[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Information


Given what I am about to offer, your Subject title "Information" is an
appropriate semantic choice as Neural Nets transform Information
into "Knowledge" and, knowledge about dependence of S on A and V
is your question.

If you have MathCAD there is an interesting (new) way to judge dependence
of your Dependent "S" upon it's two Independent variables "A" and "V".

The method applies a Neural Network of the "Backpropagation" type.
MathCAD has a Neural Net library which includes the Backprop type net.

Your indicated case would be modelled as a 2-Input, 1-Output Backprop.
You would "train" the net such that the "A" and "V" variables are individually
placed at their own "Input" while the "S" variable is placed at the Output.

The Backprop algorithm then uses these three (3) data sets to teach the
"weights" of the 2-Input, 1-Output neuron such as to minimize the error
produced at the ("desired") single Output.

An interesting property of a Backprop is that you can always ask about
the effect each Input (A or V) variable has on the generated "S" output ...
which seems to be YOUR question.

A second property of any Neural Net (not just Backprop) is that a trained
net will "guess" the output "S" that "should" be produced by a set of A and V
inputs it has NEVER seen.  An interesting way to use this Neural Net property
is to sweep the values of one Input (A or V) while keeping the other constant
and see how that one Input effects the "S" Output (V or A held constant).

Richard Fabbri
Scarce Ideas
____________________Reply Separator______________________
>I have conducted an experiment where there was a dependent variable (S)
>and two independent variables (A and V). I have carried out an analysis
>of variance, which showed a significant main effect of both variables (A
>and V) on the dependent variable (S), and a significant interaction
>between these both variables (A x V).
>I'd like to assess whether the effects of the variables A and V were
>very strong (to assess the "strength" of these effects).
>I was adviced to use a statistical variable that I do not know at all.
>This is: "omega square" for estimating the degree of association between
>independent and dependent variables. I was told that most statistical
>packages use a simpler formula (for instance, in SPSS, there is "eta").
>The problem is that my stats package is Statistica and I can't find
>"omega square" neither "eta".
>Is there someone who has yet calculated this variable (omega square?
>eta?) and who knows how calculating it with Statistica?
>Thank you in advance.
>Stephanie Viollon.