17.01.2015 Views

GPS-X Technical Reference

GPS-X Technical Reference

GPS-X Technical Reference

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

405 Optimizer<br />

To check for violation of the independence assumption, the standardized residual plots<br />

should be examined for any noticeable trends. The presence of trends provides evidence<br />

of serial correlation. Serial correlation occurs when residuals taken in sequence are<br />

correlated with each other.<br />

To check the assumption that the errors are normally distributed, the standardized<br />

residual plots should be examined to see if the residuals are randomly scattered about<br />

zero. In addition, the values of the standardized residuals should be examined to check<br />

whether approximately 95 percent of the residuals lie between +2 and -2.<br />

The standardized residual plots are scaled to fit in the Log window. The largest<br />

standardized residual for each response variable is represented by the maximum number<br />

of " * " characters, which is seven. The remaining standardized residuals for a response<br />

variable are scaled relative to this maximum residual for the purposes of calculating how<br />

many " * " characters to print. You should consult the actual values of the standardized<br />

residuals provided in the Log window to determine whether the actual magnitudes of the<br />

residuals for a response variable are large.<br />

If the assumption that the residuals are independent and normally distributed is violated it<br />

provides evidence that our model is inadequate to represent the experimental data. If the<br />

process model structure is correct, it should account for the non-random variability in the<br />

data.<br />

The assumption that the residuals should be independent if the measurement errors are<br />

independent is not strictly correct but is fine for practical purposes. The residuals are<br />

always correlated to some extent as a result of the fact that there are n measurements but<br />

only (n - p) degrees of freedom, where p is the number of parameters to be estimated<br />

(Draper and Smith, 1981).<br />

Portmanteau Statistic<br />

The Portmanteau test is used to detect trends in the weighted residuals. If trends are<br />

present, the residuals are not independent. This violates one of the assumptions of the<br />

maximum likelihood method and indicates that the model does not account for all of the<br />

non-random variability in the data. The Portmanteau test is designed for data taken in<br />

sequence (e.g. time or space).<br />

<strong>GPS</strong>-X <strong>Technical</strong> <strong>Reference</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!