted was one particular application of Rasch. We would be approaching it from the perspective of "skill in CMBN" as being the ability we are interested in. We are not comparing the German to the US side (although we certainly could if we wanted to). When I did a Rasch course, I actually used the data from the last RoW tournament to see how it compared to the NABLA system. The results were similar, but there were also some notable differences. I'm happy to dig them out if anybody is interested. It should be pointed out, however, that the 'separate independent' groups format of RoW is not ideal for Rasch, which does better the more connected the data is. So what would a Rasch CM tournament look like? The important thing is connectiveness: as long as the players are related to each other in a chain they can be compared on the same scale of ability. The more links in the chain the better the ability measure, but below is what I consider to be a minimum workable format for 16 players (A-H and a-h) with two scenarios so that a player can be axis in one and allies in the other. We can see above that every player is related to another player: for example, ‘D’ is related to ‘a’ because ‘a’ played A who played B who played ‘c’ who played ‘C’ who played ‘d’ who he did play. Ideally there would be more scenarios so that more players could play each other, and you would get less error with better linking. You would get more connections by having four scenarios with each player playing each other twice, once as axis, once as allies. The good thing about this system is that it is robust in the case of players dropping out. I would love to run a Rasch analysed tournament just to see how it goes. But at the moment don't have much time to spare.

The answer is to use Rasch statistics.

On the other hand, even in Iron mode, if your unit detects an AT gun, they are not likely to forget it. Unfortunately, a player who is doing PBEM, may miss or forget it. Isn't it reasonable and more authentic for your units to have a 'memory' of the type of unit that they've identified? Certainly more authentic than having the 'godlike' player forget! The above should only happen in Iron mode if the unit in question could absolutely ID the enemy. A minor annoyance for me in SF is that whenever your unit spots a tank and then loses sight of it, the generic '?' is displayed. In RL your units simply would not fail to notice the difference between an infantry unit and an AFV, but you as a PBEM player may. There should be "AFV?" and "Infantry?" markers for contacts that you lose sight with.

Yeah, Christchurch is my home town. Such a shock to see it get smashed up twice in a few months. I always used to joke during my visits every two years or so how little anything changes. Well, things will be different next time I go. Hope you and your best got through it okay, Crocky.

Wooo hooo! I'm in. The last ROW was my first, but it remains my most intense gaming experience to date.

This came up in the AAR thread, and hasn't received any comment, probably because others aren't interested, possibly because its been answered before but I haven't found/noticed it. Originally Posted by ClarkWGriswold - its my understanding that covered arcs are not related to spotting in this sense. They only influence they have is by facing the observers towards a certain position, but they will spot outside the covered arc. The danger of the covered arc is if something dangerous pops up outside of the arc. Then it is up to the AI to take evasive action. This leads to a couple of questions that are vitally important to PBEM play: 1) Under what conditions, if any, will the AI of an AFV respond to spotted enemy outside the covered arc? 2) What actions can the AI take to threats outside the covered arc? - and was it confirmed whether there would be separate AFV and infantry covered arcs?

- its my understanding that covered arcs are not related to spotting in this sense. They only influence they have is by facing the observers towards a certain position, but they will spot outside the covered arc. The danger of the covered arc is if something dangerous pops up outside of the arc. Then it is up to the AI to take evasive action. This leads to a couple of questions that are vitally important to PBEM play: 1) Under what conditions will the AI respond to spotted enemy outside the covered arc? 2) What actions can the AI take to threats outside the covered arc? In CMSF I've had a T72 pump rounds into an Abrams that would not respond because the T72 was outside the covered arc - but this might have been in an earlier build, and I don't have time to run a test now.

Merciless!

Take this to its logical conclusion and you would have a 'game' with a very passive player. Turn 1: this is the only turn in which you would be guaranteed complete control. The players, looking at the troops at his disposal, the terrain and the objectives, issue orders. Rest of game: the players get some reports back, and issue new orders to groups that remain in command. The orders would be things like: "there is no enemy where you are, keep advancing until you contact" or "hold the ground, you must not let the enemy control this ridge". Everything else would be done by the AI. What is being described here is a simulation where you more or less set the parameters at the beginning, and wait and see what happens. Players take pleasure in advancing individual units to the enemy, putting tanks into hull-down positions, setting target arcs... etc etc, and this would be missing from such a 'game'. Is that what you really want? I'm sure if the demand is there, somebody would make it. But probably not as popular as a game like CMBN where you have to micromanage more than is realistic.