Re: Anova with continuous variable

junkjem@xxxxxxxxx a écrit :
Hello,
Lets take for example the following experiment (which is similar to
mine but with some other variable...)
Someone has to taste a meal is estimate how spicy it is.
It is always the same meal, but the quantity of salt and pepper
varies.
In some conditions there is no pepper, in other there is the same
amount of pepper and salt.
The quantity of salt (and pepper) varied between condition, let's say
3, 5, 7, 9, and 11 mg.
So to summarize here are the value of the independent variables:

Cond Salt Pepper
1 3 0
2 5 0
3 7 0
4 9 0
5 3 3
6 5 5
7 7 7
8 9 9
I have 6 estimations per conditions.
I have done an anova with two factors: 1] quantity of salt, 2]
presence of pepper.
Of course both factor are significant. The purpose of the experiment
is to test if the interaction is significant. We want to know if the
increase in the spice estimate with the introduction of pepper is
constant over the quantity of salt.

I have done the anova using Matlab (function anovan). I have to option
to set the quantity of the salt to a continuous variable. It seems
intuitive to me to use this option. But the problem is that I don't
know exactly of the algorithm used in that case.
I understand the algorithm of a typical ANOVA, but I can't find any
information on how to treat a continuous variable.

You're barking up the wrong tree : what you're tring to do is not but
regression.

Indeed, both are special cases of general linear models. I suggest to
look that up in a good intermediate-level textbook.

I wouldn't use the general-purpose Matlab for this kind of problems, but
a stattistical package such as R.

Furthermore, there are some general (psychophysiological) reasons to
question the linearity of the relationships you're trying to discover. A
nonlinear regression might be more appropriate...

HTH

Emmanuel Charpentier
.

Quantcast