The Advanced Goodness of Fit report is a text-based report that shows detailed information on the goodness of fit of the response surface for a selected output parameter. This report can help you to assess whether your response surface is reliable and accurate enough for you to proceed with confidence. It is available only for the response surface generated by the Standard Response Surface - Full 2nd-Order Polynomials response surface.
To generate the Advanced Goodness of Fit report for a given output parameter:
Under the Quality node of the response surface Outline view, select Goodness of Fit.
In the Table view, right-click the column for the desired output parameter and select .
Once generated, the Advanced Goodness of Fit report displays in a separate window.
The following properties provide general information about the Goodness of Fit analysis:
Type of polynomial regression. Cross Quadratic Surface means
that the response surface uses constant terms, linear terms (), pure quadratic terms (
) and cross-quadratic terms (
). You cannot change this property.
Applied transformation on input parameters before solving the response surface. You cannot change this property.
Applied transformation on output parameters before solving the response surface. To edit this property, select an output parameter in the Outline view.
Algorithm used to select the relevant regression terms. For Modified Linear Forward Stepwise Regression, the individual regression terms are iteratively added to the regression model if they are found to cause a significant improvement of the regression results. A partial F test is used to determine the significance of the individual regression terms. If a regression term is not significant, it is ignored.
Used to filter out insignificant regression terms. A regression term is ignored when its probability of being zero is greater than 1-Filtering Confidence Level. For example, if the filtering confidence level is 0.95, a regression term is ignored when its probability of being zero is greater than 0.05.
The following properties provide all the scaling models, transformation models, and regression term coefficients used to build the regression model:

Scaling models used to build the regression model.
Transformation models used to build the regression model.
Regression term coefficients used to build the regression model. These are displayed in a table with auxiliary values, as follows:
Std. Dev. of Coefficient: Statistical estimation of the dispersion of the coefficient. A low standard deviation indicates stability of the coefficient: a high standard deviation indicates that the coefficient has a wide range of potential values.
Prob. Coeff. =0: Corresponds to the probability that the coefficient is zero. If this probability is high, then it suggests that the coefficient is insignificant. More significant coefficients have a lower probability of being zero.
Confidence Interval [Lower Bound; Upper Bound]: Corresponds to the 95% confidence interval of each coefficient. There is a 95% probability that the coefficient is in this range, so the smaller the confidence interval, the more likely it is that the coefficient is accurate.
The following properties provide statistical data that enable you to measure the stability of the regression model and to understand which input parameters are significant.
VIF measures the degree of multi-collinearity of the independent variable with the other independent variables in the
regression model. If the Maximum VIF for Full Regression Model value
is very high (>10), it means that there is a high interdependency among the terms and
the response surface is not unique. If the response surface is not reliable in spite of a
good R2 value, you should add more points to enrich the
response surface.
Pearson correlation of selected regression terms. A non-diagonal term measures the collinearity between two independent variables. If the absolute value of this term is close to 1, it means that the response surface is not unique. As with a high VIF value, if the response surface is not reliable in spite of the good R2 value, you should add more points to enrich the response surface.
Each row of this table corresponds to a sample point (used to build the response surface) where:
Residual Value: Difference between the Sample Value and the Approximated Value.
Approx. Std Dev: Standard deviation of the predicted value at this sample point. If the standard deviation is small, the predicted value is statistically stable and you can have confidence in the predicted values in the area close to this point.
Lower Bound and Upper Bound: Bounds of the 95% confidence interval of the approximated value. There is a 95% probability that the approximated value is in this range, so the smaller the confidence interval, the more likely it is that the approximated value is accurate.
Each row of this table corresponds to a sample point (used to build the response surface) where:
Hat Matrix Diagonal: These elements are the leverages that describe the influence each observed value has on the fitted value for the same observation
Student's Residual: Corresponds to the residual divided by an estimate of its standard deviation.
Student's Deleted Residual: Student's Residual with the observation removed.
Cook's Distance: Measures the effect of the i-th observation on all fitted values. A point with a value greater than 1 can be considered influential.
P-value: Probability of obtaining a test statistic at least as extreme as the one that was actually observed, assuming that the null hypothesis is true. This value represents the error percentage you can make if you reject the null hypothesis, where the null hypothesis is that "the sample point has not influenced the observed result." Traditionally, following Fisher, one rejects the null hypothesis if the p-value is less than or equal to a specified significance level, often 0.05, or more stringent values, such as 0.02 or 0.01.
Summary of previous data and additional metrics commonly used to estimate the quality of the response surface. For more information, see Goodness of Fit.