Re: Help please!!! Two stats questions from a biologist...

On 25 jun, 18:34, Ray Koopman <koop...@xxxxxx> wrote:
On Jun 25, 3:01 pm, fil...@xxxxxxxxx wrote:

Thanks a lot guys! What do you know about Pearson's and Spearman's
distances? Are they suitable for comparing composition profiles? Ray:
Since the differences are to the square, Hellinger distance is
insenitive to negative and positive correlations, right?

Positive correlations (larger p's paired with larger q's,
smaller p's paired with smaller q's) will give smaller distances;
negative correlations (larger p's paired with smaller q's,
smaller p's paired with larger q's) will give larger distances.

I would call (1-r) and (1-rho) squared distances, not distances.
To get the analogous squared Hellinger distance, omit the outer
square root in the expression I gave. An alternate expression for
the squared Hellinger distance is 2(1 - sum sqrt(p_i*q_i)).

Roy:Interesting. Squared Hellinger distance is much more sensitive to
the
absence of one or more components than Pearson's, and that makes
much more biological sense. Very cool.Thanks!!

.

Relevant Pages

  • Re: Luminance to candlepower and inverse square law
    ... As long as the diameter of the beam increases linearly with distance ... This condition is generally satisfied for lamp sources for distances ... the inverse-square law at large enough distances too (in the ... > square of the separation between the lamp and the reflector. ...
    (sci.optics)
  • Re: Help please!!! Two stats questions from a biologist...
    ... Since the differences are to the square, ... Positive correlations (larger p's paired with larger q's, ... smaller p's paired with smaller q's) will give smaller distances; ... To get the analogous squared Hellinger distance, ...
    (sci.stat.consult)
  • Re: SMOOTH URYSOHNS LEMMA
    ... Euclidean space R^n, ... Just square the distances. ... There's no need for piecewise -- that was an hallucination on my part. ...
    (sci.math)
  • Re: Rational Trig
    ... >> simply by replacing distances by their squares, ... > there is no reason to take a square root at all. ... > instead of angles. ... alot of stuff like trangulation without any square roots look up tables etc ...
    (comp.programming)
  • Re: mulling metrics
    ... If we arrange them as a square, than two distances must be square roots ... more dimensional objects is determined by the embedding dimension. ...
    (sci.physics)