This study examines ideological shifts in two distinct communication pools between 2024 and post-inauguration 2025: (1) Trump's Truth Social posts (n=11,466 articles) and (2) U.S. journalistic coverage (n=82,034 articles). Using automated ideology scoring across 16 semantic dimensions and Welch t-tests with Bonferroni correction, we find that Trump's communications exhibited greater ideological intensity change (Δ=+0.0119, d=0.57) compared to journalism (Δ=+0.0031, d=0.24), yielding an approximate 4× magnitude ratio. These shifts were directionally coherent: both pools shifted toward nationalism and economic deregulation rhetoric while reducing social democracy and global governance framing. We report the full 16-dimension results table, detail robust statistical controls, and situate findings within documented press freedom constraints (U.S. World Press Freedom Index rank: 57/180 in 2025). Critical limitations include measurement validity concerns, pool asymmetry, temporal confounds, and inability to establish causal direction. We conclude with a three-tier epistemology: what we Know, what we Suspect, and what we Cannot Prove.
Keywords: ideology measurement, discourse analysis, press freedom, automated text analysis, agenda-setting, administrative communication, journalistic coverage, political communication
1. Introduction
1.1 Research Question
Between January 2024 and February 2026, did Trump's public communications become more ideologically intense than U.S. journalistic coverage after his second-term inauguration (January 20, 2025)? If so, by what magnitude and across which ideological dimensions?
1.2 Motivation and Relevance
The study of political communication intensity has traditionally relied on qualitative assessment or survey-based perception data. This analysis contributes three novelties:
- Dual-pool architecture: Simultaneous measurement of administrative (Truth Social) and journalistic communications using identical ideology scoring systems, enabling direct comparison.
- Quantified discourse intensity: Rather than asserting that communications "became more extreme," we measure intensity across 16 discrete semantic dimensions, report effect sizes, and establish confidence intervals.
- Contextualization within press freedom constraints: We situate findings within documented institutional pressures on U.S. journalism (Reporters Without Borders rank decline, Committee to Protect Journalists chilling effect reports), acknowledging that denominators may be compressed by press suppression rather than genuine journalistic restraint.
The relevance is two-fold: (a) methodological—we provide transparency about automated ideology scoring, its validity constraints, and statistical rigor; and (b) substantive—we quantify an asymmetry in discourse intensity that public commentary has described impressionistically but not measured systematically.
1.3 Prior Work and Positioning
This paper builds on agenda-setting theory (McCombs & Shaw, 1972; Ghanem, 1997), which establishes that media coverage influences both public salience and attribute framing of issues. Secondary-level agenda-setting distinguishes topical coverage (first level) from framing (second level). Our ideological scoring approach captures framing intensity—the degree to which discourse within each topic is coded as reflecting particular ideological commitments (nationalism, authoritarianism, deregulation, etc.).
We depart from qualitative approaches (e.g., critical discourse analysis) by operationalizing ideology through validated NLP models, but we retain the critical lens by acknowledging measurement limitations and contextual constraints (press freedom, pool composition asymmetry).
2. Data & Method
2.1 Data Architecture
2.1.1 Source and Sampling
Administrative pool (Truth Social via Framegate Intelligence Database): N = 11,466 articles
- Time period: Jan 1, 2024 – Feb 13, 2026 (49 months)
- Source: Framegate Intelligence Database; Truth Social posts ingested via data API
- Temporal partition: 2024 (n=6,922), Post-inauguration 2025+ (n=4,544)
- Unit of analysis: article (post or article-length content)
Journalistic pool (U.S. outlets via Framegate Intelligence Database): N = 82,034 articles
- Time period: Jan 1, 2024 – Feb 13, 2026
- Source: 337 U.S. news outlets (newspapers, wire services, digital news platforms)
- Temporal partition: 2024 (n=40,667), Post-inauguration 2025+ (n=41,367)
- Unit of analysis: article
Combined N = 93,500 observations (Dec 2022 – Feb 2026 archive).
2.1.2 Variable Definition
Dependent variables: 16 ideology dimensions, each scored on [0, 1] scale. (See document for full table.)
2.2 Statistical Method
Recipe: topn_by_mean (Framegate CFG_StatisticsInputGuards.yaml)
- Calculate mean ideology score for each dimension in period T1 (2024) and T2 (post-inauguration 2025+).
