Estimating parameters for data below the MDL

• Not appropriate to simply ignore observations below MDL where computing sample statistics (mean, standard deviation)

• A number of approaches have been proposed and applied

• Three are not particularly useful

  • Replace all <MDL values with MDL
  • Replace all <MDL values with 0.0
  • Replace all <MDL values with MDL/2

Previous slide

Next slide

Back to first slide

View graphic version