Topic: Satoshi Dice -- Statistical Analysis - page 32. (Read 192889 times)

They didnt shut off betting, just payouts afaik.

I was expecting to see a dead flat spot due to yesterday's temporary hard fork.  Any idea why I don't?

Dabs it's just random probabilities. You can do it by your own.

Hi Dabs.

Sorry, I was stressed out with other business yesterday when you posted about this.

One problem with your request is that there's not really a definitive ordering on the bets.  Do we count the time the bet transaction was seen by some random host on the network?  Or the timestamp on the block it got into?  Or the depth of the block?  (Blocks aren't always in order of timestamp).  Or do we count the order that the bets were paid out?

There are lots of different ways of ordering the bets, which will change the longest sequence of losses.

Another problem, and what puts me off from looking into this, is "who cares".  Are you wanting the information to check whether the dice rolls are random?  ie. "we would expect to see a chain of 10 losses but the longest chain is 20 so SDICE is cheating"?  Because other than that I don't really see the point.  Runs of good and bad luck happen, due to the random nature of the dice rolls.  Suppose I go through the blockchain and find that once there were 21 losses on the 50% game in a row, given some arbitrary ordering of the bets.  Then what?

I guess I'll do this when I'm bored, but I don't expect it to tell us anything very useful or interesting.  Maybe I'm just being grouchy though.  That happens too. 

Thanks for the tip organofcorti - that's very kind of you. 

Hi, I understand the theoretical part (on consecutive wins or losses). What I'm looking for is actual historical data, and I think you guys already have it. (or the block chain has it.) It just needs to be sorted out. I mean, Mr. Dooglus is already posting statistics of wins vs lose. Like, for the under 48000 I can see that there have been 336630 bets of which 246989 have won and 89013 have lost. I am just curious as to how many of those losses were consecutive and what is the longest chain.

Then you can plot the theoretical formula beside the actual results to compare the "should be" and the "what actually happened already", similar to the comparison of the odds and how each 1dice has actually performed.

The first of the pair of charts actually does that - sorry if I didn't make that clear.

Got some ideas, will get something up tomorrow. I could be wrong - I might have been letting the percentile charts guide me a bit.

That's really kind from you. Respect!

What about my request? Do you think it would be more revelant to normalize with the higest month volume? It would be easier to spot the trends month to month, I think.

Thanks! Much appreciated.

It's dooglus' data and he does much more work than I do so I've passed half onto him.

What about this graph but with every "violon" as proportion to the highest month volume?

Very nice graphs.  The 2nd one reminds me of big scary needles rather than violins.


Is that true?  They increased for a few months Dec 2012 to early Feb this year, but other than that has there been an upwards trend?

A version of your first graph with the horizontal scale magnified about x10 would let us see that - just truncate the high values while the whale was playing.

That's 0.5 BTC tip my dear (I've sent the same to dooglus). Thanks a lot for your contribution!

I was thinking the same thing last night but it was getting late, so I left it until now.

The first plot below shows to volume, and the monthly volumes can be directly compared to each other (March is low because it's only one third the way through). However it doesn't really show much detail for most months which had low volumes compared to January 2013.

The second plot makes the maximum width equal for each month, meaning the volume is presented as a proportion of the total monthly volume. This means that months cannot be compared directly, but much more detail is presented for any given month.

Three things are clear:
1. The number and size of Bets on SatoshiDice are increasing;
2. People prefer to bet "round numbers" of btc - this is the reason for the "wiggliness" of the plots.
3. The maximum bet size often produces the most monthly volume; the largest portion of most monthly volumes are due to bets 1btc or greater.

Also, from these plots I'd expect the month with the greatest volume to be January 2013, followed by December 2012 and then February 2013.

I also reckon I could sell these plots to psychiatrists as Rorshach ink blots. 

We discussed this earlier in the thread, but theoretically:

1. Let total number of times you want to make a bet on this game = "bets".
2. p is probability of winning the game you're playing, q is the probability of losing the same game, q = 1 - p
3. Then expected losses in a row = -log(bets*p + 1)/log(q)

I believe you! I need to learn more about bitcoin tx minutiae before I can begin to argue this 

They were always spendable, apart from to those who wished to make a mountain out of a molehill.

I don't know if its in this thread anywhere, but is there data available that shows number of consecutive wins or losses, basically streaks? For example, in the 50% bet, what is the longest winning and losing streak? It's like heads or tails, but with the number of bets already made, has there been anything similar to the Monte Carlo gambler's fallacy of 26 consecutive losses?

Of course, you can't see this for each person unless one person uses the same address all the time (like from Bitcoin Spinner.)

I'm interested in all the other categories as well, not just the 50%. I'm curious what is the longest losing streak in the higher win odds (better than 50%) as it won't make much sense in the lower odds since the less than 1 (0.0015%) is obviously going to have 50,000+ consecutive losses.

Aha. The mist clears in this jungle.

OK, I get it now.  They always paid 0.5% for a losing bet, but they used to subtract the transaction fee of 0.0005 (50k satoshis) from that, often leaving less than 0, so they'd send you a token 1 satoshi.  Now they don't subtract the transaction fee, so the min bet (0.01) pays out the full 0.5% when you lose, which is 0.00005000 BTC, or 5k satoshis.

That will result in less dust spam, or at least more easily spendable dust spam.

Interesting! My mistake then.

I took this at face value:
https://bitcointalksearch.org/topic/m.1603935

because I kept seeing comments like this beforehand:
https://bitcointalksearch.org/topic/m.1601927

There has been some SD change because those excellent violin plots show clearly that the min bet size increased 10x at the end of 2012.