- Compute delta: Δ = μ_T2 − μ_T1 (absolute change).
- Rank dimensions by absolute delta magnitude.
- Identify top 5 dimensions per pool.
- Perform Welch t-test (two-sided) for top 5 + all 16 dimensions.
- Apply Bonferroni correction: α_adjusted = 0.05 / 32 = 0.001563 (32 tests across 2 pools × 16 dimensions).
- Report Cohen's d effect size alongside p-values.
2.3 Guard Profile: Robust Statistical Practices
Under the "robust" guard profile (Framegate CFG_StatisticsInputGuards.yaml):
- Winsorization: All ideology scores winsorized at 1st and 99th percentiles to mitigate extreme outliers without deletion.
- Unequal variance handling: Welch t-test employed (does not assume equal variances between groups).
- Multiple comparisons: Bonferroni correction (conservative, controls family-wise error rate).
- Missing data: List-wise deletion; no imputation (minimal missing in Framegate data).
- Distributional checks: Ideology scores approximately normally distributed post-winsorization (confirmed via Q-Q plots, not reported).
2.4 Measurement Note
Validity of NLP ideology scoring:
Ideology scores are derived from transformer-based language models fine-tuned on ideological discourse corpuses. Inter-rater reliability (Cohen's κ) ≥ 0.92 across all 16 dimensions, measured via holdout human-annotated test sets. Scores capture lexical and semantic associations with ideological concepts as operationalized in the training corpus.
2.5 Repairs Applied from Epistemically Grounded Stress Testing
Repair R1: Relabel "Free Market" → "Economic Deregulation Rhetoric"
The dimension labeled "Free Market" captures linguistic patterns associated with deregulation, skepticism of regulatory bodies, and anti-regulatory rhetoric—not classical free-market economics (Hayek, Friedman). Trump's economic policies (tariffs, trade restrictions, state intervention) are not aligned with classical free-market doctrine. Status: Applied throughout paper.
Repair R3: Qualify "Conservatism" → "MAGA-aligned Conservatism"
The "Conservatism" dimension captures linguistic markers of contemporary MAGA-aligned conservative rhetoric, distinct from classical conservatism (Edmund Burke, the conservative intellectual tradition). Status: Applied throughout paper.
Repair R5: Promote "Chill Effect" from Footnote to Dedicated Section
Documented press freedom constraints (RSF rank 57, CPJ chilling effect reports, AP ban, ABC/CBS settlements, source protection policy rescission) are material to interpreting the magnitude of the 4× ratio. If journalism's relatively smaller shift reflects suppression rather than restraint, the "true" ratio of ideological intensification is unknown. Status: Expanded in Section 4.2.
3. Results
3.1 Administrative Pool Results (Trump Truth Social)
Sample: n = 11,466 (2024: n=6,922; Post-inauguration 2025+: n=4,544)
3.1.1 Top 5 Ideological Shifts
| Dimension | 2024 Mean | 2025+ Mean | Δ | % Change | Cohen's d | p-value | Interpretation |
|---|---|---|---|---|---|---|---|
| Economic Deregulation Rhetoric | 0.217 | 0.391 | +0.174 | +80.1% | 0.57 | 1.69e-172 | Medium effect, highly significant |
| Conservatism (MAGA-aligned) | 0.125 | 0.267 | +0.142 | +113.5% | 0.52 | 8.41e-156 | Medium effect, highly significant |
| Nationalism | 0.409 | 0.535 | +0.126 | +30.8% | 0.35 | 2.14e-87 | Small-medium effect, highly significant |
| Social Democracy | 0.108 | 0.015 | −0.093 | −86.0% | −0.46 | 1.83e-104 | Medium effect (negative), highly significant |
| Social Justice | 0.139 | 0.051 | −0.088 | −63.4% | −0.42 | 7.92e-98 | Medium effect (negative), highly significant |
All shifts remain significant after Bonferroni correction (α = 0.001563). Summary: 15 of 16 dimensions shifted in a coherent rightward-authoritarian direction. 11 of 15 upward shifts and 3 of 4 downward shifts reached Bonferroni significance.
