The MVPDIAGNOSE Procedure
The MVPDIAGNOSE procedure accepts a primary input data set that has one of the following two types:
-
a DATA= data set that contains new process data to be analyzed by using an existing principal component model (Phase II analysis)
-
a HISTORY= data set that contains process data and the accompanying scores, residuals, and statistics that are produced by using a principal component model. The process data can be the original data that were used to create the model (Phase I analysis) or subsequent data that were analyzed by using a previously created model (Phase II analysis)
These options are mutually exclusive. If you do not specify an option that identifies a primary input data set, PROC MVPDIAGNOSE uses the most recently created SAS data set as a DATA= data set.
When you specify a DATA= data set, you must also specify a LOADINGS= data set that contains principal component loadings and other information that describes the principal component model. When you specify a HISTORY= data set, you must also specify a LOADINGS= data set if you use a CONTRIBUTIONPANEL or CONTRIBUTIONPLOT statement and specify the TYPE=TSQUARE option.
A DATA= data set provides the process measurement data for a Phase II analysis. In addition to containing the process variables, a DATA= data set can contain the following:
When you specify a DATA= data set, you must also specify a LOADINGS= data set that contains the loadings for the principal component model that describes the variation of the process. These loadings are used to score the new data from the DATA= data set. The process variables in the LOADINGS= data set must have the same names as those in the DATA= data set.
A HISTORY= data set provides the input data set for a Phase I or Phase II analysis. In addition to containing the original process variables,
a HISTORY= data set contains principal component scores, residuals, SPE and 
statistics, and a count of the observations that are used to construct the principal component model. These variables are
summarized in Table 11.5.
Table 11.5: Variables in the HISTORY= Data Set
|
Variable |
Description |
|---|---|
|
Prin1–Prinj |
Principal component scores |
|
R_ |
Residuals |
|
_NOBS_ |
Number of observations used in the analysis |
|
_SPE_ |
Squared prediction error (SPE) statistic |
|
_TSQUARE_ |
|
The score variables names must consist of a common prefix followed by the numbers 1, 2, …, j, where j is the number of principal components. By default, the common prefix is Prin. You can use the PREFIX= option to specify another prefix for score variables.
If the number of principal components is less than the total number of process variables, the HISTORY= data set should also
contain residual variables. A residual variable name consists of a common prefix followed by the corresponding process variable
name. The default residual variable prefix is R_. For example, if the process variables are A, B, and C, the default residual variable names are R_A, R_B, and R_C. You can use the RPREFIX= option to specify a different residual variable prefix.
Note: Usually you create a HISTORY= data set by specifying the OUT= option in the PROC MVPMODEL statement or the OUTHISTORY= option in the PROC MVPMONITOR statement. If the PREFIX= or RPREFIX= option is used when such an output data set is created, you must specify the same prefixes to identify the score and residual variables when you read it as a HISTORY= data set.
A LOADINGS= data set contains the following information about the principal component model:
-
eigenvalues of the correlation or covariance matrix used to construct the model
-
principal component loadings
-
process variable means used to center the variable values
-
process variable standard deviations used to scale the variable values
You can produce a LOADINGS= data set by using the OUTLOADINGS= option in the PROC MVPMODEL statement. Table 11.6 lists the variables that are required in a LOADINGS= data set.
Table 11.6: Variables in the LOADINGS= Data Set
|
Variable |
Description |
|---|---|
|
_VALUE_ |
The value contained in process variables for a given observation |
|
_NOBS_ |
Number of observations used to build the principal component model |
|
_PC_ |
The principal component number; 0 for the observation that contains eigenvalues |
|
process variables |
Values associated with the process variables |
Valid values for the _VALUE_ variable are as follows:
- EIGEN
-
eigenvalues from the principal component analysis
- LOADING
-
principal component loadings
- MEAN
-
process variable means
- STD
-
process variable standard deviations
The LOADINGS= data set contains one EIGEN observation and j LOADING observations, where j is the number of principal components in the model. The presence of a MEAN observation indicates that the process variables were centered when the principal component model was built, and the presence of a STD observation indicates that the process variables were scaled when the principal component model was built. The means and standard deviations are used to center and scale new data in a Phase II analysis.