Home / DOE / Create Designs / Modify Design

Modify Design¶

QXL DOE New > Modify Design

The Modify Design menu has thirteen entries. Nine of them open one dialog, titled Modify Design, on the page that matches the entry you chose; four act on the sheet immediately with no dialog. This page covers all thirteen.

Run any of them with the design sheet active. An inactive design sheet is not an error. If the active sheet is not a design sheet, you are asked to pick a design sheet from the workbook, and only told Workbook doesn't have design sheets. when there is none to pick.

Starting from a regression sheet runs the command, but it does not follow the sheet back to the design it came from. A regression sheet is not a design sheet, so you go through that same pick step: with exactly one design sheet in the workbook it is chosen silently, which looks like the design was tracked down, and with more than one you get a window titled Select design sheet., reading The active sheet has to be a design sheet. Select one from the list., and have to pick the right one yourself.

A regression sheet does record which design it was built from, and other commands use that record, but Modify Design asks you to choose before anything reads it. So in a workbook holding several designs, check which one you picked. This is true of all thirteen entries.

The menu's own group headings are Inputs, Outputs, Replicate Format and Randomization.

Menu entry	Opens the page headed
Inputs > Input Name, Low/High, and Type	Enter Factor Names and Coding, or the version that suits the design type
Inputs > Add/Remove Interactions	Change interactions
Inputs > Change Coding (Autocode/Standardize)	Change design coding
Inputs > Add/Remove Center Points	Axial and center points on a CCD, Center points on any other design that allows them
Inputs > Modify Axial Value (CCD Only)	the same page as Add/Remove Center Points
Inputs > Change Blocking/Folding	Enter number of replicates and number of blocks
Inputs > Add/Remove Runs	Change number of rows per replicate
Outputs > Add/Remove, Change Name/Type	Define outputs (responses)
Outputs > Add/Remove Replicates	the same page as Change Blocking/Folding
Outputs > Generate Random Output Data	no page
Replicate Format > Convert From Table To Stacked	no page
Replicate Format > Convert From Stacked To Table	no page
Randomization > Randomize Run Order	no page

Two pairs of entries land on exactly the same page, because one dialog page covers both jobs:

Add/Remove Center Points and Modify Axial Value (CCD Only). On a CCD that page carries both the center point count and the axial value, which is why the two entries cannot be separated.
Change Blocking/Folding and Add/Remove Replicates. Replicates, blocking and, on a two-level factorial, folding are all set on the one page.

The factor page depends on the design type¶

Input Name, Low/High, and Type opens whichever factor editor the design was built with, so what you see depends on the design:

Design type	Page heading
2-level factorial, CCD, Box-Behnken, Plackett-Burman	Enter Factor Names and Coding
3-level and N-level factorial	Enter factor names and factor levels
Taguchi	Factors
Custom	Factors, with the custom design factor editor

See Factor Names and Coding.

Interactions¶

Change interactions has the available interactions on one side and the model's interactions on the other. Drag between the lists, or use Add >> and << Remove. Degrees of freedom remaining: n updates as you go.

The available list is grouped, so terms can be moved wholesale: there are nodes for Quadratic, Custom and n - way interactions. Build custom interactions >> opens the Custom Interactions Builder, where an interaction is entered by typing the coded factor names.

The one check this page makes is on the degrees of freedom. If the terms in the model would use more degrees of freedom than the design has, leaving the page asks Degrees of freedom are less than zero. Would you like to go back and remove some interactions?, with Yes as the default. Yes returns you to the page; No accepts the model as it stands. While the count is negative the page also shows a red label reading You need to remove some interactions.

The hierarchy rules are not checked here. The Hierarchy warnings setting in the Options dialog governs the regression run, not this dialog, so a model that breaks hierarchy can be saved from this page and is questioned later, when the regression runs. See Regression.

