prozema
Joined: Oct 24, 2016
• Posts: 1194
May 8th, 2018 at 8:49:14 PM permalink
I didn't know whether to put this in a math thread or a slot thread, so I picked slots because, well... I'm no math genius, which I will make quite clear in this post. :-)

After a few cocktails, a thought occurred to me that some slot analysis is being done in a way that is harder than it needs to be... I fugured someone here could set me straight.

Specifically the analysis I'm thinking of is on desirable bonus games that bank spins.

I think I see the smart people counting key symbol frequency over a given number of trials to compute the probability of getting a desired result, in this case a pile of free spins that hopefully land at a profit. After they know how often they will trigger the bonus game, then they need to figure out if the bonus reels are indeed the same as the reels in the base game, which often they are not, and calculate the expected value for each spin in each bonus game. I understand how that analysis works... But it's a lot of data to gather and a lot of button pushing / coin in.

Wouldn't an ordinary least square regression essentially do the same thing but with less data gathering?

Let's keep it simple and say I have a single bet level and a single spin bank...

I would set profit on the X axis
I would set the number of banked spins on the Y axis.

Now that I write this, to capture the meter, you would have to gather data on the starting banked spins, the ending banked spins, and the total profit (positive or negative).

So I play twice... I end up with:

Starting Bank1: 7 spins
Ending Bank1: 18 spins
Profit: -\$80

Starting Bank2: 74 spins
Ending Bank2: 84 spins
Profit: \$22

That's enough to start... Then for each new trial, I recalculate the regression.... The more trials the better, but I'm only gathering 3 data points that are quick and easy to record.

I think I would either find a poor linear fit that seems to have a very high Y intercept / spin count (i.e. a game that never / rarely goes +EV) or a relatively good linear fit that has lower Y intercept / spin count (i.e. a game that has a vulture point).

I guess my question(s) is / are:

- Would this method work?
- How much accuracy would the analysis lose for getting a break even point vs the more traditional methods?

Thanks,
prozema
Joined: Oct 24, 2016
• Posts: 1194
May 8th, 2018 at 9:32:41 PM permalink
Just thinking out loud... The coefficient on the starting spin bank variable would matter a lot. I wonder how to deal with that...
prozema
Joined: Oct 24, 2016