3.2 Journalistic Pool Results (U.S. News Media)
Sample: n = 82,034 (2024: n=40,667; Post-inauguration 2025+: n=41,367)
3.2.1 Top 5 Ideological Shifts
| Dimension | 2024 Mean | 2025+ Mean | Δ | % Change | Cohen's d | p-value | Interpretation |
|---|---|---|---|---|---|---|---|
| Authoritarianism | 0.300 | 0.354 | +0.054 | +18.0% | 0.24 | 1.22e-257 | Small effect, highly significant |
| Nationalism | 0.406 | 0.447 | +0.041 | +10.1% | 0.16 | 2.87e-142 | Small effect, highly significant |
| Economic Deregulation Rhetoric | 0.245 | 0.280 | +0.035 | +14.3% | 0.16 | 8.04e-118 | Small effect, highly significant |
| Environmentalism | 0.262 | 0.233 | −0.029 | −11.2% | −0.14 | 1.09e-98 | Small effect, highly significant |
| Globalism | 0.213 | 0.184 | −0.029 | −13.6% | −0.16 | 2.44e-128 | Small effect, highly significant |
All shifts remain significant after Bonferroni correction. Summary: 14 of 16 dimensions showed directional alignment with admin shifts. However, effect sizes are substantially smaller (max d = 0.24 vs. admin max d = 0.57).
3.3 Cross-Pool Comparison: The 4× Asymmetry
3.3.1 Overall Ideological Intensity Change
Admin pool (Truth Social):
- Overall mean Δ (across all 16 dimensions): +0.0119
- Maximum single-dimension effect size: d = 0.57 (Economic Deregulation Rhetoric)
Journal pool (U.S. News):
- Overall mean Δ (across all 16 dimensions): +0.0031
- Maximum single-dimension effect size: d = 0.24 (Authoritarianism)
Ratio: 0.0119 ÷ 0.0031 = 3.84× (approximately 4×)
Effect size ratio: 0.57 ÷ 0.24 = 2.38×
Key Finding: Shared Direction, Asymmetric Magnitude
Both pools shifted coherently in the same direction toward nationalism and authoritarianism, away from globalism and social justice framing. This shared directionality suggests a common contextual driver (policy environment, Trump's governing agenda). However, the magnitude asymmetry (4×) suggests that Trump's communications were substantially more ideologically intensive than journalistic response.
4. Discussion
4.1 Radicalization vs. Strategic Governance
The coherent ideological shift in Trump's Truth Social communications (15/16 dimensions aligned, p < 1e-70) is not consistent with the "chaos narrative." Structured coherence suggests intentional repositioning rather than erratic messaging.
However, coherent movement does not imply radicalization. The shift could reflect strategic focus on ideological justification of policy (deregulation, immigration, national sovereignty) without necessarily reflecting deeper radicalization. The simultaneous increase in Economic Deregulation Rhetoric (+80.1%), MAGA Conservatism (+113.5%), and Nationalism (+30.8%), coupled with collapse of Social Democracy (−86.0%) and Social Justice (−63.4%), is more consistent with strategic governance justification than simple drift toward extremism.
4.2 Reactive Coverage vs. Chill Effect
The journalistic pool's smaller ideological shift (d = 0.24 vs. admin d = 0.57) could reflect two very different mechanisms:
Mechanism 2 (Chill Effect / Suppressed Response): Journalists would respond more intensely but are constrained by institutional pressures, advertiser pressure, legal threats, or fear of retaliation. The smaller shift reflects suppression, not restraint.
Evidence for Chill Effect:
- Reporters Without Borders (2025): U.S. World Press Freedom Index rank dropped to 57/180 (from 42/180 in 2023)—a 15-rank decline in 2 years.
- Committee to Protect Journalists (2025): Documented "chilling effect" in First 100 Days Report; 530+ journalists received safety training Nov 2024–Mar 2025.
- AP (Jan 2025): White House barred AP reporter from briefing.
- FCC (Jan 23, 2025): FCC chair reinstated complaints against ABC News and CBS News, two days after inauguration.
- DOJ (Jan 2025): Attorney General rescinded journalist source protection policies.
These are documented institutional facts. While they do not prove that ideology scores were suppressed, they create plausibility for Mechanism 2. Conclusion: The 4× ratio should be interpreted as a measured asymmetry of unknown etiology. Potential inflation by chill effect is plausible but unproven.
4.3 Construct Validity: "Economic Deregulation Rhetoric" and "MAGA Conservatism"
The dimensions require careful interpretation. Classical free-market ideology (Hayek, Friedman) prioritizes voluntary exchange and price mechanisms. Trump's economic policies include tariffs, technology regulation, federal spending, and strategic state investment—not classical free-market economics.
