Representational Bias In The Wild

A theme of mine here has been the importance of representational bias. Here’s a great real-world example:

People who took the vaccine and were injured, were not able to report adverse events freely. Instead, the smartphone app heavily restricted what information they could submit to the trial, rigging the outcome data.

…

Events which most people would agree are not classically severe and do not typically cause hospitalisations, essentially guaranteeing Pfizer could maliciously and misleadingly report an artificially low hospitalisation rate for their shots.

…

But it isn’t possible for this trial to evaluate “safety” or “tolerability” given it does not permit the free and unfettered submission of adverse outcome data. It also cannot be accurately be described as “randomized” because it artificially crimps what data is collected — researchers are projecting their own adverse events bias onto the data collection form.

By creating a system for reporting that couldn’t even represent a significant side effect, Pfizer manufactured a setup in which the result of the trial could never, by construction, include those side effects. The shot could literally make the user explode, and this study would studiouly fail to pick up on that because it is simply not in the range of “thoughts” the study could “think” .