Re: Covariate in ANCOVA question

On 24 Oct, 01:33, Richard Ulrich <Rich.Ulr...@xxxxxxxxxxx> wrote:
Sometimes, the useful idea is to figure out what the basic
"variables" ought to be, especially if they can be described
in terms that are fairly independent.

Well, WHAT do you see in your plots?

In a slightly simplified version we see in a diagram plotting call
duration on the x-axis and pitch on the y-axis that there are three
lines:
- all three lines start at high pitch for short calls and end at lower
pitch for longer duration calls
- the three lines are roughly parallel
- and now the potential problem: The three lines differ in "length",
i.e., for instance, for group 1 there are calls having duration from 1
to 4 ms, for group 2 there are calls having durations from 1 to 6 ms
and for group 3 there are call durations from 1 to 10 ms.

Thus, from the graphs one would immediately conclude that the pitch
differs between groups, despite differences in call duration. However,
the difference in call durations would introduce a severe confound if
we would just compare the pitch between groups averaged across all
call durations. How to control for that in a statistical way?


Do they start the same? That already answers one version
of "what pitch."
No, they start on different levels


They differ in duration.
Yes.

Do they differ in rate of drop? Are the drops equally
linear?
The rate of drop is linear and comparable between groups.

A big problem with covariance is that you can't trust the
simple results if the groups differ on the covariate.
Oh, I thought that's the whole point of ANCOVA? Or is it solely to
reduce unexplained variance to increase statistical power?

The
usual, easy tactic is to make plots, look at the plots, and
figure out the simplest way to describe whatever is going on.
Well, that's what I've tried with the original post :-)



.

Relevant Pages

  • Re: Covariate in ANCOVA question
    ... WHAT do you see in your plots? ... duration on the x-axis and pitch on the y-axis that there are three ... all three lines start at high pitch for short calls and end at lower ... The rate of drop is linear and comparable between groups. ...
    (sci.stat.edu)
  • Re: Serial = Random ?
    ... observes that a relation between pitch and duration must find some ... the analogy of the logarithmic division of the pitch octave. ... For many European composers, however, serialism is only distantly ...
    (rec.music.theory)
  • Re: Caryn is going HOME!!!!
    ... I just didn't have the heart to actually pitch them ... before I stored them in my basement for the duration. ...
    (rec.crafts.textiles.needlework)
  • PSOLA -- changes in duration
    ... So I thought about switching to PSOLA. ... "...the smaller signals are modified by either repeating or leaving out ... depending on whether the pitch of the target speaker ... the duration of the signal will change. ...
    (comp.dsp)
Quantcast