Re: VPC question

From: Norman Z Sent: Tuesday, October 09, 2012 7:30 AM To: [email protected] Subject: [NMusers] VPC question Dear NMusers, I am using PsN to conduct VPC. A “vpc_results.csv” was generated by PsN, where the percentiles of real and simulated data were reported. Generally, these percentiles plus the observed data were used for the VPC plot. My question is how to calculate number of observation outside the 90% prediction interval (PI). I am thinking about two way to do this. First, I may use the NPC results provided by the “vpc_results.csv” file, e.g. NPC results bin 3: 15 observations points below PI (count) points below PI (%) ... 0% PI 7 46.66667 ... 40% PI 5 33.33333 ... 80% PI 1 6.66667 ... 90% PI 0 0 ... 95% PI 0 0 ... Should the numbers in this table be used to obtain the observation outside the 90% PI? If so, how can they be used? If not, shall I calculate the observation outside the 90% PIs of each bin using R? Second, I can calculate the 90% PIs for each time point (e.g. 1.1h, 1.3h, etc) from the raw simulated data and compare the 90% PIs with the observation at one particular time point. Would the second strategy be more accurate than calculate the observation outside the 90% PIs of each binned time interval (e.g. 1-2 h, 2-3 h)? If this is the case, where can I find the raw simulated data? (I suspect the “DV_matrix.csv” and “vpc_simulation.1.npctab.dta” in “m1” folder are the raw simulated data, but want to refer to relevant document to understand how these files are organized.) Thanks for your help. Norman