Recognition heuristics
How accurate is the recognition heuristic? Figure 2a shows the
relationship between recognition validity (i.e., PPV) and
discriminability (i.e., d’ ). As expected, the greater d’ ,
the greater the separation between signal and noise. However, the degree
of separation is relatively modest, with maximumd’ ~1.0. Fig 2b displays the relationship betweend’ and TE value. The optimal cut-off is approximatelylog(OR) of 1, which is equivalent to jnd as per the
Weber-Fechner law. Similarly, the largest evidentiary support is found
for TE of about 1( =1.17) at moderate p=0.033. P-values for
larger TE, which theoretically should have higher recognition validity,
has not reached traditional statistical evidence at p<0.05
(Fig 2c), which probably explains why larger treatment effects are not
invariably associated with drug approval without testing in RCTs.
In fact, only 11 observations were
associated with effect size exceeding\(\operatorname{}\text{OR}>2\).