Home / Statistical Tools / MSA / Attribute MSA: Crosstabulations Method / Options

Options¶

Every control on the Attribute MSA dialog. It has three tabs, Data, Options and Gage Info.

Worked example

The How-To page runs a complete study from sample data and prints the numbers you should see.

Data tab¶

The whole tab is the column selector. On a run started from an MSA template this tab is hidden, because the template has already said which column is which.

Two radio buttons at the top choose where the data come from:

Excel reads the range you selected on the worksheet.
GroupBy reads the same range but splits it into groups first, and runs a separate complete study for each group.

Data Types¶

Each column carries a data type shown beside its name. Rating Columns and Standard (Optional) will take a column of any type except DateTime. Part and Appraiser will take a column of any type at all, including DateTime, because they are only labels.

Rating Columns¶

What the appraiser decided. Tick one or more; each ticked column becomes its own separate study with its own summary, its own charts and its own tables.

Part¶

Which part was being judged. One column, required.

Appraiser¶

Who judged it. One column, required.

Standard (Optional)¶

The known right answer for each part, spelled with the same labels the ratings use. Without it you still get Within Appraiser and Between Appraisers; with it you also get everything that compares an appraiser against the right answer.

Every row of a given part must carry the same standard value. A part with two different standards stops the study.

The four roles must be four different columns¶

A column claimed by one role is greyed out in the other lists, showing which role took it. The dialog will not run until the required roles are filled with different columns.

GroupBy¶

In GroupBy mode, the columns you move into GroupBy Order split the data into groups, and each group becomes a study of its own.

Options tab¶

Data Type¶

Three radio buttons, and which of them you may pick depends on how many distinct levels your data actually holds. The levels counted are those in the Rating columns and the Standard column together.

Binomial (two levels)

: Available only when the data holds exactly two levels.

Nominal (two or more unordered levels)

: Always available. The fallback the study uses when nothing else fits.

Ordinal (three or more ordered levels)

: Available only when the data holds three or more levels. Adds the two Kendall tables and enables the Levels block, because an ordinal scale needs an order.

A line under the buttons says why a choice is unavailable, in one of these words:

Ordinal needs three or more levels; this data holds two.
Binomial needs exactly two levels; this data holds 4.
Binomial and Ordinal need every Rating column to hold the same kind of level count; these columns differ.

The last of those is what you see when you tick two rating columns, one with two levels and one with five. They are separate studies, but the data type is one setting for both, so only Nominal can describe them together.

Binomial and Nominal are the same study

They do not differ in what they compute, and Binomial does not unlock the Assessment Disagreement table. That table appears whenever the data holds exactly two levels and a standard was given, whichever of the two you picked. The only thing your choice changes is the Data Type line in the report's Study Summary, which reads Binomial when you chose it and the data really does hold two levels, and Nominal otherwise.

So choosing Binomial on data that turns out to hold three levels does not fail. The study runs as nominal and prints a note saying so in the Study Summary block.

Only Ordinal changes the arithmetic, because it is the only choice that says the levels have an order.

Levels¶

Lists every distinct level found in the Rating and Standard columns together, in the order the study will use, with a caption saying how many were found. A level that appears in the standard but was never given by any appraiser is marked (standard only).

Each row carries Move Up and Move Down buttons. They are available only when the data type is Ordinal and the levels are text, because ordering is the only thing they can affect and ordering only means something on an ordinal scale.

Two cases grey the buttons even at Ordinal:

Numeric levels are ordered by value. Levels that all read as numbers sort themselves, and a note under the list says so.
There is only one level to order.

For text levels at Ordinal with no order set, the study uses alphabetical order, silently. There is no note on the report to tell you it did. Alphabetical rarely means what you want on an ordinal scale, so set the order rather than waiting to be warned.

A note is printed in one narrower case: when a saved level order exists but no longer covers the levels found in your data, so alphabetical order had to be used in its place.

