Work on proportional data is continuing. Studies of models for proportions with covariates has proceeded with respect both to theory and computer programs for applications. Studies of correlated proportions when the correlation is unknown or unreported have been extended to the case in which prior data is used to impute a lower bound for the unknown correlation and thus improve the inference. This has been applied to the evaluation of EKGs. In nonparametric and robust regression, the behavior of several robust and nonparametric estimators has been studied, with attention to the use of weights to improve the performance over the classic least squares estimates. A simulation study to evaluate various estimators in the one-sample location problem with known weights has resulted in a presentation at the Computer Science and Statistics Symposium on the Interface and a joint paper. A comprehensive study of Receiver Operator Characteristic (ROC) curves has begun. A nonparametric approach has been developed to compare two diagnostic tests using areas under ROC curves when the same patients are used for both tests. With a sparse number of points the straightforward nonparametric estimator for the area under the curve is biased since ties are not treated correctly. A parametric model using truncated exponential distributions has been developed which is robust and also indicates how to treat ties so as to reduce this bias. Applications to ECG programs and to radiology are being studied.

Agency
National Institute of Health (NIH)
Institute
Center for Information Technology (CIT)
Type
Intramural Research (Z01)
Project #
1Z01CT000111-07
Application #
3916962
Study Section
Project Start
Project End
Budget Start
Budget End
Support Year
7
Fiscal Year
1988
Total Cost
Indirect Cost
Name
Computer Research and Technology
Department
Type
DUNS #
City
State
Country
United States
Zip Code