Maryland 16-1, 6-1
My Maryland team is still the #1 RPI team in the country after this loss, but I don't think it's possible for a team to have this big of a let-down. A loss I could definitely understand. A horrific blowout like this just seems very incongruous.

maybe the 16 wins were flukes.

Surprising blowouts do happen in real life too. To pick a different sport look at the Stanford fotball team recent utter destruction of #5 ranked USC at home. If that had happened in a SIM the USC coach would be screaming bloody murder.

They do happen, but not at anywhere near the rate they do in RL. And RL includes factors that contribute heavily to such unexpected outcomes. In the case of the Stanford-USC game, the bad blood between the 2 coaches/programs was a factor. There is also momentum, team chemistry, player personal issues, etc. None of these exist in HD. In other words, the teams are pure mathematical machines, with no inconsistencies other than those produced by the RNG. In a large set of trials, outcomes will tend to cluster near an average. This does not happen in HD. Try flipping a fair coin 100 times. Then do this (100 flips) 1,000 times, or 10,000 times. This can be simulated in excel. You'll see that unexpected outcomes are very rare, when the environment is mathematically machine-like.

jskenner, i think what you claim is true in theory. i'm not sure what the actual implementation is, but there obviously is a pretty big random factor. how do they do it? it seems you would look at every situation where multiple outcomes were possible, weight them based on ratings and settings, and roll the dice. but i agree, that alone doesn't really explain what we see. there are too many dice rolls, like you explained.

so, how do the outcomes get so severe? well, i think there have to be random factors generated for games, to create a more random environment. for example, the weight of HCA in each of the outcomes might have a starting value, like 10%. but, maybe at the start of the game, a random factor is applied to it, to allow it to jump to say 30%, or -5%?

also, i have went back and forth on the idea of season-long player random factors with a small random walk after each game. some guys just play so well some seasons, and some, so ******. for a 30 game season, it seems too long for some of the seasons our guys have (including the senior slumps). but, i am pretty skeptical WIS would do that, although, it seems to mimics reality better than not. there could also be game-long player random factors that are created to cause more random outcomes on the games.

also, consider how you might implement different offenses and defenses. it probably causes different equations to be used in a bunch of places. some might be situation specific (like when 2 players go for a rebound after a missed shot), or some could be player specific. anyway, i'm not sure you would actually go and make sets of equations for all the combinations, but one element of the equation might be, for example, on a steal formula, chance of a steal = ... + .1 * (defense is press) + .04 * (defense is man) + .02 * (defense is zone). i like this place as a place to insert random factors, such as, what factor in the range [.3-3] will we multiply all offense/defense weights by? or maybe have a range for rebounds, one for making shots, and one for everything else.

anyway, i really have no idea how WIS actually does it. but i can't imagine there is no random factor like one of the ones described above. not only from the outcomes, but from a design standpoint. if you were making a sim like this, to keep the users guessing, you would basically want small random factors everywhere - otherwise, it would be too deterministic, like the excel formulas you were talking about. people would figure it out too much, instead of building a "feel". from experience, i really feel WIS has them all over, most everything i've looked at in detail seems to have more variance than what you would expect without the random factors.

also, i wouldn't blame WIS for denying there is no "upset factor". i wouldn't even call them wrong. it depends how you look at it. it seems you could put all the equations used in one game together, into one big equation for the entire game, and factor out all the randomness into an upset factor, and call it an upset factor. but, i don't think WIS does it like that. they want a complicated upset factor that effects many aspects of the box score. so, its more likely there is a random factor to make a team rebound better/worse all game (or half), and a random factor to make teams shoot better/worse all game, etc. type of thing. if you were aware of those random factors, i really don't think you would call them upset factors. i don't think you would agree an upset factor existed, either. but, i would consider the two cases mathematically analogous, although the upset factor would likely be extremely complicated.

billy,

I like a lot of your considerations here. Re: your idea on a complex set of factors, I'm not sure how I see that, as the more factors one adds, the more any such "upset/blowout" swing would be diluted toward the middle. In other words, it would be highly unlikely that a team would get the short end of most of the multiple rebounding/shooting/ball handling/passing/fouling/etc. factors, which could reasonably explain such a result as colorblind's example, which seems to occur on a much greater rate than expected. What do you think?

