This is a minor maintenance release that adds only one new feature: the ability to do one factor regressions. The remainder of the modifications focus on improvements to the setup kit and problems that some people have run into with two-factor experiments. In addition, this version eliminates the problems that some customers have experienced with generating interaction designs and withholding observations.
The primary feature of this upgrade is a brand new setup kit. The new setup utility will automatically make copies of all files replaced, a feature that has become increasingly necessary because of the rapid changes in key Windows components.
In addition, the new version features the ability to generate scatter plots directly from the XY List, elimination of unwanted duplicates in the GridSearch results, some additional fine-tuning of the contour plots features, and a significant upgrade to the help system.
Major Update!
Now you can apply GridSearch constraints to your contour plots! This is a must-have feature if you are doing experiments involving mixtures or discrete factors with more than two settings. Just define your GridSearch first, then check Use GridSearch Constraints on the Special tab of the Contour Plots Definition Form.
Note that we have modified the interface to use tabbed dialog boxes. This should provide a visual overview and it will simplify navigation within major components of the program.
We have also added the ability to change the number of contours and the ability to choose your own custom colors.
We have fine-tuned many elements of the program. Multiple linear constraints are now correctly applied in all cases, and a variety of nagging printing problems have been tracked down and eliminated. The regression output is more complete. The error message is no longer generated when you try to build contour plots for two-factor experiments.
Known problem in this release: The regression appears to correctly accept a prior s and degrees of freedom for each response, but the only numbers used are for the first response. This will be corrected soon in a maintenance release.
Major Update!
New features include unlimited additional factors for GridSearch and printing for multiple charts including the ability to impose a common color scale over a bunch of related plots.
Adding multiple additional GridSearch factors will primarily interest two groups: those doing experiments that involve multiple mixtures, and those doing experiments with Hardin-Sloane Optimal (I) designs (as generated by Gosset).
The ability to impose a common color scaling over a set of related plots is an important tool because it provides a perspective that is simply not available when the charts are scaled individually. The plots lose some detail when you do this because there are only so many colors that you can really tell apart so there is a practical limit to how many charts you want to scale as a group. The program also provides the capability to print multiple plots with a single command, a feature that has been much requested.
Fixes include the contour printing problem. The transformation functions in the spreadsheet have been made more reliable, and the overlap problem with the charts has been dealt with.
Dozens of minor conveniences have been added and the help files have been updated again.
Note that Dave has contributed a new article on optimal designs that you can get to through the STRATEGY for Windows frequently asked questions page.
The most obvious improvement is in the contour plots program. We've made it a lot more intuitive to create and view the graphs. You can now also save multiple sets of graphs for the same experiment, and you can tell which one is which without having to display each individual graphs.
A couple of irritating bugs have also been eliminated. On some systems, the program would crash if you tried to run the regression directly after creating the model. That should no longer happen in any environment. Other people reported problems with getting the correct contour plot. That problem has also been eliminated.
Finally, the continuing effort to upgrade the help files and make them more reliably available has taken a giant step forward. Dave has completed a thorough review of all of the help topics, and his suggestions have all been implemented. We also believe that we have solved the problem that arose if you installed STRATEGY into a folder whose name had a space in it.
There is also one piece of bad news to report. Our efforts to become NT 3.51 compatible have suffered a major setback. It still appears that we have some major issues to resolve, and this is an operating system that is disappearing more rapidly than we anticipated. We have temporarily put this on hold, and it will probably remain there unless one of you raises it as a priority.
The three most significant features of this upgrade are a much improved utility for displaying and printing reports, resizeable forms (where it seemed to make sense), and a major improvement to the help files.
When we asked users of STRATEGY for DOS what they liked least about the program, the answer almost unanimously was printing the regression output. It sadly was not much better in the new program, but it is much easier and much quicker in the new program. In addition, loading and unloading files is much quicker.
The new text editor is one of the several forms that have been made resizeable. Others include the GridSearch Sweet Spots form, the design list spreadsheet, the linear constraints form, and the contour plots view form.
In addition to the significant review and upgrade of the help files, we have changed the way they are called to meet Microsoft's new recommendations. They tell us that this will eliminate the "Help File not found" message that some of you have been getting.
Problems eliminated in the new version include the problem of GridSearch crashing when no Sweet Spots were found and the mislabeling of the remaining factors on the contour plots. The program now remembers the model type chosen when you reload the model file, and it remembers whether or not you suppressed any terms. You can now specify remaining factor levels to two decimal points, and you can enter decimal points on the Prior S form.