Conforming Level¶

Conforming level: names which of the two standard values counts as the good one. It is available only when a Standard column is selected and that column holds exactly two distinct values, because the idea has no meaning otherwise.

It decides the Conformance table: which errors are false alarms, a good part called bad, and which are misses, a bad part called good.

The dialog guesses the answer from the words themselves. It recognises the usual conforming words (pass, accept, conforming, ok, good, yes, true, go, and their short forms) and the usual nonconforming ones (fail, reject, nonconforming, bad, no, false, defective, and theirs). When neither list settles it, it takes the first of the two in level order. When the guess was used rather than a choice of yours, the Study Summary block says so and tells you what the other level was.

Study Options¶

Alpha:

: The significance level, which sets the width of every confidence interval on the report. Beside the box, Confidence level: 95% shows what your alpha comes to. Alpha must be greater than zero and less than one.

Gage Info tab¶

Seven free text boxes recording what the study was of and who ran it: Gage name:, Gage No.:, Gage type:, Part name:, Part No.:, Date: and Performed by:.

Nothing here is parsed and nothing here changes a number. All seven are written to the Gage Information block at the top left of the report, and all seven rows are printed whether or not you filled them in. On the finished sheet these seven value cells are left unlocked, so you can still type into them after the sheet is protected.

Save as Default Settings¶

Save as Default Settings stores two settings only: alpha and the data type. They are read back the next time you open the dialog, on this machine, for every attribute study.

Nothing else is saved, and that is deliberate. The level order, the conforming level, the column choices and the seven gage details belong to one study, not to the next one. They are stored on the worksheet instead, so Modify brings them back for that study.

A saved data type that the next study's data cannot support is quietly dropped rather than forced. A saved Ordinal opening on two-level data loads as Nominal.

What stops the dialog¶

These appear in red on the Data tab and hold Finish until you fix them.

Message	What caused it
A Rating column, a Part column and an Appraiser column must each be selected.	One of the three required roles is empty
The Rating, Part, Appraiser and Standard columns must be different columns.	Two roles are pointed at one column
The Rating column has more than 100 distinct values. An attribute study rates parts on a small number of levels.	A rating column looks like measurements, not categories
The Standard column has more than 100 distinct values. An attribute study rates parts on a small number of levels.	The same, on the standard column
The Rating column Rating holds one value, P. An attribute study needs at least two distinct values, such as a pass and a fail.	Every rating is the same, so there is no disagreement to measure

The last of those also appears as a message box, titled Attribute MSA, before the dialog opens at all. A one-value rating column beside a two-value standard is fine and is not refused: the appraisers agreed with each other and were wrong half the time, which the study can say something about.

When a study is refused¶

These stop one study after the dialog has closed. The sheet prints the heading This study could not be computed and one of these sentences, then carries on with the other studies.

Sentence on the sheet	What caused it
Alpha must be greater than zero and less than one.	Alpha: is outside that range
This study needs at least 2 parts with ratings.	Fewer than two parts survived
This study needs at least 1 appraiser with ratings.	No appraiser had any ratings
This study needs at least 1 trial with ratings.	No trial had any ratings
A rating is missing for part 7, appraiser Dana, trial 2. Every appraiser must rate every part on every trial.	A hole in the grid
The standard is missing for part 7. Give every part a standard, or leave the Standard column empty for every part.	A partly filled standard column
Ordinal data needs at least three distinct values in the ratings and the standard together. This study has 2. Choose Binomial or Nominal for two-level data.	Ordinal chosen on two-level data
Every appraiser must rate every part the same number of times. Appraiser Dana rated part 7 3 times where 2 were expected.	An unbalanced study
Part 7 has more than one standard value. Every row of a part must carry the same standard.	Two different standards given for one part

Two more messages replace the whole worksheet rather than one study:

No data was found for analysis. when the run produced no studies at all.
Too many results. Please reduce the number of groups. past 100 studies. Refused studies count.