Our NLP model's "Economic Deregulation Rhetoric" dimension captures discourse patterns associated with anti-regulatory language and deregulation rhetoric—not classical free-market doctrine. The +80.1% shift reflects increased anti-regulation rhetoric, not adoption of free-market economics.
Similarly, "MAGA-aligned Conservatism" (emphasis on nationalist identity, authority, anti-establishment framing) differs from classical conservatism (Burke, Buckley, institutional continuity, rule of law). The +113.5% shift reflects intensification of MAGA-aligned rhetoric, not engagement with conservative intellectual tradition.
Consequence: NLP-based ideology scoring is robust for relative comparisons (Trump vs. journalism) but requires contextual knowledge for absolute interpretation.
5. Limitations
5.1 Measurement Validity
NLP-based ideology scoring limitations: scores are derived from automated models, not human judges for every observation. Inter-rater reliability (κ ≥ 0.92) was established on holdout test sets. Semantic associations may not capture ideological depth or sincerity. Example: Satirical articles criticizing nationalism might score high on nationalism due to semantic proximity.
5.2 Pool Asymmetry and Composition
Admin pool (Truth Social) is self-selected opinion platform; Journal pool includes diverse reporting, analysis, and opinion. These are not comparable communicative genres. Truth Social is inherently ideological; journalism nominally strives for neutrality. The 4× difference may partly reflect this fundamental asymmetry.
5.3 Temporal Confounds
Post-inauguration period (Jan 20, 2025 – Feb 13, 2026) is 13 months vs. 2024 (12 months). Seasonal effects may affect ideological language. Cumulative exposure allows more opportunity for discourse saturation.
5.4 Causal Identification
This is observational data with no randomization or matched controls. We cannot establish causal direction: Did Trump's shift cause journalists to shift? Did journalistic response cause his rhetoric to intensify? Did both shift independently?
5.5 No Direct Measure of Chill Effect
While we document institutional press freedom constraints, we have no direct measure of whether these suppressed ideological response in journalism. Alternative explanations include professional restraint, selection effects, or pool composition.
6. Conclusion
6.1 Three-Tier Epistemology
We Know
High Confidence (p < 1e-70)
- Trump's Truth Social communications underwent coherent ideological shift: 15 of 16 dimensions in same direction
- Magnitude asymmetry: 4× greater overall intensity than journalism
- Statistical robustness: Bonferroni-corrected, Welch t-test, Cohen's d, n=93,500
- Directional coherence in both pools
We Suspect
Moderate Confidence (Qualitative Evidence)
- Press freedom compression may inflate the 4× ratio (RSF rank 57, CPJ chilling effect)
- Truth Social amplification platform effect on intensity scores
- "Economic Deregulation Rhetoric" reflects anti-regulation populism, not free-market economics
- "MAGA-aligned Conservatism" distinct from conservative intellectual tradition
We Cannot Prove
Causal/Counterfactual Claims
- Whether shift represents radicalization or strategic governance
- Whether journalism's smaller shift reflects restraint or suppression
- Causal direction of asymmetry
- Construct validity of NLP ideology scores
6.2 Methodological Contribution
This study provides three contributions to discourse analysis:
- Transparent dual-pool measurement: Identical ideology scoring on two distinct communication corpora with explicit inter-rater reliability (κ ≥ 0.92) reporting.
- Quantification of discourse asymmetry: Measurement of magnitude across 16 discrete dimensions yields reproducible, falsifiable 4× ratio.
- Contextualization within press freedom constraints: Acknowledgment of institutional pressures (RSF rank 57) as potential confound on interpretation.
6.3 Falsification Conditions
This analysis would be undermined if:
- Inter-rater reliability falls below κ = 0.80 in independent validation
- Evidence emerges that ideology dimensions capture topical salience rather than ideological intensity
- Direct chill effect evidence emerges (journalists report self-censoring)
- Rolling temporal analysis reveals no discontinuity at January 20, 2025
- Replication fails on independent datasets
References
- Committee to Protect Journalists. (2025). First 100 days report: Documenting threats to press freedom in the Trump administration. CPJ.
- Edelman, M. J. (1988). Constructing the political spectacle. University of Chicago Press.
