Analysis

Analysis Options

Maximum Coded Value: This value helps to detect typos. If the actual value converts to a coded value with an absolute value larger than this setting, an error message appears. For instance, say that the factor range is 0-10, which is coded -1 to +1. When a factor level setting of 15 is entered into the Design sheet it gets converted to a coded value of 2; 25 is coded to 4; 30 is coded 5. A coded value of 5 or higher will trigger the error message at the default setting. Increasing the limit will allow for a larger extrapolation range. Stat-Ease does not recommend drawing conclusions from any extrapolated prediction.

Significance Threshold for Power: This sets the alpha-risk level for estimating the power for a factorial design and the evaluation.

Significance Threshold: This sets the alpha risk level for most of the estimates and graphs throughout the program . This is used in places like the ANOVA, designations of significant/non-significant, the LSD bars and confidence bands on the Interaction Graphs, the CI, PI, and TI values on the Point Prediction screen. This setting is not used to estimate power nor algorithmically select effects.

Stepwise Selection Alpha: This sets the default alpha risk level for the algorithmic model reduction methods.

Tolerance Population Proportion: This sets the population’s proportion of interest contained by the tolerance interval. The larger the value the wider the tolerance interval will become.

Coefficient Sum of Squares: The defaults are the generally preferred methods. See Sum of Squares for details.

Split-Plot

Maximum Likelihood Precision: Moving right towards Full will increase the precision of the estimates and increase the computation time. The default “almost full” setting is a good trade-off.

Maximum Iterations: The maximum number to times the ReML and ML algorithms will iterate before stopping even if some improvement is found in the fit of the model.

Information Matrix for F-Test: Choose which Fisher Information Matrix to use for testing the random effects and model coefficients.

Observed: Recommended setting to use for generating the Kenward and Roger estimated F-Test variance-covariance matrix.

Expected: An alternate (not recommended) method.

Default to Numeric Model Selection: Instead of using the half-normal and other graphical model selection tools for factorial designs, use the numerical display and algorithms to select the model for factorial designs.

Click the Defaults button to return to the original installed settings for this screen.