Design & Analysis of Experiments Startup Panel - Advanced Tab

Select the Advanced tab of the Design & Analysis of Experiments Startup Panel to access the design of experiments described here. See also Experimental Design and Analysis - Index.

2**(K-p) standard designs (Box, Hunter, & Hunter)

Select 2**(K-p) standard designs (Box, Hunter, & Hunter) to generate or analyze standard fractional factorials and full factorials with two levels (and, optionally, center points), with or without blocking (see, for example, Box, Hunter, & Hunter, 1978, Montgomery, 1991). For a description of these designs, refer to the Introductory Overview.

2-level screening (Plackett-Burman) designs

Select 2-level screening (Plackett-Burman) designs to generate or analyze designs for screening a large number of two-level factors (optionally, with center points). Statistica generates and analyzes Plackett-Burman (Hadamard matrix ) designs as well as saturated fractional factorial designs with up to 127 factors. For a description of these designs, refer to the Introductory Overview.

2**(K-p) max unconfounded or min aberration designs Select 2**(K-p) max unconfounded or min aberration designs to search for and analyze the 2(k-p) fractional factorial design of up to 128 factors or 2048 runs, with properties that best suit your needs. A variety of options and criteria are available for specifying the design search. For a general description of 2(k-p) designs, refer to the Introductory Overview, and for a description of 2(k-p) design search procedures, refer to the Overview of 2(k-p) maximally unconfounded and minimum aberration designs.

3**(K-p) and Box-Behnken designs Select 3**(K-p) and Box-Behnken designs to generate and analyze designs with 3-level factors. Specifically, Statistica generates Box-Behnken designs as well as blocked 3(k-p) full and fractional factorial designs as enumerated by Connor and Zelen (see McLean and Anderson, 1984) for the National Bureau of Standards of the U.S. Department of Commerce. In the analysis, the main effects and interactions can be partitioned into linear and quadratic components. For a description of these designs, refer to the Introductory Overview.

Mixed 2 and 3 level designs

Select Mixed 2 and 3 level designs to generate and analyze designs with 2 and 3 level factors. Specifically, Statistica generates full and fractional factorial designs as enumerated by Connor and Young (see McLean and Anderson, 1984) for the National Bureau of Standards of the U.S. Department of Commerce. In the analysis, the main effects and interactions can be partitioned into linear and quadratic components. For a description of these designs, refer to the Introductory Overview.

Central composite, non-factorial, surface designs Select Central composite, non-factorial, surface designs to generate or analyze first- and second-order central composite (response surface) designs. For a description of these designs, refer to the Introductory Overview.

Latin squares, Greco-Latin squares

Select Latin squares, Greco-Latin squares to generate or analyze Latin-squares, Greco-Latin squares, and Hyper-Greco Latin squares designs. For a description of these designs, refer to the Introductory Overview.

Taguchi robust design experiments (orthogonal arrays)

Select Taguchi robust design experiments (orthogonal arrays) to generate orthogonal arrays for Taguchi robust design experiments. Experiments with up to 65 factors and 100 runs can be analyzed with this procedure. When analyzing data, Statistica automatically converts the data to the selected signal-to-noise (S/N) ratios for the analysis. The procedure also analyzes categorical frequency data (accumulation analysis) as well as user-defined S/N ratios. For a description of robust design techniques, refer to the Introductory Overview.

Mixture designs and triangular surfaces

Select Mixture designs and triangular surfaces to generate or analyze experiments for mixtures, where the sum of the component settings must be constant (e.g., 100%). Statistica generates the standard simplex-lattice and simplex-centroid designs, and can handle lower bound restrictions on the components (for lower and upper bound restrictions, select Designs for constrained surfaces and mixtures, see below). The results can be computed for the original component settings as well as the pseudo-component. For more information concerning these designs, refer to the Introductory Overview.

Designs for constrained surfaces and mixtures

Select Designs for constrained surfaces and mixtures to find vertex and centroid points for constrained surfaces and mixtures, following an algorithm suggested by Piepel (1988) and Snee (1985). Statistica can process lower and upper bound constraints for mixtures, and/or linear constraints of the form A1x1+...+Aixi+A0 ³ 0, defining a convex hyperpolyhedron. The algorithm is described in the Introductory Overview.

D- and A- (T-) optimal algorithmic designs Select D- and A- (T-) optimal algorithmic designs to select from a candidate list of points a fixed number of runs, so that they (potentially) extract the maximum amount of information, given the respective model that is to be fit to the data. You can choose both the D (determinant of X'X) optimality criterion as well as the A (trace of (X'X )-1) optimality criterion. Available search (optimization) algorithms include the sequential or Dykstra method, the simple exchange (or Wynn-Mitchell) method, the DETMAX algorithm, the Fedorov simultaneous switching method, and a modified Fedorov simultaneous switching method. These techniques are discussed in the Introductory Overview. Note that you not only can generate designs from candidate lists, but also "repair" existing designs via these methods by forcing existing experimental runs into the final selection.

D-optimal split plot design Select D-optimal split plot design to generate a D-optimal split design. Statistica can generate split plot designs for multiple easy and hard to change factors and covariates. This flexible design generation is based upon minimizing the volume of the joint confidence region of the parameter estimates.  Options for generating design syntax for subsequent GLM analyses as well as options for saving Variance Estimation and Precision designs to the design spreadsheets make it easy to analyze these designs once the experiment is performed and data is collected.

D-optimal split plot analysis Select D-optimal split plot analysis to analyze a D-optimal split design. By default, Statistica will analyze the split plot design using the Variance Estimation and Precision module. The Variance Estimation and Precision module is a powerful analytic tool that enables you to analyze the split plot design in the presence of both the whole plot and sub plot error. If the Variance Estimation and Precision module is not available, Statistica will analyze this design using the GLM module. For more information on the difference between GLM and Variance Estimation and Precision, see Variance Estimation and Precision vs. GLM for more details.

Experimental Design Builder. Select  Experimental Design Builder to generate an optimal design flexible to fit the needs of your experiment. You can tailor the design to include any mix of continuous and categorical predictors. Linear constraints and factor combination restrictions can also be specified. D and I Optimal designs are supported.

Full factorial design. Select Full factorial design to generate a full factorial design, which is an efficient type of design that allows a researcher to investigate numerous factors simultaneously. In a full factorial design, the effects of every combination of each level of each factor are studied. This type of design allows a researcher to investigate interactions of factors.

Note: most of these designs can also be analyzed via General Linear Models (GLM) or General Regression Models (GRM), or via (nonlinear) Generalized Linear/Nonlinear Models (GLZ).