🚀 go-pugleaf

Lyle Brenner writes:
 > What have people found to be the best way to describe (conceptually)
 > degrees of freedom?  I've never come across a fully satisfactory
 > clear, understandable, intuitive, non-mathematical "definition."

This might be of some help ....

Although degrees of freedom can be explained in the context of, "I have a
divided 100 people into six groups and know five of the group sizes,
therefore I must know the size of the sixth group", I wonder if another
perspective might be more functional for neophyte statisticians?

Why not define degrees of freedom as the number of parameters that need to
be estimated to allow us to create the current estimator or model.  Thus,
the degrees of freedom for the mean is N as we have had to estimate no
other parameters;  the degrees of freedom for the variance/standard
deviation is N-1 as we have had to estimate the population mean with X-bar;
the degrees of freedom for the MSE term in simple (1 predictor) regression
is N-2 as we have had to estimate two population parameters (the y-
intercept and the slope) via the y-intercept and the slope of the estimated
regression line. The same logic holds true for Chi-square, log-linear,
ANOVA, etc.


Geoff Selig        Phone (514)-848-7666 |        _,  /|     The Cat's Meow
Academic Services    FAX (514)-848-7622 |        \ o.O'
Department of Computing Services        |        =(___)=
Concordia University   (H928)           |            U
Montreal, Quebec   H3G 1M8              |  ilfc594@vax2.concordia.ca
                                        |  ilfc594@conu2.BITNET