14.03.2014 Views

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Chapter 9 Performing Nonlinear Regression 271<br />

Statistical Details<br />

Figure 9.20 Solution Table for the Poisson Loss Example<br />

Statistical Details<br />

This section provides statistical details <strong>and</strong> other notes concerning the Nonlinear platform.<br />

Profile Likelihood Confidence Limits<br />

The upper <strong>and</strong> lower confidence limits for the parameters are based on a search for the value of each<br />

parameter after minimizing with respect to the other parameters. The search looks for values that produce<br />

an SSE greater by a certain amount than the solution’s minimum SSE. The goal of this difference is based<br />

on the F-distribution. The intervals are sometimes called likelihood confidence intervals or profile likelihood<br />

confidence intervals (Bates <strong>and</strong> Watts 1988; Ratkowsky 1990).<br />

Profile confidence limits all start with a goal SSE. This is a sum of squared errors (or sum of loss function)<br />

that an F test considers significantly different from the solution SSE at the given alpha level. If the loss<br />

function is specified to be a negative log-likelihood, then a Chi-square quantile is used instead of an F<br />

quantile. For each parameter’s upper confidence limit, the parameter value is increased until the SSE reaches<br />

the goal SSE. As the parameter value is moved up, all the other parameters are adjusted to be least squares<br />

estimates subject to the change in the profiled parameter. Conceptually, this is a compounded set of nested<br />

iterations. Internally there is a way to do this with one set of iterations developed by Johnston <strong>and</strong> DeLong.<br />

See <strong>SAS</strong>/STAT 9.1 vol. 3 pp. 1666-1667.<br />

Figure 9.21 shows the contour of the goal SSE or negative likelihood, with the least squares (or least loss)<br />

solution inside the shaded region:<br />

• The asymptotic st<strong>and</strong>ard errors produce confidence intervals that approximate the region with an<br />

ellipsoid <strong>and</strong> take the parameter values at the extremes (at the horizontal <strong>and</strong> vertical tangents).<br />

• Profile confidence limits find the parameter values at the extremes of the true region, rather than the<br />

approximating ellipsoid.

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

Saved successfully!

Ooh no, something went wrong!