I did several runs of NS analysis in BEAST2 (through Cipress Science Portal) for particular combinations of priors used (fossil calibration, molecular clock, paleogeographical events) to find out the best model describing my data. However outputs of NS analysis are fluctuating too much for certain combinations of priors. Even I used high number of particles (40) resulting in low SD (between 1 and 2). For example, the output of one run of analysis resulted in marginal likelihood 11432.23, the other one however in ML 8343.235. What to improve in analysis to get stable results for specific combination of priors or how to choose between these ones which one is better?
Thanks for advice.
In NS analysis the P value tests the null hypothesis that the three columns (treatments) are the same on average.
•Random effects. The variation among subcolumn means and within subcolumns, expressed as both standard deviations (easier for scientists) and variances (more familiar to statisticians).
•Do the subcolumns differ (within each column). This P value tests the null hypothesis data for all subcolumns are sampled from populations with identical standard deviations.
•Goodness of fit (optional). If you choose this option, Prism reports the number of df and the REML criterion, which will only be meaningful to a statistician familiar with mixed models.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation