marksolberg
marksolberg
  • Threads: 26
  • Posts: 205
Joined: Nov 14, 2009
October 20th, 2011 at 11:42:15 AM permalink
I find this interesting. There can be small differences between the return to player (RTP) percentages between what a game manufacturer provides on a par sheet and what a gaming lab certifies for the exact same game. I find it interesting because in the old 3 reel 22 stop days this would never happen. With the advent of the new multi-line bonus slots it can occur. They can differ because the way a gaming lab calculates a bonus may be different than the way the manufacturer did. If you think about a bonus slot that awards free spins and has a locking wild feature you can see where the number of possible outcomes for the bonus grows very quickly. It would be unreasonable to "calculate this by hand" so programs are written to iterate through every possible combination of results to arrive at a RTP number. Since the manufacturer doesn't share, nor should they, their method for calculating the RTP number the gaming lab must do it independently. My best guess is that the variances may come down to things like the precision of math of the program or computer it's run on. Anyone have any other explanation?

Mark
Paigowdan
Paigowdan
  • Threads: 115
  • Posts: 5692
Joined: Apr 28, 2010
October 20th, 2011 at 12:15:57 PM permalink
When games are mathematically examined, they are usually run through a simulations of 'n' number of random hands - to produce a session of 500 million hands or so. Different runs produce slightly different results, as different hands may occur, but with so many sample hands examined, they fall within the statistical boundaries of the game's performance. You might have a 2.871150% HE in one case, and a 2.880017% HE in another sample run.
Of course, a static algebraic analysis would always produce the same baseline result, but simulation runs are informative and required.
Beware of all enterprises that require new clothes - Henry David Thoreau. Like Dealers' uniforms - Dan.
Ayecarumba
Ayecarumba
  • Threads: 236
  • Posts: 6763
Joined: Nov 17, 2009
October 20th, 2011 at 12:23:06 PM permalink
Are these simulations done on the bench with the actual chips? or is it performed virtually, using software supplied by the applicant?
Simplicity is the ultimate sophistication - Leonardo da Vinci
marksolberg
marksolberg
  • Threads: 26
  • Posts: 205
Joined: Nov 14, 2009
October 20th, 2011 at 1:05:26 PM permalink
Certainly any simulation may result in a variance between different trials. My assumption has always been that manufacturers "do the math" to come up with the actual theoretical return percentage. I'm sure they also run simulations to verify results and look at different aspects of the play of the game. I can't imagine a manufacturer relying on a simulation to generate an RTP number. Send the game into the field and find that none of their simulations happen to hit the billion to one winning combination?

I know a gaming lab does the algebraic analysis of the game to get their results. I assume, perhaps incorrectly that the manufacturer does the math also. I find it interesting that their results can vary.

Mark
Paigowdan
Paigowdan
  • Threads: 115
  • Posts: 5692
Joined: Apr 28, 2010
October 20th, 2011 at 2:19:34 PM permalink
Quote: marksolberg

Certainly any simulation may result in a variance between different trials. My assumption has always been that manufacturers "do the math" to come up with the actual theoretical return percentage. I'm sure they also run simulations to verify results and look at different aspects of the play of the game. I can't imagine a manufacturer relying on a simulation to generate an RTP number.


Well, imagine it: they do indeed do it this way. The math report I sent to Nevada Gaming and to GLI for EZ Pai Gow was required to have a simulation run example - not the flat algebraic math. The game has been out in real casinos for over two years now, and is doing fine.
Quote: marksolberg

Send the game into the field and find that none of their simulations happen to hit the billion to one winning combination?


Games enter the field and go live all over the place. Casino operators track the coin-in, coin-out, and amount kept.
For table games, the table hold is examined, and is expected to have an aggregate table hold performance in the 5% to 35% range over time, the closer to 21% or so the better.
Let's put it this way: when car manufacturers (GM, Ford, Chrysler) measure a model's mpg, they do it by driving the car over many miles with diagnostic equipment attached to the car examing the car's performance, not just algebraicly (by computing the car's weight, engine displacement, areodynamic fatures, etc.)
Quote: marksolberg

I know a gaming lab does the algebraic analysis of the game to get their results. I assume, perhaps incorrectly that the manufacturer does the math also. I find it interesting that their results can vary.


Well they do, and it's normal. So as long as it is within reasonable parameters, things are fine.
Distributors generally hire a known mathematician to perform the analysis, or have an on-staff mathematician do it.
Beware of all enterprises that require new clothes - Henry David Thoreau. Like Dealers' uniforms - Dan.
CrystalMath
CrystalMath
  • Threads: 8
  • Posts: 1909
Joined: May 10, 2011
October 20th, 2011 at 2:55:36 PM permalink
If GLI can do the math without a simulation, they will. I have a game submitted to GLI right now in which I performed a simulation to get the results. GLI has also informed me that they will do a simulation.

I also know, from working at GLI, that they are not concerned with a difference of 0.1% on a skill game since the game won't play at theoretical anyhow. For a non-skill game, I would not expect their numbers to ever be more than 0.01% different than the manufacturers math. GLI will always put their number on the certification report in the event of a difference.

There is a game I'm working on now that I have estimated will take 65 days of calculations running 2 threads to find an exact solution. Instead, I'm opting for a week of simulation.
I heart Crystal Math.
marksolberg
marksolberg
  • Threads: 26
  • Posts: 205
Joined: Nov 14, 2009
October 20th, 2011 at 7:36:18 PM permalink
Thanks for sharing this information. I wouldn't have thought it was done this way but now I know.

Mark
  • Jump to: