The Global Election Influence Formula
After 2010
A Study on the Weight Restructuring of Electoral Outcome Determinants in the Social Media Era
The Global Election Influence Formula After 2010
A Study on the Weight Restructuring of Electoral Outcome Determinants in the Social Media Era
This paper advances a core thesis: since the full-scale entry of social media into political communication in 2010, the determinants of democratic election outcomes worldwide have undergone a fundamental weight restructuring. The two variables that carried the highest weight in traditional electoral political science models—”governing performance” and “campaign funding”—have been surpassed by “a candidate’s personal social media influence,” which has become the single highest-weight variable determining electoral victory or defeat.
Through efficiency-weighted campaign finance analysis of multiple cases—including the 2024 U.S. presidential election, the 2025 Wisconsin Supreme Court election, the 2025 New York City mayoral election, and the 2026 Hungarian parliamentary election—as well as experimental data from a Nature-published X platform algorithm field experiment and PNAS-published Meta deactivation experiment data, this paper constructs a four-dimensional election influence formula. The formula models election outcomes as a weighted combination of social influence, economic/livelihood sentiment, efficiency-weighted campaign funding, and traditional governing performance, with social influence accounting for approximately 40–50% of the weight in non-presidential elections and approximately 25–35% in presidential elections.
Keywords: Social media influence · Election outcome prediction · Campaign finance efficiency · Algorithmic political bias · Information bombardment · Fake social influence
I. Introduction: The Overturned Laws of Elections
In November 2024, Kamala Harris’s campaign spent nearly $2 billion—the most expensive presidential campaign in human history. Her campaign committee directly raised approximately $1.15 billion, while external super PACs spent an additional $850 million[1]. She outspent Donald Trump on advertising by a ratio of approximately 3:1[2]. Yet she lost.
In June 2025, Zohran Mamdani—a 33-year-old New York state legislator—defeated former Governor Andrew Cuomo and his billionaire-backed super PAC television ad blitz with less than $8 million in campaign funds to win the New York City mayoralty[3].
On April 12, 2026—the day before this paper’s publication—Péter Magyar defeated Viktor Orbán, who had been in power for 16 years, by an overwhelming margin of 53.6% to 37.8%, winning a supermajority of seats in the Hungarian parliament[4]. Magyar’s party, built from scratch in just two years, defeated the Orbán regime, which controlled all of Hungary’s traditional media, enjoyed the full support of the state apparatus, received a personal endorsement from Trump, and benefited from Russian intelligence social media interference[5].
These cases constitute a set of natural experiments: different countries, different systems, different political camps, yet the only variable that can systematically explain victory or defeat is the candidate’s personal social media influence.
II. Theoretical Framework: The Four-Dimensional Election Influence Model
2.1 Limitations of Traditional Models
Traditional electoral research treats campaign spending as the most critical predictive variable. OpenSecrets data shows that in the 2020 elections, 89.1% of the highest-spending House candidates and 69.7% of the highest-spending Senate candidates won their races[6]. However, this correlation masks a critical question of causal direction. The entry of social media after 2010 further deconstructed this relationship—candidates can bypass the “fundraise → buy ads → reach voters” chain and directly reach millions of voters at zero cost through social platforms.
2.2 Proposal of the Four-Dimensional Model
M = Efficiency-weighted campaign funding | P = Traditional governing performance | ε = Random disturbance term
Stratified Weight Estimates
Based on a comprehensive analysis combining the PNAS deactivation experiment and multi-case comparisons, this paper proposes stratified weight estimates—variable weights differ significantly between presidential and non-presidential elections[7]:
| Variable | Presidential Elections | Midterm/Local Elections | Explanation |
|---|---|---|---|
| α (Social influence) | 0.25–0.35 | 0.40–0.50 | Information saturation is high in presidential elections, reducing social media’s marginal effect; in local elections, information scarcity makes social influence more decisive |
| β (Economic/livelihood) | 0.30–0.35 | 0.20–0.25 | In presidential elections, voters “judge” the incumbent party by economic performance; in local elections, economic issues are overshadowed by candidates’ personalized narratives |
| γ (Campaign funding) | 0.10–0.15 | 0.10–0.15 | The efficiency-weighted funding weight remains relatively stable across both election types |
| δ (Governing performance) | 0.10–0.15 | 0.05–0.10 | Performance retains some explanatory power in presidential elections; in local elections it is almost entirely replaced by social narratives |
This stratified design responds to the key finding of the PNAS large-scale deactivation experiment: deactivating Facebook and Instagram for six weeks before the 2020 presidential election had effects on voter political attitudes that were “precisely near zero”[8]. Stanford’s Gentzkow explained that in presidential elections, people’s cognition of political issues is already quite thorough, so social media’s marginal effect is small; however, in other elections, social media may have a greater influence[9]. The stratified weights in this paper are based precisely on this insight.
