From: John Dunlop [jdunlop@acoustoscan.com.au]
Sent: Dienstag, März 30, 2004 09:00 
To: HJK
Subject: NBI report

Hans,
thank you for the new copy of the NBI report, which I also received from
another source. I have made some initial comments in the short time in
case meetings are coming up. As you can see I don't think their formula
is useable.

The NBI report on Ball Roll is an invaluable report as it provides a
scientifically reliable reference source on which to base testing
methods. I hope that it is posted permanently on the website open to
inspection and criticism, and not restricted as it is to a few selected
people.

It is unfortunate that when the work was commissioned a choice was made
to use 2 m gates in direct contradiction of FIFA's established 1 m
system.

The summary gives the trend line for the averages of all the surfaces -
with a correlation coefficient of 0.87. This is somewhat misleading as
individual types of surface give different trendlines with correlation
coefficients ranging from 0.48 to 0.89. This indicates that the
uncertainty in the average trendline is fairly high. Thus this trendline
cannot be used for the basis of a test method with any great certainty
(or accuracy).

Perhaps the solution to the question is in the direct measurements
themselves. Rarely in soccer play does the ball roll to a stop. However
its deceleration (or speed) along the surface is important to play. So
why not use the measurement of delta V to characterise this property of
the surface, rather than degrade the information by converting it to
roll distance. It would also be difficult to determine uncertainty
intervals for roll distance which is necessary when testing for
compliance to a specification.

If we look at the natural surfaces tested, delta V for the system ranges
from 0.55 to 0.87. The average value with 95% confidence intervals is
however 0.68 (0.06) or 0.62 to 0.74. So we might assume a suitable range
for a test method would be 0.5 to 0.8.

If we use this range, only 4 of the 12 sets of artificial surface
measurements pass (the Salter PE of 0.46 (0.04) cannot be failed at the
95% confidence level).
We might entend the test range to 0.4 to 0.8 to give more allowance to
artificial surfaces and to compensate for the heavy slow northern
European grasses used in the study.
However this improves the pass rate to only 5 out of 12.
It seems that only the Salter products come near to replicating the roll
properties of natural grasses. If Salter can do it then we should force
the other products to do so by setting the delta V limits accordingly.