MLR Model Need Help

[wkcalvin]

Hi,

I am using SPSS buliding a horse racing model with MLR. I have a problem that is the factors will neutralized each others.

Example:

Factor 1: FP - nomalized finish position of a horse
Factor 2: BL - nomalized beaten length of same horse

If I use FP or BL alone, works fine, but if I use both together, they did not boost each other but neutralized each other.

If I use one of them, the model is better, but use both, the model worsted, any body have idea about this issue ?

[OTM Al]

They are strongly correlated variables is what is causing the problem. You could make a single variable by multiplying the 2 non-normalized variables together and then normalizing it by dividing by the size of the field.

[arkansasman]

Calvin,

As you go forward and add factors to your model, you will find it harder to find factors that show significance on "out of sample." That is because your core group of factors already contain the most significant measures of horse performance.

Look at your T-values. If you add a factor and it shows significance "in sample" but then your significance on "out of sample" degrades, look at your interaction among your factors. The factor is obviously stealing T-value from another factor.

John