2.3 Operationalization of Social Influence (S)
S is decomposed into three measurable sub-dimensions:
| Sub-dimension | Definition | Measurement Indicators | Typical Representatives |
|---|---|---|---|
| S₁ Candidate native influence | The reach and engagement power of a candidate’s personal social media | Cross-platform engagement ratio (engagement/followers), secondary content sharing rate, organic reach impressions | Trump, Mamdani, Magyar |
| S₂ Platform algorithm control | The ability to control platform recommendation mechanisms | Change in political content share within algorithmic feeds, CUSUM change-point detection | Musk’s adjustments to the X platform |
| S₃ Earned media value | Free media exposure obtained through topic generation | Volume of candidate-related news coverage, media mention frequency, equivalent advertising value (EAV) | Trump’s controversial statements |
The Hungarian election monitoring organization 20k provided a practical measurement example for S₁ in the 2026 election: through cross-platform (Facebook, TikTok, Instagram) engagement comparisons, Magyar’s post engagement was twice that of Orbán’s[10]. This measurement method can serve as a reference benchmark for the operationalization of S₁.
Critical Distinction: Authentic Social Influence vs. Fake Social Influence
S₁ measures only a candidate’s authentic, native social influence. Externally purchased fake influence (such as bot accounts, AI-generated fake videos, and state-level information manipulation) has been empirically proven ineffective. The 2026 Hungarian election provides the most comprehensive evidence: Russia’s Storm-1516 unit deployed 17 AI-generated TikTok channels, produced AI-fabricated execution videos, and used the Matryoshka bot network to spread anti-Magyar disinformation on a massive scale—yet none of these operations prevented Magyar from winning by a 16-percentage-point margin[11]. Similarly, Fidesz and its proxy organizations spent 2.5 times more on social media advertising than all opposition parties combined[12], but purchased exposure cannot replicate the trust generated by a candidate’s authentic interactions.
2.4 Decomposition of Efficiency-Weighted Campaign Funding (M)
Research from the University of Arizona provides the quantitative basis for efficiency weighting[13]:
| Funding Source Type | Effect Coefficient | Relative Efficiency |
|---|---|---|
| Candidate campaign committee direct spending | 0.0381 | 1.00 (baseline) |
| Party and traditional PAC spending | 0.0166 | 0.44 |
| Super PAC independent expenditures | 0.0137 | 0.36 |
III. Empirical Testing: Natural Experiments with Controlled Variables
3.1 The “Trump Control Group” Within the Republican Party
The most powerful empirical evidence comes from within the Republican Party—the same party, the same policy platform, similar donor networks, with the only systematically varying variable being the candidate’s personal social influence.
