3. Reports and Graphics › 3.3 Statistical Analyses › 3.3.6 Correlation Report

3.3.6 Correlation Report

Correlation measures the closeness of a linear relationship
between two variables.  The Correlation Report helps you
identify relationships between measurement items by showing
the correlation coefficients among a group of selected
variables.  It is useful in initial data analysis.

The report can be extended to up to five variables correlated
against five different variables.  Statistics that you can
request include the following:

o  Pearson correlations
o  Spearman coefficients
o  Kendall tau-b coefficient
o  Hoeffding's D statistic
o  Descriptive statistics
o  Significance probabilities
o  Sums of squares cross product matrices
o  Covariance matrix

You can specify one of four variance calculations, whether or
not to accept missing values, and whether you want a standard
correlation matrix or a ranked (best N fits) correlation
matrix.

For example, you might perform a correlation analysis against
transaction rate and response time for multiple CICS regions
to identify regions whose transaction volume impacts response
time in other regions.

Correlation reports can help you determine what, if any,
relationship exists between work measures and response times.
In a tuning study, a correlation report can help you
determine where to concentrate efforts to improve response
times.  For example, Figure 3-19 shows two sample correlation
reports.  Both correlate the following:

o  Average concurrent CICS users (CSYAVUSR)
o  Average CPU time per CICS transaction (CSYAVCPU)
o  Percent CPU usage by CICS (CSYPCCPU)
o  Average short CICS response time (CSYAVSTM)
o  Percent short CICS responses less than or equal to
   response threshold 1 (CSYPCSR1)
o  Average total CICS response time (CSYAVTTM)
o  Percent total CICS responses less than or equal to
   response threshold 3 (CSYPCTR3)

Observations are weighted by the number of CICS transactions
(CSYTRANS).  The response time thresholds are user-defined.
For example, response threshold 1 might be 1 second and
response threshold 3 might be 5 seconds.

The first sample in Figure 3-24 shows a standard correlation
report.  The second sample illustrates the same analysis
using the rank option to show only the best three
correlations for each independent variable.

To create a correlation report, select the Correlation Report
option from either of the following menus:

o  Direct Inquiry Statistical Analysis Menu
o  Structured Inquiry Statistical Analysis Menu

PEARSON CORRELATION COEFFICIENTS
/ PROB > |R| UNDER H0:RHO=0 / NUMBER OF OBSERVATIONS / FREQ VAR=CSYTRANS

                                  CSYAVSTM CSYPCSR1 CSYAVTTM CSYPCTR3


        CSYAVUSR                  -0.04225 -0.51245  0.02830 -0.34808
        AVG. CICS USERS             0.0108   0.0001   0.0878   0.0001
                                     65732    65732    65732    65732

        CSYAVCPU                   0.27517 -0.35798  0.26270 -0.13491
        AVG. CPU TIME/TRANSACTION   0.0001   0.0001   0.0001   0.0001
                                     65732    65732    65732    65732

        CSYPCCPU                  -0.02321 -0.04564 -0.04694 -0.01163
        PCT CPU USAGE               0.1617   0.0059   0.0046   0.4832
                                     65732    65732    65732    65732


                               Sample Standard Correlation Report




PEARSON CORRELATION COEFFICIENTS
/ PROB > |R| UNDER H0:RHO=0 / NUMBER OF OBSERVATIONS / FREQ VAR=CSYTRANS

            CSYAVUSR   AVG. CICS USERS

             CSYPCSR1 CSYPCTR3 CSYAVSTM
             -0.51245 -0.34808 -0.04225
               0.0001   0.0001   0.0108
                65732    65732    65732

            CSYAVCPU   AVG. CPU TIME/TRANSACTION

             CSYPCSR1 CSYAVSTM CSYAVTTM
             -0.35798  0.27517  0.26270
               0.0001   0.0001   0.0001
                65732    65732    65732

                                 Sample Ranked Correlation Report

Figure 3-24.  Sample Correlation Reports