JASP_Bayesian repeated measures ANOVA

Comments

• EJ

  September 2018

  Hi Rachellql,

  The results in the Bayesian analysis change from one run to the next because they are obtained using a numerical approximation routine. Under Advanced Options you can increase the samples that the routine draws in order to obtain a more stable estimate. The Auto setting may sometimes give %errors that are a little to big for comfort. You should be able to up the number of samples used in the approximation routine and thereby lower the %error.

  Cheers,
  E.J.
• JAH

  September 2018

  Greetings.
  I have just started using JASP to run Bayesian repeated measures ANOVAs, and i have noticed some discrepancies in the values in the output table. In these cases, the BF_10 for the interaction does not correspond to what i calculate using the ratio of the p(M|data) for the interaction and the null. For example, in the table pasted below, the p for the CS*Session interaction is 0.002, and the p(H0) is 6.811E-33. By my calculation, the ratio of these is 2.936E+29, whereas the table shows the BF_10 for that model as 2.323E+29. I presume this discrepancy is a rounding error since the ratio of the p's gets closer to the BF in the table if i let p(M|data) be 0.0015 instead of 0.002.
  My question then is should i use the BF values in the table in preference to the p(M|data) values if i wish to calculate the evidence for the interaction beyond the model that includes the two main effect factors? Using the probability values in the table, i get a BF (favouring no interaction) of 84.5 (=0.169/0.002), but i get a BF of 106.6 (=2.477E+31/2.323E+29) if i use the BF_10 values in the second last column.
  Any advice would be greatly appreciated.
  Cheers,
  Justin

  
• EJ

  September 2018

  This must be due to rounding, in which case it is best to use the BFs. (I'll double check, thanks for bringing this up).
  Cheers,
  E.J.
• JAH

  September 2018

  Great. And thanks for such a quick reply!
  Justin