Funding: Efficiency-weighted approx. $816M (inferior to Harris’s $1.456B)[14]
Funding: Total spending nearly $50M (exceeding opponent Crawford’s $40M)[16]
Funding: TV ad spending only about one-ninth of opponent Taylor’s[17]
3.2 Mirror Verification from the Democratic Camp
Funding: Efficiency-weighted approx. $1.456B (1.78× Trump’s)[1]
Funding: Stopped fundraising after reaching public matching fund cap, total approx. $8M[3]
3.3 Cross-National Verification: Hungary’s 2026 Parliamentary Election
Orbán: Controlled all traditional media for 16 years, full state apparatus support, social ad spending 2.5× all opposition combined, received Trump/Vance personal endorsement in Budapest, Russian GRU intelligence interference[5]
An NPR commentator observed: “If you consider all the political headwinds Magyar faced—funding asymmetry, fighting the state apparatus, restricted media access, a party only two years old—his overwhelming two-thirds majority victory is truly unprecedented.”[18]
3.4 Case Matrix Summary
| Combination | Representative Cases | Result |
|---|---|---|
| High social + low spending | Trump 2024, Mamdani 2025, Magyar 2026 | Won |
| High social + high spending | Crawford 2025 | Won decisively |
| Low social + high spending (incl. state apparatus) | Harris 2024, Schimel 2025, Orbán 2026 | Lost |
| Low social + low spending | Lazar 2026 | Crushed |
IV. Information Bombardment: Quantifying Platform Algorithm Control (S₂)
4.1 Experimental Evidence of X Platform Algorithmic Bias
A field experiment published in Nature in 2026 provides the most rigorous causal evidence: merely seven weeks of exposure to an algorithmic feed can significantly shift users’ political attitudes in a conservative direction, and this effect is unidirectional[19]. Research from Queensland University of Technology further quantified this bias: after Musk endorsed Trump on July 13, 2024, Musk’s account saw a 138% increase in views, a 238% increase in reposts, and a 186% increase in likes[20].
4.2 Equivalent Advertising Value of Information Bombardment
A CBS investigative analysis revealed that among the approximately 17,000 posts Musk published on X in 2024, those touching on election topics averaged 9.3 million views each[21].
Including the algorithm’s systematic boosting of all Republican accounts, X platform’s contribution may equal $3–5 billion in equivalent advertising
Cost to the Trump campaign: zero, and not subject to FEC regulation
4.3 Why Fake Social Influence Fails
However, not all platform manipulation is effective. Hungary’s 2026 election provides counter-evidence: Russia’s Storm-1516 unit deployed 17 AI-generated TikTok channels using fake personas—a young girl, an elderly professor, a football fan—to push anti-Magyar messages. NewsGuard estimated that pro-Orbán network videos accumulated at least 10 million views[22]. Yet Magyar still won by a 16-percentage-point margin.
This reveals the essential distinction between S₁ (authentic native influence) and fabricated social influence: authentic social influence is built on a long-term trust relationship and emotional connection between the candidate and voters, while fake influence lacks this trust foundation. Voters can discern—or at least collectively “filter”—externally manipulated information, but are genuinely moved by a candidate’s authentic personal charisma and narrative ability.
4.4 Generational Comparison with Cambridge Analytica
Christopher Wylie documented in Mindf*ck how Cambridge Analytica achieved precision voter manipulation by harvesting 87 million Facebook users’ data[23]. By owning the X platform itself, Musk can legally and at massive scale achieve even more powerful effects. The weapons of information warfare have been upgraded from “pistols” to “nuclear weapons,” but the regulatory framework remains stuck in the era of regulating “pistols.”
V. Dialogue with Counter-Evidence: The PNAS Deactivation Experiment
The strongest challenge to this paper’s core thesis comes from the large-scale deactivation experiment published in PNAS by Allcott et al. in 2024. The study randomly deactivated the accounts of 19,857 Facebook users and 15,585 Instagram users for six weeks. The core finding was that the effects of deactivation on affective polarization, perceptions of electoral legitimacy, candidate favorability, and voter turnout were “precisely near zero”[8].
On the surface, this appears to directly negate the importance of social media to elections. However, this paper argues that this finding actually supports—rather than undermines—our stratified weight model, for the following reasons:
First, the experiment was conducted in a presidential election context—precisely the election type where social influence carries the lowest weight in our model (α = 0.25–0.35). Gentzkow himself noted: “Even if Facebook and Instagram didn’t exacerbate polarization in the run-up to the 2020 election, they might have more influence in other contexts where people’s beliefs are less entrenched.”[9]
Second, the experiment measured the effect of deactivating an existing platform, not the effect of a candidate’s personal social influence. Even if deactivating Facebook had no impact on voter attitudes, this does not mean that the influence of a Mamdani-caliber social media native candidate is zero—because such influence continues to ferment through interpersonal communication, offline word-of-mouth, and secondary coverage in traditional media, and does not disappear when a single platform is deactivated.
