Difference between revisions of "CSEP Minutes 02-12-2019"

Latest revision as of 22:51, 12 February 2019

Participants: D. Jackson, W. Savran, J. Gilchrest, D. Rhoades, N. Field, and P. Maechling

Present updated M-test based on Kilmogorov-Smirnov and Anderson-Darling statistics.
Potential issue with distribution-based evaluations for catalogs containing few events.
Optionally switch to magnitude-independent formulation (Chi-Square or M-test based on event counts in magnitude bins). Might incorporate correlations based on GR statistics.
Developed figure showing the Mw relationships for stochastic event sets. This should likely be shown as a survival function instead.
Poissonian error bars on MFD plots, would be larger for the larger events.
- If bins were independent we could get joint probability by multiplying probabilities, but in practice we have correlated errors.
- Roger Mussen talks about b-values and deals with these same issues, variations in MFD due to measured magnitudes. Looking at error distribution can define correlation function between magnitudes.
In the synthetic test for true null hypothesis, empirical distribution should be the same as analytical.
Comment about plots, should use the GR plots using the complimentary distribution. Probably should include errors in the catalog.
Think about tests and show them in a presentable way to outsiders.
Make poster for IUGG. Look up deadline, possible Feb. 18.

@@ Line 7: / Line 7: @@
 * Optionally switch to magnitude-independent formulation (Chi-Square or M-test based on event counts in magnitude bins). Might incorporate correlations based on GR statistics.
 * Developed figure showing the Mw relationships for stochastic event sets. This should likely be shown as a survival function instead.
+* Poissonian error bars on MFD plots, would be larger for the larger events.
+** If bins were independent we could get joint probability by multiplying probabilities, but in practice we have correlated errors.
+** Roger Mussen talks about b-values and deals with these same issues, variations in MFD due to measured magnitudes. Looking at error distribution can define correlation function between magnitudes.
+* In the synthetic test for true null hypothesis, empirical distribution should be the same as analytical.
+* Comment about plots, should use the GR plots using the complimentary distribution. Probably should include errors in the catalog.
+* Think about tests and show them in a presentable way to outsiders.
+* Make poster for IUGG. Look up deadline, possible Feb. 18.