23 lines
No EOL
2.6 KiB
Markdown
23 lines
No EOL
2.6 KiB
Markdown
---
|
|
type: claim
|
|
domain: internet-finance
|
|
description: Randomly implementing only some approved policies to create counterfactuals fails at realistic randomization rates because selection signal dominates causal signal
|
|
confidence: experimental
|
|
source: Nicolas Rasmont (LessWrong), analysis of randomization fix
|
|
created: 2026-04-10
|
|
title: "Post-hoc randomization requires implausibly high implementation rates (50%+) to overcome selection bias in futarchy"
|
|
agent: rio
|
|
scope: functional
|
|
sourcer: Nicolas Rasmont
|
|
related_claims: ["[[conditional-decision-markets-are-structurally-biased-toward-selection-correlations-rather-than-causal-policy-effects]]"]
|
|
related:
|
|
- Conditional decision markets are structurally biased toward selection correlations rather than causal policy effects, making futarchy approval signals evidential rather than causal
|
|
- Hanson's decision-selection-bias solution requires decision-makers to trade in markets to reveal private information and approximately 5 percent random rejection of otherwise-approved proposals
|
|
reweave_edges:
|
|
- Conditional decision markets are structurally biased toward selection correlations rather than causal policy effects, making futarchy approval signals evidential rather than causal|related|2026-04-18
|
|
- Hanson's decision-selection-bias solution requires decision-makers to trade in markets to reveal private information and approximately 5 percent random rejection of otherwise-approved proposals|related|2026-04-18
|
|
---
|
|
|
|
# Post-hoc randomization requires implausibly high implementation rates (50%+) to overcome selection bias in futarchy
|
|
|
|
Rasmont analyzes the proposed fix of randomly implementing approved policies to create counterfactual data for causal inference. The mechanism is that if only X% of approved policies are actually implemented, the market can compare outcomes between implemented and non-implemented policies to isolate causal effects. However, Rasmont argues this requires 'implausibly high randomization rates - perhaps 50%+' before the causal signal overwhelms the selection signal. At realistic randomization rates (5-10%), the selection bias still dominates because the correlation between approval and fundamentals is stronger than the causal effect of most policies. This means the fix would require organizations to randomly not implement half of their approved policies, which defeats the purpose of having a decision mechanism. The alternative fix - random settlement regardless of outcome - eliminates the information aggregation purpose entirely by transforming markets into influence-buying mechanisms where capital rather than information determines outcomes. |