1963 MODEL MATERIALS

This folder contains analytical output from the 1963 Harris–Newsweek proof-of-concept
used in "Beyond Isolated Headlines: A Framework for Coupling Scorekeeping and Explanatory
Modeling in Public Political Polls."

FILE
  jfk1963-variable-ranking-standalone-incremental.xlsx

PURPOSE
  Export from the JFK Model Builder's Standalone vs Incremental view for all 123 scored
  candidate variables. It supports the manuscript's discussion of the Variable-Ranking
  Tool, Table 2's standalone-versus-incremental distinction, and the disclosures concerning
  party identification (q3) and ideological self-placement (q19).

WORKBOOK CONTENTS
  Run info
    Records the dataset (JFK), headline outcome (Approval), six-variable reference model,
    full-sample scope, candidate count, generation time, and definitions of the displayed
    measures.

  Standalone vs Incremental
    Reports, for each candidate variable:
      - Data column: the variable's column name in 1963-replication/opster6.csv. Racenew is
        the author-created two-category race variable. The two "Grandparent Birth Country:
        Poland" rows are q33b_5 and q33b_11, because the printed questionnaire lists Poland
        twice (Supplementary Appendix S1).
      - Standalone: one-variable Tjur R².
      - Incremental: change in Tjur R² when the variable is added to or dropped from the
        reference model.
      - n_standalone: respondents with valid approval, weight, and the variable; the sample
        for the Standalone fit.
      - n_incremental: respondents with valid approval, weight, the six reference
        predictors, and the variable; the common sample for the Incremental contrast
        (1,159 for the six reference predictors). Splitting the 116 omitted candidates at
        1,000 cases reproduces the Appendix E.4 subset correlations.
      - Overlap: largest pairwise association with any single predictor already in the
        model: Spearman's rho when both variables are ordered (no-opinion answers set
        aside), Cramér's V when either is unordered, eta for an unordered variable against
        a continuous one; survey-weighted (Appendix E.4). The Run info sheet's one-line
        description ("max correlation") refers to this measure.
      - In model / Why not in model: the Model Builder's status and plain-language prompt.
        The Why not in model column follows the standalone–incremental diagnostic's
        0.005 band for negligible incremental contribution, not the 0.008 threshold the
        candidate cards use (Appendix E.4); party identification, for example, is labeled
        "Adds signal" at 0.0055.
      - The original three-category race variable (q40) shows "—" for Incremental: it is
        collinear with Racenew, so no incremental value is defined (Appendix E.4).

PARTY IDENTIFICATION AND IDEOLOGY
  q3, party identification ("Party Identification" in the workbook):
      Standalone Tjur R²    0.0620
      Incremental ΔTjur R² 0.0055
      Overlap              0.153
  q19, ideological self-placement, final item "Yourself" (q19_7 in the data; "Own
  Ideological Self-Placement (Liberal, Conservative, or Middle of the Road)" in the workbook):
      Standalone Tjur R²    0.1068
      Incremental ΔTjur R² 0.0221
      Overlap              0.216

INTERPRETIVE CAUTION
  Incremental values are not all the same kind of comparison: included predictors use
  drop-one contributions, while omitted candidates use add-one contributions. The workbook
  is an analytical output for inspection and disclosure, not a claim that the reference
  model was optimized or uniquely correct.

PROVENANCE
  Generated by the companion Model Builder on September 14, 2026 at 10:55:36 PM, as recorded
  in the workbook's Run info sheet. This export replaces the August 17, 2026 export, which
  predated the overlap measure described above and the current codebook labels; its
  standalone and incremental values are unchanged.
  The Data column, n_standalone, and n_incremental columns were added on September 15,
  2026, computed from 1963-replication/opster6.csv under the same row rules, and the
  Race (q40) Incremental value was changed from 0.0000 to "—" because none is defined.
  All other values are unchanged from the September 14 export.
