The NTRP algorithm was explained in this thread from a while back:
http://tt.tennis-warehouse.com/showthread.php?t=72175
The article that was cited in the thread appears to be gone now, but it went into great detail about the formula, which appears to look like this:
(PRD - CRD) divided by 2 = RDD (which is reapplied to the winner and loser's rating number)
The PRD is the "player rating differential", which is calculated by subtracting the difference between the higher rated player's NTRP and the lower rated player's.
The CRD is the "computer ratings differential", which is a chart the USTA has devised which shows the number that should be used in the formula depending on the score from the match. This chart is reported to assign the following numbers for the match score:
Table 1
Match Results = CRD
7-6, 7-6 = 0.06
7-6, 6-4 = 0.09
6-4, 6-4 = 0.12
6-3, 6-4 = 0.15
6-3, 6-3 = 0.18
7-5, 6-2 = 0.18
6-2, 6-3 = 0.21
6-2, 6-2 = 0.24
6-3, 6-1 = 0.24
6-1, 6-2 = 0.265
6-3, 6-0 = 0.27
6-2, 6-0 = 0.295
6-1, 6-1 = 0.295
6-1, 6-0 = 0.325
The RDD ("ratings differential discrepancy") is the difference between the PRD and CRD, divided by two. This number is then applied to the winner and loser's rating.
I'm not sure if this formula is 100% correct, but it seemed to work in the example in that previous thread.
In your hypothetical scenerio, the exact calculation would depend on the score. However, let's say the 3.0 beat the 5.0 by a score of 6-4, 6-4. Here is how that would effect their ratings:
Lower rated player's rating (2.75) minus the higher rated player's rating (4.75) = 2.00, which is the PRD.
With a 6-4, 6-4 score, the CRD would be 0.12.
2.00 - 0.12 = 1.88 divided by 2 = 0.94 for the RDD, which is added to the winner's rating and subtracted from and loser's.
Winner = 2.75 + .94 = 3.69
Loser = 4.75 -.94 = 3.81
I really wish that I had the original source article that explained the formula because I could then know if I've done this correctly. However, it does look like that one loss could potentially drop a 5.0 down to a 4.0 dynamically and raise the 3.0 to a 4.0. Such a scenerio is very, very unrealistic, but I suppose it's possible. On the other hand, I've also heard that there is a piece of programming within the NTRP computer that throws out abnormal scores like this to eliminate blatant manipulation of rating via thrown matches or a one-time injury. Since the system is supposed to give you a year end rating that represents your
average level of play, I could see how they might throw out your two highest and two lowest scores within TennisLink to balance the rating appropriately (but I don't know if they do this for sure).