Success or failure is measured by computing how closely (i.e. to how many significant digits) the estimated coefficients, standard errors, standard error of estimate, and R-squared produced by the statistical program match the actual certified values.
RATS performs very well on the Linear Regression tests. The only program where the RATS results do not match the certified results closely is the FILIP data set, where RATS flags the regression as suffering from perfect multicollinearity, as indicated by the results of zero for both the last two coefficients and corresponding standard errors.
* NA's indicated that the certified value is exactly zero, and it is not clear how to compute a significant digits result for this case. (RATS does match the zero result to at least 4 decimal places for WAMPLER1, at least 9 places for WAMPLER2.)
The programs, data files, and output files are available in the following Zip file: RatsLinear.Zip
Click on NIST Results to return to the main NIST benchmark page.