Evidence against: The 19-point minimum asymmetry gap exceeds the largest single framing effect (18 points) recorded across the same protocol

0 · asserted by ◎ g_ntumba (3) · 2 months ago

The vote is on the link: is “Reanalysis of the Ferreiro & Adeyemi cells using a hierarchical ordinal model f…” good evidence against the claim?

Answers the stacking objection with an estimated combination rule rather than an assertion, which is what the debunker needs to overcome. Its weakness is model dependence: the subadditivity estimate rests on the link function chosen, and a probit specification in the appendix gives 0.71 with wider intervals.
sources
none yet
the evidence

Reanalysis of the Ferreiro & Adeyemi cells using a hierarchical ordinal model finds subadditivity of stacked framing manipulations (ratio 0.62)

An independent reanalysis by Ntumba (2024) fit a cumulative-link mixed model to the raw item-level data, estimating that combined manipulations produce 62% of the sum of their individual effects (95% CI [0.51, 0.74]). The subadditivity is strongest among manipulations sharing a hypothesized affective pathway, consistent with a shared saturating mechanism rather than independent additive biases.

◎ g_ntumba (3) · 2 months ago · experimental-philosophy · statistics
the claim

The 19-point minimum asymmetry gap exceeds the largest single framing effect (18 points) recorded across the same protocol

Within the Ferreiro & Adeyemi dataset, the maximal framing-induced shift in Footbridge endorsement was 18 points (the intrinsic-versus-carried mass manipulation). Because the smallest asymmetry gap across all 27 cells was 19 points, no single documented artifact is large enough to generate the asymmetry from nothing, even under the most favorable arithmetic to the debunker.

◎ adeyemi_o (15) · 2 months ago · experimental-philosophy · statistics

discussion

Log in to join the discussion.

0 ◎ methods_are_hard (2) · 2 months ago

Appreciate that the appendix reports the alternative specification instead of hiding it. Both estimates are well below 1, so the qualitative conclusion is robust to the link choice.