Not to quibble, but again I believe that you are misstating the
facts. The rationale for developing the normalized power algorithm
was *precisely* to be able to obtain an estimate of someone's
performance, i.e., the isopower output that they could have
maintained *for the same degree of physiological (again, esp.
metabolic) strain*. It should be obvious that this only the same as
their *maximal* performance ability if the powermeter file being
analyzed originates from a situation in which the athlete was pushing
themselves maximally. Even in cases where they were not, however,
calculating the normalized power can provide useful insight, e.g., as
a way of comparing the intensity of training sessions with
dramatically different distributions of power output (e.g., interval
vs. steady-state training, hilly vs. flat rides, group vs. solo
efforts).

To derive the normalized power algorithm, I used a first principles,
or "top down", approach, which any modeler will tell you is just as
valid as the empirical, or "bottom up", approach that you advocate.
Indeed, in this particular case I believe that the first
principles/top down approach is superior, simply because it would be
impossible to collect enough data under the nearly infinite number of
situations in which the normalized power algorithm might be applied
to obtain truly precise estimates of best-fitting constants. Thus,
the empirical/bottom up approach that you advocate would likely
simply serve to provide a false sense of accuracy and/or precision,
without providing any real increase in such.

As you indicate, one benefit to performing a study such as you
suggest is that it can provide a formal test of the
accuracy/precision of the normalized power algorithm, regardless of
how the constants themselves are derived. Indeed, it is for this very
resaon that Lindsay Edwards and Simon Jobson in the UK are presently
conducting just such a study, some of the results of which I included
in my presentation at ACSM. As you may or may not recall, these
preliminary data are quite encouraging (and consistent with the
experience of numerous coaches and cyclists already using normalized
power), although of course more data needs to be collected to
complete the picture.

Finally, while the notion of attempting to capture more information
about training demands using a zone-based approach to powermeter data
is appealing, at present it is unclear whether such an approach is
even possible, much less whether it provides any benefit over a more
global approach based on just the normalized power and duration of a
workout (i.e., training stress score, or TSS). Indeed, as Dave Martin
related to me at ACSM, the AIS has tried just such an approach, and
found it so unwieldy that they were unable to ascertain any
relationship between such zone-based measures of the training load
and performance. On the other hand, even cruder metrics than TSS,
e.g., Banister's heart rate based training impulse, or TRIMP, have
been repeatedly shown to work quite well when used as the input
function to the impulse-response model. Given that TSS is "immune" to
the numerous factors that can influence heart rate during exercise at
a given intensity and that it gives credit for exercise at >100% of
VO2max (which TRIMP does/can not, since heart rate cannot exceed 100%
of maximum), I would *hypothesize* that it would work just as well,
if not better, as TRIMP as a measure of the training load when
attempting to model the relationship between training and
performance. Perhaps some reader of this list would be interested in
testing this hypothesis.

Andy Coggan