Heritage Anchored to DW-Nominate Scores Setting Heritage Scores as 92% Accurate Reflections of Heritage's Ideology
This is a thought experiment based on work that Tim Groseclose, Steven D. Levitt, and James M. Snyder Jr. (1999) did using ADA and ACU scores to evaluate members of Congress to determine where the politicians fit within ideological space. They argued that while these scores were useful in measuring ideology, that the criteria behind the scores might change over time and thus need to be adjusted. In 2005, Tim Groseclose and Jeffrey Milyo expanded on the use of ADA scores to measure the partisan leaning of members of the media in their paper "A Measure of Media Bias."
While there are academic critiques of these papers, the novel use of partisan think tank scores to measure ideological positions within political "ideological space" is an interesting experiment and useful for the objective practice of political science. For example, a recent research paper by James Manzi published in Theory and Society examined "The Ideological Orientation of Academic Social Science Research 1960-2024." The paper used a large language model to analyze research papers in the social sciences and to score them from 0 (Far Right) to 10 (Far Left) based on the written political positions of politicians and political research instititutions and think tanks as expressed in 2025. This was done in order to evaluate politics in the political equivalent of "Modern Dollar" to account for how changes in ideological drift of thes political actors over time. There has been some pushback to the paper's findings based on the specific anchors used to represent each political label. For example, Pippa Norris who has done extensive research in comparative politics, civic engagement, and Democracy & Populism questioned the use of the term "Conservative" as an anchor for the Heritage Foundation. Based on the tone used in the critical tweet, I interpreted her critique to be that Heritage should have a different ranking and this is an important question and critique of Manzi's article as it opens up the question of subjectivity with regards to the anchors themselves.
While there are academic critiques of these papers, the novel use of partisan think tank scores to measure ideological positions within political "ideological space" is an interesting experiment and useful for the objective practice of political science. For example, a recent research paper by James Manzi published in Theory and Society examined "The Ideological Orientation of Academic Social Science Research 1960-2024." The paper used a large language model to analyze research papers in the social sciences and to score them from 0 (Far Right) to 10 (Far Left) based on the written political positions of politicians and political research instititutions and think tanks as expressed in 2025. This was done in order to evaluate politics in the political equivalent of "Modern Dollar" to account for how changes in ideological drift of thes political actors over time. There has been some pushback to the paper's findings based on the specific anchors used to represent each political label. For example, Pippa Norris who has done extensive research in comparative politics, civic engagement, and Democracy & Populism questioned the use of the term "Conservative" as an anchor for the Heritage Foundation. Based on the tone used in the critical tweet, I interpreted her critique to be that Heritage should have a different ranking and this is an important question and critique of Manzi's article as it opens up the question of subjectivity with regards to the anchors themselves.
In order to address this question, I began this little thought experiment and decided to do the inverse of what Groseclose et al. did in 1999. Instead of using ADA scores to evaluate where politicians fall in ideological space, I decided to use those Heritage Scores and compare them to DW-Nominate and Nokken-Poole Scores in order to determine where the Heritage Foundation fell in ideological space of several recent Congresses (112-118). I manually collected the data for the scores from the Heritage Foundation for Congressmembers (Senate and House 112-118 Congresses) website by cutting and pasting and then merging the files and downloaded data for all Congresses using the VoteView website. Because this is a thought experiment, I used Claude AI to run simple regression models to predict where Heritage Foundation would fall within ideological space using the basic assumption that Heritage was 92% accurate in reflecting its sincere political positions. I also ran one where they were 100% accurate, but have not included that here though it is in the full conversation. You can read the chat I had with Claude, and download the code, here.
The results for the finding are below. The Heritage Foundation consistantly falls within a range of 0.752 and 0.546 (when anchored solely to Republican politicians due to the lack of variance in Democratic assessments). Whether this falls within the range of "Conservative" as opposed to Right Populist, Hard Right, or Far Right, I will leave to you, but if you look at the Overlay Tabs below you will see that Heritage tends to consistantly to the Right of the Median Republican Member of Congress.
The results for the finding are below. The Heritage Foundation consistantly falls within a range of 0.752 and 0.546 (when anchored solely to Republican politicians due to the lack of variance in Democratic assessments). Whether this falls within the range of "Conservative" as opposed to Right Populist, Hard Right, or Far Right, I will leave to you, but if you look at the Overlay Tabs below you will see that Heritage tends to consistantly to the Right of the Median Republican Member of Congress.
Political Science · Roll Call Analysis
Heritage Foundation Scores,
Anchored to DW-NOMINATE
Per-Congress OLS regressions bridge Heritage Action scores (0–100) onto the DW-NOMINATE and Nokken-Poole scales. Primary model uses Republicans only to avoid inflating fit via between-party variance. Full model (D+R+I) shown as a robustness comparison. Congresses 112–118.
