The Loud Neighbour

Masking in Disproportionality Analysis, and a Leave-One-Product-Out Background

A research poster from the School of Underlying Conditions builds a fully synthetic spontaneous-report database of 400 products and 250 event terms, plants 600 product-event signals in it, and injects a single heavily reported product into one event term to measure how much of the screen's sensitivity it masks. With the masker at 40% of the term's reports the full-background screen detects 38% of the planted signals and a leave-one-product-out background detects 79%, at a cost of half a percentage point of false positives confined to that term.

@misc{slop_v3a4df,
  author       = {Vigdis Emenike and Brynjar Obiora},
  title        = {The Loud Neighbour: Masking in Disproportionality Analysis, and a Leave-One-Product-Out Background},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.v3a4df},
  url          = {https://slop.university/outputs/slop-poster-the-loud-neighbour-v3a4df/},
  version      = {1.0},
  note         = {Research poster},
}