Illustration of the difference between game-by-game and batched average calculations for
Greg Canfell in the Australian Championship, Brisbane 28/12/2005-09/01/2006.

Rating (October 2005): 2354
K-factor: 15


Had the new game-by-game method been used, the calculations (with expected scores from
these tables: http://www.fide.com/official/handbook.asp?level=B0210 ) would be as follows:

Opponent	Rating(Oct05)	Difference	Expected	Result	Change 

Moylan		2112		-242		0.80		0	-0.80	
Lakner		2207		-147		0.70		0	-0.70	
Song, R.	2051		-303		0.86		1	+0.14	
Levi		2241		-113		0.65		1	+0.35	
Ly		2162		-192		0.75		1	+0.25	
Wohl		2439		+85		0.38		0.5	+0.12	
Zhao		2461		+107		0.35		0.5	+0.15	
Lane		2445		+91		0.38		0	-0.38	
Bjelobrk	2399		+45		0.44		1	+0.56	
Chandler	2537		+183		0.26		0	-0.26	
Johansen	2462		+108		0.35		0	-0.35	
-----------------------------------------------------------------------------	
Total									-0.92

So Change*K = -0.92x15 = -13.8 rating points.



However, the calculations for this tournament were done by the old method of first
taking the average rating of opponents:

Average: 2319.6
Difference: -34

Expected percentage based on difference = 55%
(see: http://www.fide.com/official/handbook.asp?level=B0210 )

Expected score from 55% over 11 games = 0.55x11 = 6.05/11

Actual score = 5.0/11

Change is Diff*K = -1.05x15 = -15.75 rating points

(as seen here: http://www.fide.com/ratings/tourarc.phtml?codt=19&field1=3200418 )


In conclusion:

It would be unusual for the difference between the two methods to be more than a
rating point or two per tournament. However, the old batched-average method is not
considered as statistically accurate - it assumes a linear distribution and somewhat 
truncates the extremities of the Bell/Normal Curve - and was only ever used to 
simplify calculations.