https://community.jmp.com/t5/Irish-JMP-Users-Group/Variance-Components/m-p/110615#M43 <P>From a statistical viewpoint, what are the dangers with treating a negative variance from the EMS method as zero? And why would you use REML or Bayesian Instead?</P> Tue, 12 Feb 2019 16:40:18 GMT JDore 2019-02-12T16:40:18Z https://community.jmp.com/t5/Irish-JMP-Users-Group/Variance-Components/m-p/110615#M43 <P>From a statistical viewpoint, what are the dangers with treating a negative variance from the EMS method as zero? And why would you use REML or Bayesian Instead?</P> Tue, 12 Feb 2019 16:40:18 GMT https://community.jmp.com/t5/Irish-JMP-Users-Group/Variance-Components/m-p/110615#M43 JDore 2019-02-12T16:40:18Z https://community.jmp.com/t5/Irish-JMP-Users-Group/Variance-Components/m-p/110705#M45 <P>For EMS vs REML, take a look <A href="https://www.jmp.com/support/help/14-2/mixed-models-and-random-effect-models.shtml" target="_self">here</A>, particularly the section ''REML vs. EMS . . . '. For the Bayesian approach used in JMP, see Laura's 2015&nbsp;<A href="https://community.jmp.com/t5/Discovery-Summit-2015/Strictly-Positive-Estimates-of-Variance-Components-for/ta-p/22945" target="_self">Discovery paper</A>.</P> Wed, 13 Feb 2019 10:38:54 GMT https://community.jmp.com/t5/Irish-JMP-Users-Group/Variance-Components/m-p/110705#M45 ian_jmp 2019-02-13T10:38:54Z