But here, with a little more detail compared to my original answer.

For years Yahoo has been using the FIBS rating formula for calculating ratings, but half the equation had been incorrectly programmed from the start, as it did not reward lower rated victors with the correct number of points. Obviously, it is more difficult for a weaker player to defeat a stronger opponent than the other way round, and so he deserves more Rating Points for doing so. This always happens in other Yahoo Games, but up until last week was not ocouring in backgammon. They have recently corrected this (in the last week), and the results will bring ratings more into line with other 2 player games ratings, such as for Pool, Dominoes, Gin etc

As a result, all higher rated (>1502) players will see ratings eroded until they reach a likely range of 1300 – 2300. So if you are 3500 (as I am), save it, as you will never get there again.

In a 1 point match the ‘baseline figure’ for the points at stake is 2 Rating Points; this is unchanged.
This ‘baseline figure’ is gained by the winner, and lost by the loser, but only when the players have identical ratings (Rating Difference = 0). Due to ’rounding – up’ of ratings in your profile, this is likely also to to be the case when RD is less than 200.

However, if there is a bigger difference in rating between the contestants, the RP at stake (which is usually the same for each player) changes.

If the higher rated player wins, he gains between 0 and 2 points (with the lower rated player losing an identical amount), depending on the RD. Typical figures are RD of 1000 gives only 1 RP, RD of 2000 gives 0.4 RP, RD of 3000 gives 0.1 RP. This is not new either.

However, if the lower rated player wins, then the maximum points at stake are double the above ‘baseline figure’, so will be between 2 (minimum) and 4. Mirroring the above RD, if RD is 1000 then the lower rated player stands to gain 3 RP, with his opponent losing the same figure; an RD of 2000 generates 3.6 RP, and an RD of 3000 would cause the higher rated player to lose 3.9 RP.

Therefore, for a 1 point match, a higher rated victor can earn between 0 and 2 points (maximum), whereas a lower rated victor will always earn between 2 and 4 points.
The exception to this is when either player has low ‘experience’: in that cases the value is multiplied by the ‘experience factor’ [(500-E)/100], where E, the experience is simply the total of points played. This can multiply the value by up to 5 times, but only for the ‘inexperienced’ player (E < 400).

In a multi-point match, the points at stake are greater: simply multiply 2 by the square root of the match length (easy to do for 4,9,101) to get the ‘baseline figure’. Naturally, as in all cases from now on, the higher rated player has much more to lose than his opponent: a maximum of double the ‘baseline figure’.

Here are links to some Java FIBS Ratings calculators, based on work by Francois Hochede at FIBS: http://home.nordnet.fr/~fhochede/#RATING
http://www.netadelica.com/bg/fibscalc.html
You simply type in the ratings, match length, and experience, and it tells you the points to be won or lost.
Note that, as Yahoo ratings are recorded to an exact figure, but rounded up when displayed, your ‘displayed rating’ may differ by 1 RP either way from your ‘expected rating’.

The first calculator mentioned can be saved on your dektop and used offline.
The second one is for online use only, but seems to have some extra features, such as probability of winning matches with various lengths/rating differences.

There is a table of RP to be won at an older related question:
http://answers.yahoo.com/question/index;_ylt=AtjpJux.IAuXW_lkHPOZXFrty6IX;_ylv=3?qid=20090715073806AAmYQ9a&show=7#profile-info-LWCl9HAYaa
The results here still hold for when the higher rated player wins. But to get the values for when the lower rated player wins, some numerical manipulation is needed: you need to subtract the column figure from [4 x Sqr(match length)].

Some more detail in my other answer to a similar question at:
http://answers.yahoo.com/question/index?…