You can now go back to the Sweet Spots form after you close it. Start GridSearch. When the GridSearch Start Form loads, choose Review Last Search from the File Menu. The GridSearch Sweet Spots form will be loaded with the last successful GridSearch results. If there are no previous successful results, the most recent GridSearch Summary Report for that experiment will be loaded in the text editor.
Note this is the first version that has run successfully on Windows NT 3.51, but the testing is far from complete.
This is a minor upgrade unless you do mixture designs. We have added the ability to generate Simplex Lattice Designs (STRATEGY for Windows could already analyze these designs, of course). If you do mixture designs, you should definitely plan on downloading this version as well as reading Dave's notes on how to handle mixture designs in STRATEGY for Windows.
There are also some significant minor fixes. The check for replicates now excludes any rows where the response values are not provided. You can now enter decimal numbers for both prior s and degrees of freedom.
GridSearch is a little friendlier. In particular, the program doesn't blow up when you enter illegal constraints, it correctly handles omitted factors, and the constraints are described more clearly in the summary output.
Contour Plots now correctly handle changes to the levels for the last remaining factor.
Major Upgrade featuring all new graphics, ability to reload an entire project with a single command, better handling of left-out factors in GridSearch, and literally dozens of minor improvements. Speed throughout has been improved and the program is shrinking somewhat in size despite the addition of many new features.
Note that there are still several known kinks in the new graphics program. In particular, Print As Shown prints a very tiny image of the graphic, and neither the remaining factor levels nor the confidence limits appear on the current print out. We recommend either using the full page print scale or copying the charts to another application for printing at the present time.
The program is more forgiving of variances when importing files from spreadsheets or other tools. The problem with the suppressed terms form has been repaired. Major additions and updates of the help files are included, and the GridSearch forms show the result of Dave's attention.
All contour plots features are now working, and we've added the ability to set limits on the primary axes. GridSearch allows stopping by resolution (Quarters, Eighths, etc.). Improved error trapping added.
Hardin-Sloane I-Optimal cubical and spherical designs can now be generated for 2-7 factors (almost any size design can be analyzed).
The new version already includes a much faster version of the graph viewer. The problems with scaling and orientation of the graphs have also been eliminated since the last beta version. The levels for remaining factors now show on the 3-D surface graph view.
Among the added user conveniences is the ability to save any of the graph views to bitmap files so that they can be included in presentations and reports. Help for the graphing form and the new Hardin-Sloane Design form is now included. We have adopted a more consistent use of common dialog boxes for open and save routines as many of you requested.
Known problems: Wordpad is not found in all cases on NT 4.0. NT 3.51 does not include Wordpad.
The new regression engine has been integrated into STRATEGY for Windows. Example files for Contour Plots were upgraded.
The major change is a more seamless integration between the front end and the STRATEGY engine. Three of the four pieces of the STRATEGY engine are now so integrated. The new regression engine will be on-line by the next release. These new components are not only more aesthetic, but computer processor-intensive functions like GridSearch should run much faster and take better advantage of all of the memory you have.
The interim regression engine allows you suppress any of the parts of the output that you do not need.
Printing is now available from the STRATEGY Spreadsheet. At the present time, it only prints to the default system printer so you must change your default if you wish to print to another printer.
The problems that GridSearch had with some models has now been corrected. GridSearch should work correctly with all models now.
The Contour Plots are almost ready, and they are ready enough so that you can try them out. Contour Plots now offer Surface Graphs, Color Contour Plots, and the cubic landscapes available in the old STRATEGY program. Note that our graphing form (Contour Plots, View) is in an early stage of development and we know that it is really slow. A primary focus of our production fine-tuning centers around the performance of this part of the program.
Many user conveniences have been added. Chief among these is that you now have the option to immediately open any of the files you have created with a single mouse-click or keystroke. You will also see that the program is better at remembering your previous choices if you are simply making an adjustment to an experiment. Editing in the STRATEGY Spreadsheet has also been improved.
The help files have been significantly updated. Context-sensitive help is now available from every screen except the new graph forms.
After a year and half of development, STRATEGY for Windows is released to a small core of our most loyal and motivated customers. Although the contour plot portion of the program is not yet ready and we have not solved all of the printing problems, most of the program has reached the "expert-friendly" stage. We now need customers to provide feedback to make sure that the development is in synch with their needs.