7. UNIVARIATE MODEL FORECASTING › 7.3 Standard Output › 7.3.1 Model Analysis Report

7.3.1 Model Analysis Report

The Model Analysis Report provides a summary of the
regression analysis that was conducted for the particular
model that you specified.  Figure 7-1 illustrates the Model
Analysis Report for a study that relates channel SIO counts
to the passage of time.  You can use this report to evaluate
whether or not the model produced by Univariate Model
Forecasting acceptably represents the historical data series.
This decision is based on the r-squared and F values
discussed in Section 7.2.

                                          CA MICS Capacity Planner

                                   ANALYSIS OF UNIVARIATE FORECASTING MODEL

               MODEL OF: CHAN5CNT
               BASED ON: LINEAR
                                                        ------------TIME ELEMENTS------------
                 R**2        F        P     INTERCEPT     NAME    COEFFICIENT     T       P
               --------   -------   -----   ---------   --------  -----------  -------  -----
                 0.75      30.84    .0004    9736797    LINEAR        587938     5.55   .0004

 Figure 7-1.  Model Analysis Report

The Model Analysis Report shown in Figure 7-1 presents the
following information:

MODEL OF:    The data element name to be forecast.

BASED ON:    The time elements (linear, quadratic and/or
             cubic) that you used to develop the model.

The first four columns of the report are discussed below:

R**2:        The r-squared value for the model.  We recommend
             R-squared values of 0.70 and above for accepting
             models that are produced using Univariate Model
             Forecasting.

F:           The F statistic for the model.  As we discussed
             in Section 7.2, the F statistic is calculated
             from the ratio of sums of squares of the model.
             Basically, larger F values indicate a more
             reliable model.  The F statistics may be
             unreliable for models developed with a small
             number of points (that is, less than 30).

P:           The probability that the model proposed is
             significant (that is, more reliable than simply
             forecasting the average value of the historical
             observations).  Although most statistics texts
             recommend testing the null hypothesis to a 0.001
             level, 0.01 has been recommended for models
             based on computer measurement data (SAR79).

INTERCEPT:   The mean value of y when x is zero.

The next four columns are produced for each time variable
that is used in the model. (For example, if a model specified
that CPUHRS is to be forecast based upon linear, quadratic
and cubic, time elements, then three lines would appear under
the heading TIME ELEMENTS.)  These columns contain the
following information:

NAME:        The name of the time element to be analyzed.
             This name can be either LINEAR, QUADRATIC, or
             CUBIC.

COEFFICIENT: The coefficient for the time element in the
             regression equation.

T:           The T statistic.  The T statistic is similar in
             nature to the F statistic.  Larger values
             indicate "goodness."  Note that the value may
             be misleading for models that are based on small
             samples.

P:           The probability that the time element you
             specified makes a significant contribution to
             the model to be developed.  Once again, we
             recommend a 0.01 acceptance value.