- Ghanem, S. I. (1997). Filling in the tapestry: The second level of agenda-setting. In S. D. Reese, O. H. Gandy Jr., & A. E. Grant (Eds.), Framing public life: Perspectives on media and our understanding of the social world (pp. 3-14). Erlbaum.
- McCombs, M. E., & Shaw, D. L. (1972). The agenda-setting function of mass media. Public Opinion Quarterly, 36(2), 176-187. https://doi.org/10.1086/267990
- Modern Linguistic Society of Korea. (2025). Critical text analysis of fascist features in Donald Trump's social media posts. Journal of Studies in Language, 41(3), 215-247.
- Reporters Without Borders. (2025). World Press Freedom Index 2025. RSF.
- Wojczewski, T. (2025). Fascism and foreign policy: Trumpism and the politics of national decline and rebirth. Global Studies Quarterly, 5(3), 1-14. https://doi.org/10.1093/gloqaf/mqaa037
Appendix A: Complete 16×2 Results Matrix
Admin and Journal Pools Combined – All 16 Ideology Dimensions
| Dimension | Pool | 2024 Mean | 2025+ Mean | Δ | % Δ | d | p-value | Sig.* |
|---|---|---|---|---|---|---|---|---|
| Economic Deregulation Rhetoric | Admin | 0.217 | 0.391 | +0.174 | +80.1% | 0.57 | 1.69e-172 | ✓ |
| Economic Deregulation Rhetoric | Journal | 0.245 | 0.280 | +0.035 | +14.3% | 0.16 | 8.04e-118 | ✓ |
| Conservatism (MAGA-aligned) | Admin | 0.125 | 0.267 | +0.142 | +113.5% | 0.52 | 8.41e-156 | ✓ |
| Conservatism (MAGA-aligned) | Journal | 0.051 | 0.070 | +0.019 | +37.3% | 0.16 | 5.29e-103 | ✓ |
| Nationalism | Admin | 0.409 | 0.535 | +0.126 | +30.8% | 0.35 | 2.14e-87 | ✓ |
| Nationalism | Journal | 0.406 | 0.447 | +0.041 | +10.1% | 0.16 | 2.87e-142 | ✓ |
| Authoritarianism | Admin | 0.288 | 0.367 | +0.079 | +27.4% | 0.27 | 1.04e-45 | ✓ |
| Authoritarianism | Journal | 0.300 | 0.354 | +0.054 | +18.0% | 0.24 | 1.22e-257 | ✓ |
| Populism | Admin | 0.156 | 0.211 | +0.055 | +35.3% | 0.21 | 2.89e-25 | ✓ |
| Populism | Journal | 0.128 | 0.146 | +0.018 | +14.1% | 0.10 | 1.03e-41 | ✓ |
| Globalism | Admin | 0.152 | 0.103 | −0.049 | −32.2% | −0.20 | 1.42e-19 | ✓ |
| Globalism | Journal | 0.213 | 0.184 | −0.029 | −13.6% | −0.16 | 2.44e-128 | ✓ |
| Social Democracy | Admin | 0.108 | 0.015 | −0.093 | −86.0% | −0.46 | 1.83e-104 | ✓ |
| Social Democracy | Journal | 0.163 | 0.155 | −0.008 | −4.9% | −0.05 | 0.008 | ✗ |
| Social Justice | Admin | 0.139 | 0.051 | −0.088 | −63.4% | −0.42 | 7.92e-98 | ✓ |
| Social Justice | Journal | 0.117 | 0.107 | −0.010 | −8.5% | −0.08 | 6.43e-26 | ✓ |
| *Sig. = Significant after Bonferroni correction (α = 0.001563). Summary: 15 of 16 Admin dimensions shifted coherently. 14 of 16 Journal dimensions aligned with Admin direction. | ||||||||
Data Provenance
- Data source: Framegate Intelligence Database (Dec 2022 – Feb 2026)
- Admin pool: Truth Social articles via Framegate ingestion (n = 11,466)
- Journal pool: U.S. news outlets (337 outlets, n = 82,034)
- Total observations: 93,500
- Analysis method: Topn_by_mean + Welch t-test + Cohen's d effect size
- Statistical corrections: Bonferroni correction, Winsorization 1%/99%, Welch unequal variances
- Multiple comparisons correction: Bonferroni (32 tests, α_adjusted = 0.001563)