to Dave's research index.
GPS seconds time scrambler for Point-Source blindness
5 November 2001
to Dave's research index.
There has been much debate in the collaboration over the implementation of various blindness schemes for upcoming analysis. Pursuant to the analysis call decision, I present here a scheme for blinding a point source search analysis.
In general, the concern of this discussion is that a human analysis of data is succeptible to bias via the tuning of cuts. In the case of a point source analysis the particular concern is that the analyzer could tune his or her cuts in order to artificially enhance an excess of events arriving from one direction in the sky. The most obvious way to protect the data would be to somehow randomize the arrival direction of each event before the cuts are developed. A practical blindness protection scheme must accomplish this goal while leaving unhindered the analyzer's ability to study the background and systematic detector effects.
Given our detector's special location at 90 degrees South lattitude, the arrival time of an event translates directly into right ascention. Thus, randomizing the arrival time of each event is a viable way of scrambling the arrival directions while at the same time preserving the ability to investigate zenith dependent systematics.
One would also like to investigate time-dependent systematics. Since most observed time-dependent effects happen on a time scale of days or more, it should pose little hinderance to such an investigation to scramble the seconds of the arrival time, leaving the day intact. Scrambling just the seconds of the arrival time within the day will randomize the right ascension into the set of all possible right ascentions.
I have written a perl script (available here) which
randomly scrambles the GPS seconds
of the day on an event by event basis. The program accepts standard
input and writes to standard output for easy insertion into existing or
future analysis scripts. The program writes the following warning message
conspicuously in the header of the output:
HI ************************************************
HI *
*
HI *
*********** NOTICE!***********
*
HI *
*
HI * The seconds of the event GPS times in *
HI * this file have been scrambled for blindness! *
HI *
*
HI ************************************************
and also tags each event with the following user line:
US scrambled 1
We propose that a point source analysis would be sufficiently blind even if it were to use 100% of the data processed through this script at a sufficiently low cut level, until all cuts are finalized. A practical place to run the script would be after post-level 2 reconstructions, but before post-level 2 cuts.
As a demontration that this script works, I plot below a histogram of event times in an arbitrary file of data. The solid lines represent the true recorded time while the dotted represent the scrambled times. The true times are bunched together in groups because these data comprise events merged from non-contiguous files of a run.
to the research index.