|
Analyze Factorial DesignTwo-Level Factorial Designs |
For each main effect term and each interaction term, there is a coefficient and an effect. The effect is two times the coefficient.
Look at the effects and coefficients to determine the relative strength of the factors. The value and the sign are both important in the determination:
- A positive sign indicates that the high factor setting results in a higher response than the low setting.
- A negative sign indicates that the low factor setting results in a higher response than the high setting.
Because these coefficients are estimated using coded units, putting the uncoded factor values into an equation with these coefficients yields incorrect predictions about strength. If the model is hierarchical, then the regression equation displays the uncoded coefficients.
In theory, any variance inflation factor (VIF) value greater than 1 can inflate the variance of the coefficients so much that statistical significance is a less useful way to identify candidate models. In practice, values greater than 5–10 can produce unstable coefficients that are difficult to interpret and, thereby, prompt corrective measures
Example Output |
Coded Coefficients
Term Effect Coef SE Coef T-Value P-Value VIF Constant 56.0 21.0 2.66 0.056 MeasTemp -1.229 0.979 -1.25 0.278 5.87 Material 5.316 2.658 0.678 3.92 0.017 3.23 InjPress 5.645 2.822 0.401 7.04 0.002 1.13 InjTemp 4.355 2.177 0.378 5.76 0.005 1.00 CoolTemp -3.457 -1.729 0.420 -4.12 0.015 1.24 Material*InjPress -0.723 -0.361 0.415 -0.87 0.433 1.21 Material*InjTemp -1.025 -0.512 0.443 -1.16 0.312 1.38 Material*CoolTemp -0.208 -0.104 0.510 -0.20 0.848 1.82 InjPress*InjTemp -0.837 -0.419 0.510 -0.82 0.458 1.82 InjPress*CoolTemp -0.055 -0.027 0.382 -0.07 0.947 1.03 InjTemp*CoolTemp 1.933 0.966 0.381 2.54 0.064 1.02 |
Interpretation |
For the insulation data, the effects suggest the following interpretations:
The VIF for the covariate is 5.87. The temperature at the time of measurement is moderately correlated with the material and more weakly correlated with the other factors and interactions. The temperature could have an effect that the multicollinearity hides. The coefficient could also change dramatically as the model changes.