Center points and the axial value¶

Which designs allow center points at all is fixed:

Design type	Add/Remove Center Points
CCD, Box-Behnken	allowed
2-level factorial	allowed, as long as at least one factor is quantitative
everything else	refused

A two-level factorial whose factors are all categorical gives Center points are not available when all inputs are qualitative. Any other design type gives Center points are not available for the design type design., naming the type, so Plackett-Burman, Taguchi, 3-level, N-level, Custom and D-Optimal designs all refuse.

Center points asks for one number. Its label depends on the design type and the replicate format, not on how many replicates or blocks the design has:

Design	The label
2-level factorial, Stacked format	Number of center points per block:
2-level factorial, Table format	Number of center points per replicate:
every other design type	Extra number of center points:

Below the box are two total lines, and both update as you type. With a single replicate and a single block they read Total number of center points: and Total number of runs:. Once the design is replicated or blocked they show the arithmetic instead, as Total center points: n replicates * m center points = t (or n blocks * ... in the stacked multi-block case) and Total runs: n replicates * m = t.

On a CCD the same page carries the axial value, with the same four choices the builder offered. See Central Composite Design.

Coding¶

Change design coding offers three radio buttons: Autocode, Standardize and Do not code the design.

Standardize is not always available. If any factor in the design is categorical the radio button is greyed out, and a design that was stored as standardized comes up showing Autocode selected instead.

This is the one modification that changes no data. Only the design's stored coding setting is rewritten. Run the regression again from the design sheet to see the new coding take effect: a regression started from an existing regression sheet uses the coding recorded on that sheet, not the setting you just changed. See Design Coding.

Runs, replicates and blocks¶

Change number of rows per replicate takes one number, in a box labelled Enter the number of runs per replicate:. Use it to extend a design with extra rows or to trim it. The entry must be a whole number greater than 2, otherwise Finish is blocked with Number of runs must be an integer greater than 2.

Adding or removing runs is one-way

Once a design carries added or removed runs, the count is stored with the design and any later modification that has to rebuild the design is refused. That covers changing a factor's low, high or levels, changing folding, and changing blocking or replicates. The refusal is Design cannot be modified when it has runs added or removed., or, when the design has more than one block, The current design is blocked. Adding or removing runs would change the blocking. Either remove blocking from the design or recreate the design from scratch. Recreating the design from scratch is the only way back.

Enter number of replicates and number of blocks is the page described in Define Replicates and Blocks, reached from either Add/Remove Replicates or Change Blocking/Folding.

Outputs¶

Add/Remove, Change Name/Type opens Define outputs (responses), so outputs can be added, renamed or retyped after the design exists. The weights or events and sample size columns can be turned on here only when the design is already in stacked replicate format; in table format the checkbox is greyed out and the note Weights and sample size columns are only available with stacked replicates appears instead. The Design format choice itself is hidden on this version of the page, so the format cannot be changed from here: use Modify Design > Replicate Format for that. See Define outputs.

The four commands with no dialog¶

Generate Random Output Data fills the response columns with made-up values, which is how you get a design sheet you can practise the analysis on before the real data exists. What it writes depends on the output's type, and it is not numbers in every case:

Output type	What goes in the column
Quantitative	whole numbers from 100 to 199. A weights column gets 1 to 99
Binary entered as one column	the words Male and Female, picked at random per row
Binary with events and sample size	events 10 to 89, sample size 100 to 199
Nominal	weekday names, Monday through Sunday. A weights column gets 10 to 99

It asks first only if the outputs already hold data: Outputs already contain data. Do you want to overwrite it with random data?, Yes or No. On empty output columns it writes without asking.

Randomize Run Order shuffles the rows of the existing design sheet. It asks nothing and acts immediately.

Convert From Table To Stacked and Convert From Stacked To Table rewrite the sheet in the other replicate format. They also act immediately, unless the design is already in the format you asked for, in which case nothing happens and you get Design is already in table format. or Design is already in stacked format. Note that the weights and events and sample size columns exist only in the stacked format, so converting to table drops them. See Stacked vs. Table.

After a modification¶

Any of these changes the design sheet, not the regression sheet. A regression sheet and the charts built from it are not updated by a design change, so run the regression again after modifying a design, and rebuild any chart that came from the old regression sheet. See Chart Source.