What asserted weights are worth: pricing a receptivity index in households and pre-specifying its test

Abstract Agencies and firms allocate by rankings, so a specification error becomes a targeting error. We audit GCEIA, a deployed five-dimension composite of 3,106 U.S. counties. On the pre-specified proxies (2020 Census self-response and turnout, 3,097 counties), fitted models beat the composite (gradient boosting ρ = 0.758 and 0.770 against 0.634 and 0.157). Two dimensions do not earn their weight on one proxy, and the proxies disagree in sign about two. We revise the index on theory, not on those outcomes, and freeze it (GCEIA-R). The revision improves the Census correlation (ρ = 0.666, difference + 0.032, 95% CI [0.018, + 0.047]) and is neutral on turnout. On turnout the asserted and fitted top deciles share only 30 of 310 counties. Priced in households, rank disagreement overstates the cost: a plan contacting the top tenth of households reaches 73.9% self-responders under the asserted index against 67.7% untargeted, and neither the fitted model nor GCEIA-R differs detectably (at the one-tenth budget). The decisive test has not been run. We pre-specify it as a protocol: a five-year geo-randomized design, hierarchical weight learning from the GCEIA-R prior, and a top-decile capture bar, sized by design-based power analysis. Its contribution is the frozen design, not a result.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11543865/v1
Primary Topic
Census and Population Estimation
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

What asserted weights are worth: pricing a receptivity index in households and pre-specifying its test

Nicholas Sokol
Research Square
Census and Population Estimation
preprint

What asserted weights are worth: pricing a receptivity index in households and pre-specifying its test

Nicholas Sokol
preprint en

Abstract

Abstract Agencies and firms allocate by rankings, so a specification error becomes a targeting error. We audit GCEIA, a deployed five-dimension composite of 3,106 U.S. counties. On the pre-specified proxies (2020 Census self-response and turnout, 3,097 counties), fitted models beat the composite (gradient boosting ρ = 0.758 and 0.770 against 0.634 and 0.157). Two dimensions do not earn their weight on one proxy, and the proxies disagree in sign about two. We revise the index on theory, not on those outcomes, and freeze it (GCEIA-R). The revision improves the Census correlation (ρ = 0.666, difference + 0.032, 95% CI [0.018, + 0.047]) and is neutral on turnout. On turnout the asserted and fitted top deciles share only 30 of 310 counties. Priced in households, rank disagreement overstates the cost: a plan contacting the top tenth of households reaches 73.9% self-responders under the asserted index against 67.7% untargeted, and neither the fitted model nor GCEIA-R differs detectably (at the one-tenth budget). The decisive test has not been run. We pre-specify it as a protocol: a five-year geo-randomized design, hierarchical weight learning from the GCEIA-R prior, and a top-decile capture bar, sized by design-based power analysis. Its contribution is the frozen design, not a result.

Research Square
University of Tennessee at Knoxville (US)
Reduced inequalities
Census and Population Estimation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon