Record linkage procedures based on the Fellegi-Sunter Theory (JASA, 1969) require the estimation of the conditional probabilities of the agreement patterns. Under the assumption of conditional independence, this reduces to the estimation of the conditional probabilities of the agreement of the individual matching fields. We consider methods for using value-specific, frequency-based methods to modify the agreement probabilities according to the rate of recurrence of the common matching field value in the matching set. We compare and analyze the effects of the methods when applied to Census data sets, and assess their value and usability.
