I play on FIBS and am rated about 1750.

I recently started analyzing my matches with gnu. I had for awhile played gnu using tutor mode.

The one thing I think that must be a an imperfection with gnu is that when it analyzes one of my FIBS matches it almost always determines that the loser played a better match. Implying that the winner played poorer but was lucky I guess.

Now the odds of that being true on a constant basis seem unlikely to me, particularly when you play someone who is around your rating. So it must be some sort of incorrect assumption in the program. Maybe when you are losing you have less decisions to make so the % of correct decisions goes up.

The other thing is that when GNU says my play was "doubtful" and I compare it to what GNU would have done, I really don't have the ability to see the difference. I basically have given up on that and just try to reduce the decisions that GNU says are "bad" or "very bad."

Curious if anyone feels the same or had thoughts on that or had suggestions.

Post some of the 'doubtful' positions here and we can explain what's wrong with your play.

It's well known that the most lucky player wins in the short run. In the long run the best player usually overcomes the short term luck factor.

I ran an experiment setting XG to play itself at XG+ level for 101 games. One side had an overall luck advantage of 0.0037 (the sum of positive and negative luck). The lucky side always won and the average luck per game was 1.29.

"Almost always" is not a well defined phrase. "Constant" is, but I suspect that is an exaggeration. Can you give us some actual figures? Say, run your entire collection of recorded matches through GNU, check every one, and report here how many showed this effect.

For example, are we talking about 3 of 4 matches, or 75 of 100? Also, how long are these matches? 5 pointers are much more susceptible to such luck than 21 pointers. But online matches are most often 5 or 7, and rarely more than 9.

But yes, it happens. I run my matches from SHG through GNU, and indeed sometimes the player with a higher error rate won. That's part of the game, a feature not a bug.

Ok so I analysed my matches for Jan and Feb. I typically play 5 or 7 and I typically try to get someone as close to my rating as possible. Of course, this is not always possible. I will play lesser ranked players if they invite me and have significant experience.

I played: 111 matches
Total won: 63 matches
% I won: 57%
% GNU said I played better: 61%
overall (me or opponent)
the % GNU say better player won: 57%
% lesser player won: 43%

43% of time gnu says the loser played better than the winner. Seems higher than I expected. I am thinking the explanations are 1) if I play similarly ranked players, our skill level will be similar, therefore, luck is a major factor in deciding who wins, 2) when you are losing, you have less decisions to make, so GNU counts your error rate as less, 3) backgammon is just a game of luck.

What do you think?

Your points 1, 2, and 3 all have some validity.

But I think that the main lesson here is the difference between subjective experience and actual data analysis. Your original perceptions of the lower performing player winning "almost all" or "constant" turned out to be very inaccurate when checked dispassionately with your own data.

Your willingness to assess the actual data and post the result is commendable. This is too high a hill to climb for most of those people who complain of outcomes that they feel are unjust.

Seeing the statistics, I would reword my original post. But I still think that 43% of losers playing better than winners is surprisingly higher than what I expected before I started analyzing my matches on gnu. So I guess my original point is still valid.

Not sure what to think about that and backgammon. I guess you would say if you don't like it take up chess.

It's 5/7-point matches. I mean I can beat GNU in plenty of those even though I suck. Clearly dice can outrun even a massive skill difference over a few games.

It's true there's a lot of luck in backgammon. However, playing very short matches (5-7 points) doesn't give the skill a chance to show itself and accentuates the luck factor. Most Open divisions of tournaments nowadays play 11-15 point matches, which I think is still pretty short but which does allow skill to play a bigger role.

Old-time tournaments used to play 15-point matches in the first round and work their way up to a 25-point final. Again, skill plays a much bigger role in that case.

When money players play a chouette, a chouette session can go on for many hours and might involve 50-100 games. Again, skill plays a major role.

Yep, but I analyzed 111 matches not 3. So I think the number of matches I analyzed makes up for the match length.

No, not at all. If I played 111 matches against GNU, you'd still see a bunch of matches where the worse player (me) won. It's not like I stop winning at all because I played a bunch of small matches?!?!??!? If somebody good played GNU a bunch of 5/7 point matches, you'd see in the 40%s. If you're playing people of similar skill, then seeing in the 40s isn't strange at all for 5/7 point matches. It's basically exactly what you should expect.

He's right. Try playing 21-point matches against GNU and see how many you win. My guess will be 10-15%.

Hah, tough luck in short matches confirmed. Two match results from yesterday on safe harbor. Error rates are from GNU 2 ply "snowie error rate" - not sure how accurate this is or how it maps to XG PR. Still:

7 point match
my rate: -5.7
opponent rate: -17.6
I lost 7-4.

5 point match
my rate: -5.5
opponent rate: -19.1
I lost 5-4.

Admittedly it is a bummer in cases like this where I played somewhat well and the opponent played .. just plain badly. But that's the game. Of course I also have records where I was on the upside of such a situation. But I didn't play at -17 !

But I still got some learning value. In one of the matches I made a very large checker play error in a mutual scramble. I anchored in the opponents outfield when I should have ran the front man, holding one back in his homeboard. And .. I still won the game. Luck is a fickle beast.