Skip to content
 

On deck this week

Happy spring!

Mon: Noise noise noise noise noise

Tues: Thinking about this beautiful text sentiment visualizer yields a surprising insight about statistical graphics

Wed: In defense of endless arguments

Thurs: Multilevel regression

Fri: Data-dependent priors as an approximation to hierarchical priors

Sat: Objects of the class “Pauline Kael”

Sun: What is a Republican?

One Comment

  1. Anoneuoid says:

    This may be interesting discussion, there is a nice example of “garden of forking paths” with regards to the LIGO detection:

    “LLO – September 14, 2015, 09:53:51 UTC
    – Alex Urban, Reed Essick:
    The Coherent WaveBurst (cWB) data analysis algorithm detected GW150914. An entry was recorded in the central transient event database (GraceDB), triggering a slew of automated follow-up procedures. Within three seconds, asynchronous automated data quality (iDQ) glitch-detection followup processes began reporting results. Fourteen seconds after cWB uploaded the candidate, iDQ processes at LLO reported with high confidence that the event was due to a glitch. The event was labeled as “rejected” 4 seconds afterward. Automated alerts ceased.

    Processing continued, however. Within five minutes of detection, we knew there were no gamma-ray bursts reported near the time of the event. Within 15 minutes, the first sky map was available.

    At 11:23:20 UTC, an analyst follow-up determined which auxiliary channels were associated with iDQ’s decision. It became clear that these were un-calibrated versions of h(t) which had not been flagged as “unsafe” and were only added to the set of available low latency channels after the start of ER8. Based on the safety of the channels, the Data Quality Veto label was removed within 2.5 hours and analyses proceeded after restarting by hand.”
    http://www.ligo.org/magazine/LIGO-magazine-issue-8.pdf

    So if they stuck to the “pre-registered” data processing pipeline, which was apparently discovered to be faulty while collecting data, GW150914 would have been missed. So are the reported p-values valid or invalid?

Leave a Reply