Design-Expert v22.0

Design-Expert » Transformation

Transformation

Unless previous knowledge supports applying a transformation, leave the Transformation set to “None”. Come back to the Transformation tab if there are indications in the analysis and diagnostics that a data transformation is required.

  • Towards the bottom of this screen, look at the ratio of the maximum to minimum response - larger ratios indicate a higher likelihood that a transformation may help the analysis.

  • Once a transformation is selected use the Constant “k” entry box to add a constant to the response data. Transformations require all positive data.

  • Most of the time a common Power Law transformation will be appropriate. Power law transformations raise the response data to the specified power. The one labeled Power can be used to pick a custom exponent.

  • The last two options are non-power law transformations. They will not be suggested by the analysis or diagnostics.

    • Use ArcSine Square Root for proportion data between 0 and 1 (eg. Fraction defective).

    • Use Logit for data that approaches but never quite reaches boundary conditions (eg. yield, shrinkage). The boundary conditions must be entered.

Options Button:

Fit Summary

Max Process Model Order: For very large designs it can take a great deal of time to compute all the possible model orders for the fit summary. This option can be used to limit the computations to lower-order models to save time. If higher-order models are necessary, they can still be fit on the model screen.

Analysis

Coding for Analysis: Coding for Analysis: For most analysis use the Coded setting. The Actual setting computes the least squares fit using the factor settings as entered. Advanced modeling techniques require fitting a model in the actual scale or real scale for mixtures.