Too many Jokers on-line?
Gregg Cattanach
'there are too many jokers'. This comment seems especially prevalent
related to GamesGrid (but probably only because I have lots of friends
that play there), but I gather a variety of players at any BG server
often make this complaint. I hear this even from some very top-notch
players. I've always been skeptical of the notion that the dice are
somehow skewed, either intentionally or through some defect in the
random number generator. Granted pseudo-number generators do NOT
produce a truly random sequence of numbers, but as long as there is no
way to identify the difference between a pseudo-random sequence and an
actual random sequence of rolls, then the pseudo-random sequence is
adequate for the task.
I decided to collect some data (what a concept!) to test the 'too many
jokers' theory. I found 20 long matches (several over 25 games per
match) played over the board with real dice (various web sources), and
compared them to an equivalent number of matches (similar number of
total dice rolls) on matches I'd saved from GamesGrid. The matches
were selected at random before looking at the results. Using Snowie, I
then did a simple joker count for each player and added them up. I
used a threshold of 0.350 for the jokers which is lower than the
suggested Snowie default value. This seemed to work well, because it
captured more 'joker' rolls without trivializing the concept. A swing
of 0.350 or more on a single roll is usually perceived as quite a good
(or bad) roll by most players. Also, the joker calculation is not
impacted by the skill of the players. A roll is counted as a joker if
the equity swing is 0.350 or more when played in the best way
according to Snowie.
Here are the results:
#rolls Jokers PL1 Jokers PL2 Joker every X rolls
REAL DICE: 10325 486 497 10.50
GAMESGRID: 11394 491 500 11.50
In this sample, the joker frequency is actually LOWER at GamesGrid
than compared to these real dice matches.
There isn't any bias toward the winner or the loser of each game as
both sides are being counted. Also, Snowie adds to the joker count
the 'anti-jokers' which are rolls that are bad for the roller. These
are counted as jokers for the non-roller. This isn't a gigantic
sample, but I think if there was some real bias for GamesGrid to be
giving someone 'exactly what they need' too often, it would appear in
these numbers.
So why do lots of people think this about the Grid (and servers in
general)? I think one possibility is that it is easy to trust dice
you throw yourself, but playing on-line is like playing a match where
a third person is throwing dice in a box and announcing what your roll
is, without you seeing the numbers. Also, the speed of play on-line
can give the emotional experience of a lot of good and bad rolls
coming quickly, when the frequency of these exceptional rolls is
nearly identical to what you get over a real board. The
over-the-board game generally is slower, if only because of the time
taken to shake, roll, and move the checkers.
I thought this was interesting and wanted to share with the group :)
Gregg Cattanach
gcattana...@prodigy.net
Nis Jorgensen
wrote:
>One common complaint I hear from some players about on-line dice is
>'there are too many jokers'.
I will not comment here on the rest of your post. But I find it
somewhat remarkable that the same is the case in bridge: people seem
to believe that computer-dealt cards result in more "freak hands" -
especially long suits (=many cards in the same color). I don't know if
any statistics has been made, though.
There is, however, a good reason why computer-dealt hands should be
more "freaky": if you do not shuffle a deck of cards well enough, the
suits will be more evenly distributed than in a random deal! This is
because of the way hands are dealt in bridge - one card at a time.
Also, the psychological benefit of being able to blame a computer
should not be underestimated.
For clarity: The above mentioned computer-dealt hands are being played
by physical people with physical card decks, so the analogy to online
bg is not that good.
--
Nis Jorgensen
Amsterdam
Please include only relevant quotes, and reply below the quoted text. Thanks
Frank Mazza
I don't perceive any difference - jokers, doubles, whatever, when I
play online or against Snowie or use real dice. I haven't done an
analysis, but jokers and freak doubles seem to come up as often over a
real board as on the Grid.
Last time I played over a real board doubles seemed to come every
other roll for several games. A skeptic of online dice seeing those
game rolls on the Grid would think she had her convictions of flaky
dice confirmed.
I think your analysis of the psychological factors involved in
suspicion of online dice is probably close to the mark. Especially
with the Rodentbots moving so fast you have to keep your eyes pealed
on the outer board just to try to catch a glimpse of what they rolled!
Plus, people sometimes forget that random dice doesn't mean freaky
rolls won't happen, or even happen for a while, but that over time
everything evens out statistically.
I know that is hard to swallow when GGRaccoon has just joker rolled
you out with multiple doubles three games in a row in non-contact
bearoffs where in each game you had a significant lead (happened to me
last week), but that is dice and backgammon!
Frank Mazza
On 7 Sep 2001 06:02:12 -0700, Zox...@excite.com (Gregg Cattanach)
wrote:
>One common complaint I hear from some players about on-line dice is
Tapio Palmroth
Hmmmmmm,,,,,
is it possible for a bot , if enough matches are played , to "remember"
the sequence of rolls to come ?