And I agree, I have NO problem with HD doing something like this to simulate such variability. I can't think of a better way to do it, unless you introduced RL factors such as momentum, player personal issues, rivalries, team chemistry, etc. Now that would be better, but very difficult to design/manage. I think WIS has an exceedingly tough job to produce ongoing reasonable results, and they do a pretty darn good job of it, even if we're not always happy about some of the engine, etc.

Interesting to hear both js and billy endorse such an idea, I obviously respect both of your opinions. JS have talked about this issue on our CC and sitemail a bit. I don't think there is an "extra" random factor involved.

We're talking about an engine that routinely produces stud big men that shoot 40% for an entire season. Why is it so hard to believe that the same, skewed engine would produce a bizarre result like the one above completely on its own? To me it seems logical that would be the case.

daalter, think of it this way. Let's say a 99 LP C (with high ATH, etc) SHOULD have a 60% chance of hitting a shot (on average, with an engine that works like we want) and the 80 LP guy has a 45% chance. IF that's the case, then over a season's worth of attempts, the better C would almost always shoot better, even though both Cs would have some variability. Sometimes the better C would shoot 53% and sometimes the lesser would shoot 53%, but for the most part, the better C would shoot significantly better over time. I contend that the HD engine problems puts too little weight (or in some other way mismanages such ratings) so that the better player actually has more like a 52% and the lesser has a 48% chance. Something like that. In such a case, there will be a much greater chance that the lesser will outperform the better player. The better player will, on average, tend to outperform, but it will be less significant, and he will more often underperform the lesser (than he otherwise would, IF the engine treated the ratings more properly).

What I and billy are talking about is the fact that the better 52% player has much greater variability, so that a 90% confidence interval is something like 38%-58% instead of what it would be without our theorized "extra" game randomization factors, more like 47%-54%. And this affects all aspects of game play (at a significant level in about say, 5 games a year), so that we get colorblind's example above. In short, the fact that the engine is off does not explain such regularly occurring wild swings. I'm not doing a great job of espousing my views, and I don't mean to take a stand on this. But I just think there is something extra that would need to be included that would explain such VARIABILITY.

Quote: Originally posted by jskenner on 12/01/2009billy,

I like a lot of your considerations here. Re: your idea on a complex set of factors, I'm not sure how I see that, as the more factors one adds, the more any such "upset/blowout" swing would be diluted toward the middle. In other words, it would be highly unlikely that a team would get the short end of most of the multiple rebounding/shooting/ball handling/passing/fouling/etc. factors, which could reasonably explain such a result as colorblind's example, which seems to occur on a much greater rate than expected. What do you think?

And I agree, I have NO problem with HD doing something like this to simulate such variability. I can't think of a better way to do it, unless you introduced RL factors such as momentum, player personal issues, rivalries, team chemistry, etc. Now that would be better, but very difficult to design/manage. I think WIS has an exceedingly tough job to produce ongoing reasonable results, and they do a pretty darn good job of it, even if we're not always happy about some of the engine, etc.

you are right, the more factors, the more it would swing to the middle. that produces a desired effect of limiting extreme outlier cases. it would still add more volatility than not having random factors, but the extremely unusual cases will still be pretty unusual. you mentioned an upset factor, 5 or 10 times a season, just going off and causing a crazy game. that could still reasonably happen when a number of random factors line up on one side of the ball. i figure there have to be a limited number of such factors, where each is used many, many times throughout the game. like one for shooting (could even be a global that effects both teams equally - sometimes it seems like if one team can't make a shot, the other team tends to struggle as well. i know when my best team's offense starts to suck it up, my defense regularly seems way ahead of the curve, and that is the only thing i can think of to explain it). maybe 3-10 make up 80% or more of the random from random factors. i think that would be few enough to explain cases like colorblinds.

