You say that so dismissively. Do you have a cite that Vegas machines don’t attempt to use random numbers?
It’s not realistic to game a non-random system, either, because it’s impossible to observe a machine long enough to decide whether it’s likely to be “hot” or not. I have actually seen security escort a man off the property who was trying to do this very thing. Even if they hadn’t nailed him, it’s still not practical to observe a machine over enough time to determine whether it’s more or less likely to hit a jackpot.
Another way that a non-random system could be impossible to “game” would be to link two or more visually unrelated machines. I don’t know if casinos do this, but it seems easy to have two or more machines programmed and plugged in to the same (non-random) number generator that generates a particular payout for all of them as a group. The entire casino could be linked to a system like this, with a different number generator for each advertised payout. For example, if all of the 99% payout machines were linked to a single generator, then an observer trying to beat the system would have to observe ALL of the 99% payout machines throughout the whole casino at once, for an impractical amount of time.
Back when machines took and gave out coins, they would sound a call attender buzzer and light if a jackpot too big was won, or if it ran out of money. I’ve seen this, so I doubt very much the state of the coin slots affected anything.
I’ve heard that some machines are set to give bigger average payouts (in high traffic areas, for instance) but I don’t have a cite - except that my mother believed this. 
The casino could go through all that trouble, but they don’t, because it’s tons easier to achieve the desired (long-term average) payout with random numbers, and random numbers make the system provably non-gamable (not just difficult-to-game).
I think we’re probably each talking about different kinds of machines - the ones I (briefly) worked on just wouldn’t generate any wins if the coin tubes were empty, and would not permit a win of magnitude sufficient to more than deplete the coin tubes.
It seems to me if you were at this wheel, the last number to bet on would be 30. You don’t know the algorithm, but you know that based on a large number of trials, 30 is coming up less than expected. Based on the results you’ve seen, I would extrapolate out that the algorithm is potentially flawed and that 30 is likely to continue being avoided. Or at least I would choose that hypothesis over the idea that the flaw leads to 30 being “due”.
While I love the above discussions, I think there’s a simpler answer:
Even if the software has a memory and “corrects” over the long term, the Gambler’s Fallacy still applies, because your sample is inadequate. You don’t know if the machine was hitting 30s heavily last month – or the hours before you arrived. Your “pattern” could be a deviation OR a correction without you knowing.
Put another way: self-correcting or not, the long term sequence should be equally valid forwards or in reverse, unless the self-correction is over a very brief period (1000 spins is very short term for a wheel with 38 pockets, but it’s a full 24 hrs of roulette play, allowing 90 secs for betting, spin and payouts – plenty of time to go broke even with an ‘edge’)
Maybe it’s really short on 40s over the past year, and over quota on 30s. You’re betting 30s and thinking that betting on the visible excess of 40s in the current spins is a fool’s quest, because you know the software remembers and corrects, but you may actually have it exactly backwards. Certainly you’d have been better off betting on 40s during that run, so who’s the fool?
An inadequate sample will as happily mislead as reveal. That’s what makes it gambling
Actually having a chance to pay of programmed percentages is exactly how they work.
otherwise it would be illegal
Anything other than a single random number at that moment would violate the "dependancy upon a previous game outcome " section.
Nope. It would be 90% if all the numbers 0 through 9 were to be chosen an equal number of times across a given section of time, but if it’s truly random, the odds of each number being picked exactly 10% of the time approaches 0%.
Now, theoretically you could have machines that could have payouts hand-tuned by the management, so that every machine would use a good pseudorandom number generator. Then, an individual machine wouldn’t adjust how its paying out based upon how well it’s doing. Instead, management could keep tabs on how well their fleet as a whole was paying out, and tweak the whole fleet based upon the payouts to date. Then you couldn’t watch a given machine and know that it was due for a big payout.
Sure, the odds of each number being picked exactly 10% of the time approaches 0 as the number of plays increases (for starters, it fails to be true whenever the number of plays is not divisible by 10…). But, for any epsilon, the odds of each number being picked between 10-epsilon% and 10+epsilon% of the time approaches 1. That is to say, every number’s frequency approaches certainty of being within arbitrarily tight tolerance of 10%. It seems perfectly fair to describe this in ordinary language as the machine being engineered to pretty much guarantee 10% frequency for each number.
[QUOTE=Punoqllads]
Nope. It would be 90% if all the numbers 0 through 9 were to be chosen an equal number of times across a given section of time, but if it’s truly random, the odds of each number being picked exactly 10% of the time approaches 0%./QUOTE]
Isn’t that exactly the opposite of how random numbers work? Isn’t it that if the generator is truly random, given enough time the distribution will approach 10% for each number?
I mean…if you flip a coin (a random generator with 2 values) – after a million flips you may not get exactly 500,000 heads, but it’ll be close enough to not really care.
Yeah, you’ll very probably not get exactly 500,000 heads, but you’ll also very probably be very close to 50% heads. Specifically, the expected absolute difference between the number of heads and number of tails grows on the order of sqrt(N), where N is the number of trials. Indeed, as time goes on, you’ll expect to see a greater and greater difference between the number of heads and number of tails. However, it follows that, in terms of the proportions, the expected absolute difference between the proportion of heads and the proportion of tails is on the order of 1/sqrt(N), which of course approaches 0 as N gets large.