ESTIMATION OF TREATMENT EFFICACY WHEN NUMBERS OF TEST SUBJECTS ARE UNKNOWN

Procedures are described for analysis of number of survivors from bioassays done to determine the efficacy of quarantine treatments of fruits artificially infested with insects. Analysis of the observed survivors with Poisson or Poisson-lognormal models is preferable to the use of probit analysis. Simulation studies demonstrated that estimates of treatment levels necessary for 99.9968% mortality from survival analysis are superior (i.e., have smaller mean square errors) to those obtained from probit regression. Furthermore, analysis of the observed number of survivors avoids the problem of negative mortality estimates encountered in probit mortality methods. Based on these results, we recommend the use of data on observed survival rather than estimated mortality for the study of quarantine treatment efficacy.