This is backgammon and odd things do happen often , as i wittnessed just on
real board with real dice , odd indeed it was and this really can happen in
real world too.
Sure that psyco-affect is very true also .
But , as playing against a bot that has played very very much it seems that
when i win , i am extremely lucky in that and when the bot wins , same thing
, extremely lucky , especially when running from behind and early cubes
both sides ).
When playing a bot with lot less experince , the games seem to be more as
normal games , although that bot is to be a bit better than the first one.
And with that first mentioned bot i sure am having more laughs.
I have quitted playing that bot many times , but i do need my laughs.
tapio
Douglas Zare
[...]
> Also, the joker calculation is not
> impacted by the skill of the players.
That's not correct. The style of play can affect the number of jokers per
game and per move. For example, if one rolls a full prime home against one
checker there may be no jokers for either side for a very long sequence of
rolls. If your play leads to such positions often, then you may have a
lower joker count per move than a pair of people who blitz or get blitzed
in most games. If your play leads to two-way gammonish positions, (or if
you just gammon your opponent a lot) there will be more jokers than if you
play a high-anchor holding game.
If you don't double immediately at 2-away 2-away, but wait for a potential
market loser, the normalized equity changes get magnified by a factor of
3. Therefore there will be more jokers, even if there is no real
difference in the game.
> A roll is counted as a joker if
> the equity swing is 0.350 or more when played in the best way
> according to Snowie.
>
> Here are the results:
>
> #rolls Jokers PL1 Jokers PL2 Joker every X rolls
> REAL DICE: 10325 486 497 10.50
> GAMESGRID: 11394 491 500 11.50
>
> In this sample, the joker frequency is actually LOWER at GamesGrid
> than compared to these real dice matches.
GamesGrid is really fast (if you have a good connection to it). Therefore
people often play out positions where one side would resign over the
board. No jokers are possible if the position is gin, but the number of
moves increases. I don't think that this accounts for all of the
difference between 11.5 and 10.5 (tacking on 2-3 moves to the end of each
normal game played to completion), but it might even be more.
Similarly, by drawing out a race where you are ahead by a huge amount
(e.g., play 3-2 4/1 3/1 at 0 0.01 99.99 - 0.01 0 0), or postponing an
optional double-out, you can get a better "millipoints per move" rating
from Snowie.
> There isn't any bias toward the winner or the loser of each game as
> both sides are being counted. Also, Snowie adds to the joker count
> the 'anti-jokers' which are rolls that are bad for the roller. These
> are counted as jokers for the non-roller. This isn't a gigantic
> sample, but I think if there was some real bias for GamesGrid to be
> giving someone 'exactly what they need' too often, it would appear in
> these numbers.
>
> So why do lots of people think this about the Grid (and servers in
> general)? I think one possibility is that it is easy to trust dice
> you throw yourself, but playing on-line is like playing a match where
> a third person is throwing dice in a box and announcing what your roll
> is, without you seeing the numbers. Also, the speed of play on-line
> can give the emotional experience of a lot of good and bad rolls
> coming quickly, when the frequency of these exceptional rolls is
> nearly identical to what you get over a real board. The
> over-the-board game generally is slower, if only because of the time
> taken to shake, roll, and move the checkers.
>
> I thought this was interesting and wanted to share with the group :)
Yes, but it is still hard to extract a conclusion from it.
If someone is unlucky in 75% of their 1-point matches, this probably means
they are a terrible player. This can happen with fair dice if they are
unlucky by 25% mwc when they lose and lucky by 3 times as much when they
win because they make 25% mwc more mistakes than their opponents. On the
other hand, it is possible to detect whether there are unusually many or
few rolls which allow the underdog to turn the game around, if one is
careful to use an unbiased indicator. Unfortunately, Snowie does not
provide these.
Douglas Zare
Douglas Zare
Tapio Palmroth wrote:
> Hmmmmmm,,,,,
> is it possible for a bot , if enough matches are played , to "remember"
> the sequence of rolls to come ? [...]
It would be an amusing backgammon variant to roll openly before your opponent
decides on the checker play for the previous move; it would be a huge handicap
if this were one-sided. If one analyzed such games using Snowie, I don't think
the player(s) with extra information would be rated well at all, as one could
leave tons of extra shots that would simply not be hit, and avoid splitting
precisely when it would get pointed on (or let your opponent decide to hit
loose, in the 2-way version). The commendations that Snowie gives to other bots'
play indicates that they do not use extra information often.
Some people may think that bots leave way too many blots and when these are not
hit it might be an indication that the bot _knew_ it was safe. For example,
rather than playing 13/6 with an opening 4-3... These people are wrong, of
course, but then most people are wrong.
Douglas Zare