I intend to have items or articles at Sportscience about these updates,
but meantime...

Reliability (xrely.xls): I updated this one extensively a few months ago by
adding new analyses and improving the cosmetics and instructions. A few
days ago someone contacted me with a problem: the standardized typical error
was showing up as #NUM!. I had been standardizing the error by dividing it
by the "pure" between-subject standard deviation, which I believe is the
correct approach. The pure between-subject SD is the square root of (the
observed SD squared minus the error squared), but because of sampling
variation it is possible to get an error greater than the observed between,
so the result will be the square root of a negative number, hence #NUM!.
The same thing would occur in the retest correlation was negative,
presumably. At this stage I have altered the spreadsheet to use the
observed SD instead of the pure SD, but that means the standardized error
will usually top out at around 1.0. I will resolve the problem and get it
all peer-reviewed soon.

The controlled-trial spreadsheets (xParallelGroupsTrial.xls,
xPrePostCrossover.xls, xPostOnlyCrossover.xls): if you had a blank column
of values for one of the trials or treatments with these spreadsheets, a
Bill-Gates effect occurred somewhere in the log-transformed blocks or plots.
I have now provided advice in the spreadsheets on how to avoid this
annoyance: basically you just make sure you don't have blank columns, by
filling them with data from one of your other columns.

The pre-post crossover spreadsheet (xPrePostCrossover.xls): this one had
labels for "groups" when they should have been "treatments".

The post-only crossover spreadsheet (xPostOnlyCrossover.xls): some of the
comments referred to "trials" instead of "treatments".

I depend on you people to give me feedback on such problems. Please don't
hesitate.

I'm also hoping to write an article on converting validity statistics from
one population to another and for getting validity statistics for a generic
population that would allow sensible meta-analysis of studies of the
validity of a measure. I reported on these spreadsheets at the ECSS
conference. If anyone wants to try these out, please contact me.

By the way, it was third time unlucky for my proposal on measures of
performance at next year's ACSM. Sigh... ECSS has a better focus on
sport science, methinks.

Will
Will G Hopkins, PhD FACSM
Contact info: http://sportsci.org/will
Sportscience: http://sportsci.org <http://sportsci.org/>
Statistics: http://newstats.org <http://newstats.org/>
Be creative: break rules.