Primary: R-only regression | Robustness: Full (D+R+I) regression
Heritage Action Score vs. NOMINATE Score
Each dot is a member of Congress. X-axis = Heritage Action score (0–100). Y-axis = Nokken-Poole (primary) or DW-NOMINATE (robustness) Dim. 1. Regression line fit to Republicans only. Heritage Foundation estimated position shown at X=92 (self-assigned score).
Score Type
Congress
Chamber
Show Parties
Democrats
Republicans
R-only regression line
Full model regression line
Heritage est. (R-only)
Heritage Foundation Overlaid on Congressional Score Distributions
Kernel density of actual scores by party. Vertical lines mark Heritage's estimated position under both models.
Score Type
Congress
Party Filter
Democrats
Republicans
Heritage R-only (±1 SE)
Heritage Full model (±1 SE)
Per-Congress OLS Regression Coefficients
Teal columns = Republicans-only model (primary). Purple columns = full D+R+I model (robustness). Note the lower R² in the R-only model — this honestly reflects within-party variance, not between-party separation.
| Congress | N (full) | N (R) | R-only NP (Primary) | R-only DW (Robustness) | Full Model NP | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| NP R² | Heritage→NP | SE | DW R² | Heritage→DW | SE | NP R² | Heritage→NP | SE | |||
Methodology
How Heritage Foundation is placed on the DW-NOMINATE scale, and why Republicans-only is the preferred specification.
The Core Problem: Democratic Score Compression
Heritage Action scores for Democrats are essentially degenerate as a predictor. Across all seven congresses (112–118), Democrats' mean Heritage score is 7.8 with a standard deviation of 7.3. The 95th percentile is 19. The median is 6.
This creates a measurement problem: when you include Democrats in the regression, the model is exploiting the massive between-party gap (Democrats ≈ Heritage 6, Republicans ≈ Heritage 71) to drive a high R². But that gap is largely just a party indicator — it tells you little about where within the conservative space Heritage itself sits.
Democrats-only R² by congress ranges from 0.0004 to 0.17. Heritage scores explain almost nothing about variation within the Democratic caucus, precisely because there's almost no variation in Heritage scores to exploit.
The Solution: Republicans-Only Regression
The R-only model asks the right question: given that Heritage is a conservative institution operating within Republican political space, where within that space does it fall?
NOMINATEit = αt + βt · Heritageit + εit
estimated on Republicans only
HeritageNP = α̂t + β̂t · 92
estimated on Republicans only
HeritageNP = α̂t + β̂t · 92
The plug-in value is 92 — Heritage's own self-assigned score — rather than 100. Plugging in 100 places Heritage at a hypothetical perfect-conservative position no real member occupies and overestimates the position. Using 92 anchors the estimate to where Heritage actually claims to sit.
R-only R² ranges from 0.43 to 0.79 — lower than the full model, but this is honest. The model is explaining within-party ideological variance, not leveraging party membership as a lever.
Nokken-Poole as Preferred Specification
Nokken-Poole scores are estimated separately for each Congress, directly parallel to Heritage's per-congress rescoring. This alignment makes Nokken-Poole the primary specification: both Heritage scores and Nokken-Poole scores are congress-specific measurements of the same time window.
DW-NOMINATE career estimates are used as a robustness check. Across the 113th–115th Congresses, Nokken-Poole yields higher R² (0.75–0.81 vs. 0.65–0.69 for DW-NOMINATE), consistent with better alignment between per-congress behavior measures.
- Nokken-Poole (Nokken & Poole 2004): Congress-specific ideal points. Primary.
- DW-NOMINATE (Poole & Rosenthal 1985): Career ideal points. Robustness.
- Both scores bounded approximately [−1, +1]: −1 = most liberal, +1 = most conservative.
Connection to Groseclose, Levitt & Snyder (1999)
Groseclose et al. mapped interest group ratings (ADA scores) to a common ideological scale via OLS regression, and Groseclose & Milyo (2005) extended this to place news outlets on the same scale via citation patterns. This analysis applies the same calibration logic: Heritage's own voting scorecards serve as the bridge between an external actor and the NOMINATE scale.
The R-only specification is more analogous to their design — Groseclose anchored media outlets to legislators in the relevant comparison set, not the full ideological range. Including Democrats here would be equivalent to anchoring a conservative outlet's score using liberals' citation behavior.
Interpretation & Caveats
- Plug-in value is 92, Heritage's self-assigned score, not 100. Using 100 would extrapolate to a hypothetical position no member occupies and systematically overestimate Heritage's placement.
- Nokken-Poole is primary; DW-NOMINATE is robustness. Both per-congress measurements align with Heritage's annual rescoring cycle.
- R-only estimates (NP: 0.55–0.75) are lower than full-model estimates (0.55–0.89), correctly reflecting within-party variance rather than between-party separation.
- The 116th–117th Congress dip (NP R-only: 0.55–0.59) reflects real distributional compression as partisan sorting reduced within-Republican Heritage score variance.
- Member-level predictions: R members use the R-only model; D and I members use the full model. Both are retained in the CSV.
- The 112th Congress is included; data source is the same as 113th–118th (heritagefundamentals.com scrape).