Model Parameters in 8.0 -- Guidelines

Training Issue -- Specify Model Parameters. Many training participants reported problems specifying estimates of match parameters, especially for error probabilities and X Factors. This process is both more automatic and more accurate in Version 8.0. Probability Different has been changed from a source parameter to a match parameter. This is because underlying probabilities of misreporting do not change for different matches but probabilities of correct but different do change, such as for event dates in Crash to EMS, Crash to Hospital, and EMS to Hospital.

X Factors must used to correct calculated match probabilities when data to be linked violate one or more assumptions made in the linkage model. Agree X Factors no longer have to be used to adjust for comparisons with tolerances because the linkage model now incorporates such comparisons explicitly. This is done automatically during each match pass which increases the run time but gives more accurate results. Agree and Disagree X Factors still have to be used to adjust for dependent outcomes. Algorithms that improve on those in the earlier Excel worksheet for estimating these X Factors have been incorporated into the Markov Chain process. Cross-tabulations and other calculations are done automatically. Taken together, these changes to the MCMC process seem to have improved convergence. More details from Markov Chain iterations are tabulated for easy review, includes prior and posterior estimates for many match parameters.