also, cb lost on the road. of all the random factors that could exist, the one i feel most strongly it does exist, is one on HCA's impact. way too many times, home court seems to cause a 40 point difference between 2 games in conference play. WAY too many. with so many randoms, even the randoms outside random factors (if they exist), its tough to ever say for sure off "feel". i bet a stat analysis of a thousand games could prove it one way or the other, though, if a good subset was on the neutral court.

the things you mention at the end, momentum, rivalries, etc. are all interesting concepts. it would be really tough to make them realistically. i also agree it could be better, but it would be damn hard. i wonder, along those lines, if an attempt was made with a season-long randomly modified player constant, that effects their overall performance. that kind of thing could capture a player having off-the-court issues, or being really focuses/dedicated, which theoretically would have to be random or largely random in the first place. i also wonder, with all the complaints about the senior slump, if that is taken as the result of the experiment, marking it a failure, and will cause WIS to avoid attempting to simulate momentum and such down the line. most likely, its all in my head, but you never know. i change distro based on actual production all the time, and it appears to help me, but there are a number of other things that could explain why that seems to work.

To put it in fair-coin terms, take a fair coin, with a 50% chance of heads. Flip it 100 times, and you expect that in 96.5% of times of 100 flips, you'd get between 40 and 60 heads. Now, change the coin so that you have a 45% chance of heads. That would shift the 96.5% confidence interval downward to something like 35-55. However, the VARIABILITY is unchanged.

I am contending, that in relation to such an example, WIS introduces a factor that increases the variability such that a fair coin will now produce 40-60 heads in a significantly less percent of trials, say 85%. But many of us think the coin is not quite fair (engine issues), and this factor increases THAT variability as well, so that 35-55 heads happens much less than 96.5% of the time, and you get unusually variant outcomes such as 23 and 68, much more than one would expect.

Great points, billyg. Re: the HCA factor you mention, I believe this DOES account for some of what we're discussing. But I also see results, on a reasonably regular basis, where the ROAD team gets the good end of such an odd result. It may be that in some cases, HD uses HCA to put a negative weight for the home team, leading to such unexplainable results for the visiting team. I think you'd mentioned that before, and I think this is reasonable.

Quote: Originally posted by dalter on 12/01/2009Interesting to hear both js and billy endorse such an idea, I obviously respect both of your opinions. JS have talked about this issue on our CC and sitemail a bit. I don't think there is an "extra" random factor involved.We're talking about an engine that routinely produces stud big men that shoot 40% for an entire season. Why is it so hard to believe that the same, skewed engine would produce a bizarre result like the one above completely on its own? To me it seems logical that would be the case.

i could definitely see it producing it. once though, or not very often, but the overall experience does not add up, IMO, without "additional random factors".

for starters, consider recruit rankings. i am pretty convinced there is an additional random factor on top of ratings, that goes into those rankings. i seriously cannot fathom any continuous formula that would explain it otherwise, and i've looked at top ranked players for pretty long, trying to figure out if it could possibly be continuous. if you agree, there is a random factor, then i take this as proof on concept - in the game design, it is an example of WIS using a random factor when it is not necessary for function, only, to throw the user off and increase volatility. i submit that the existence of one such use greatly increases the possibility of the existence of others.

also, at the end, when you talk about bigs shooting 40%, i don't see how this weighs on the possibly existence of random factors. if you were to graph big man shooting % compensating for opponents and team mates, on a long term scale, i would imagine a small amount of volatility, i.e. a really sharp bell curve. i think the center is off, big men shoot like crap. but, i don't see this as having any impact on if random factors are used in the engine simulation. so, when you say if the engine can product 40% shooting bigs, couldn't it make this crazy result? sure, but there is nothing suggesting the engine producing 40% shooting bigs was not using random factors.