 |
Example
Calibration of Crash Dummies in Automobile Safety Tests
Source: Hardle and Stoker, 1989
Cytel's LogXact 5 vs. SAS's PROC LOGISTIC:
Data from 58 simulated car crashes were analyzed. The relationship between the crash outcome (fatal, non-fatal) and 3 covariates was modeled.
ACL = acceleration
VEL = velocity
AGE = designed "age" of the crash dummy
The results are presented below. SAS fails to fit the logistic regression model -- only LogXact 5 can do it.
LOGXACT 5 RESULTS
=============================================================================================
Parameter Estimates
=============================================================================================
Point Estimate Confidence interval and P-value for Beta
Type Beta SE(Beta) Type 95.0% C.I. Pvalue SE
Lower Upper 2*1-sided
=============================================================================================
acl MLE 0.0175 0.0146 Asymptotic -0.0112 0.0462 0.2319
CMLE 0.0129 0.0136 Monte Carlo -0.0128 0.0434 0.3402 0.0075
( Seed=1038337854,Samples=10000 )
age MLE -0.1398 0.3285 Asymptotic -0.7837 0.5040 0.6704
MUE -0.1512 NA Monte Carlo -INF 0.9986 0.8364 0.0099
( Seed=1038337906,Samples=10000 )
vel MLE -0.0663 0.2521 Asymptotic -0.5605 0.4279 0.7925
CMLE -0.5566 0.6415 Monte Carlo -2.6393 0.3476 0.5474 0.0089
( Seed=1038337921,Samples=10000 )
age.vel MLE 0.0068 0.0073 Asymptotic -0.0075 0.0211 0.3485
CMLE 0.0067 0.0074 Monte Carlo -0.0071 0.0220 0.3578 0.0077
( Seed=1038337935,Samples=10000 )
%CONST MLE -5.4304 11.2677 Asymptotic -27.5148 16.6540 0.6298
=============================================================================================
Analysis time: 00:01:34
=============================================================================================
SAS's PROC LOGISTIC RESULTS
WARNING: There is not enough memory available for exact computations.
Try it yourself! The links below include data files and SAS code that you
can download.
LOGXACT 5--
Download LogXact data
View Instructions for LogXact 5 analysis
PROC LOGISTIC--
Download SAS code (includes data)
|
|
|
