Final Update, July 2003
Prior to the start of the Supreme Court's Term in October 2002, we created a computer model comprised of a series of classification trees to forecast the Justices' votes. We coded several characteristics (e.g. Circuit of origin) of each pending case and then entered those characteristics into the model to produce the machine's forecasts. Prior to oral argument, we posted the machine's forecasts on this website.
Recently, we became aware of an error in the computer code that was used to input the case characteristics into the model. As a result several of the forecasts originally posted on this website did not actually reflect the output of the model as written. We have corrected that programming error and reposted the correct forecasts here. In most cases, the machine's forecasts did not change, although in seven cases the machine correctly predicted reversal when the original (incorrect) prediction was affirmance, and in six cases the opposite occurred.
Please note that the computer model (i.e. the classification trees) used to generate forecasts has not changed since the start of the Term, nor has the manner in which the characteristics of the pending cases was coded. What occurred, and has now been corrected, was a programming error that erroneously classified cases as they were run through the model. (For example, a case from the 9th Circuit was erroneously entered into the model as if it had originated in the 5th Circuit).
For more technical information about the bug, please download and upack this file. You can also access code necessary to replicate the correct forecasts here.
We have now posted the estimated classification trees used to forecast the cases below. Click here to view the pdf.