Third, the Nature 2026 X platform experiment reached conclusions different from PNAS: algorithmic recommendations did shift political attitudes[19]. This suggests that effects may vary significantly across platforms—Meta’s algorithm is relatively neutral, while the X platform algorithm under Musk’s direction exhibits a clear political tilt.
VI. Global Verification: Disruptive Elections After 2010
| Year | Country/Election | Key Features |
|---|---|---|
| 2011 | Tunisia/Egypt | Facebook and Twitter organized the overthrow of regimes that had been in power for decades |
| 2014 | India (Modi) | Twitter/YouTube campaigns transformed the electoral landscape, defeating the Congress Party |
| 2016 | United States (Trump) | Zero political experience; relied on Twitter influence to defeat Clinton |
| 2018 | Brazil (Bolsonaro) | WhatsApp group viral spread bypassed traditional media |
| 2019 | Ukraine (Zelenskyy) | Comedian won with 73% of the vote, crushing the incumbent president |
| 2023 | Argentina (Milei) | YouTube/TikTok economist defeated two traditional major parties |
| 2025 | U.S. New York (Mamdani) | $8M + social media defeated a billionaire-backed former governor |
| 2026 | Hungary (Magyar) | 2-year-old party defeated 16-year incumbent + state apparatus + Russian interference + Trump endorsement |
Klüver’s 2025 study published in Comparative Political Studies theoretically validates this pattern: social influencers affect elections through dual pathways of “digital opinion leadership” and “media agenda shaping”[24].
VII. Deeper Mechanisms: Voter Age Stratification and the Collapse of Information Intermediaries
7.1 Age as a Key Moderating Variable of Social Influence
Hungary’s Median polling firm provided the most intuitive age-stratified data: 67% of voters under 30 supported Magyar’s Tisza party, while only 8% supported Orbán’s Fidesz[25]. This near-absolute generational split demonstrates that the effective weight of social influence fluctuates dramatically with the voter age structure. In constituencies dominated by young voters, α may reach 0.60 or higher; in constituencies dominated by elderly voters, the weight of traditional funding (television ads, direct mail) may still approach or even exceed that of social influence.
Pew Research’s 2024 analysis also showed that right-wing influencers post approximately 2.5 times more frequently than their left-wing counterparts (183 times per week vs. 72), yet Trump’s support among Gen Z voters was the highest for any Republican candidate since 2008[26]—indicating that social influence is not merely about “who posts more” but about “whose content can penetrate the attention barriers of the target audience.”
7.2 The Collapse of Information Intermediaries and Emotional Voting
Before 2010, the information transmission in elections was “intermediated”—candidates had to reach voters through television stations, newspapers, and party organizations. Governing performance mattered because intermediary institutions allocated exposure based on performance. After 2010, social media shattered the intermediary monopoly. What determines elections is no longer “what you have done” but “whether you can make voters feel you.”
Facebook’s randomized experiment on 61 million users during the 2010 congressional Election Day proved that political mobilization messages spread through close friends were four times more influential than generic messages[27]. The core advantage of social media lies precisely in its simulation of the “friend recommendation” trust mechanism.
VIII. 2026 Midterm Election Outlook: Structural Confrontation
8.1 The Republican Party’s Structural Dilemma
The Republican Party’s influence structure is extremely centralized—Trump is the sole super-node. But in the 2026 midterm elections, Trump is not on the ballot. 435 House and 35 Senate candidates must fight with their own social influence. Trump’s endorsement efficiency coefficient is only about one-third that of a candidate’s direct communication[13]. The RNC holds $95 million in cash versus the DNC’s mere $14 million[28], but this funding advantage carries a weight of only 10–15%.
The Hungarian election once again demonstrates: Trump/Vance’s external endorsement was of no help to Orbán. The non-transferability of external social influence is a rule with global applicability.
8.2 The Democratic Party’s Distributed Advantage
The Democratic Party’s influence structure more closely resembles a “distributed network”—social media native candidates inspired by the AOC and Mamdani effect. Run for Something reported 10,000 registrations within two weeks of Mamdani’s primary victory[29].
8.3 The Compounding Effect of Energy Geopolitics
The $100+ oil prices resulting from the 2026 Strait of Hormuz blockade provide Democratic candidates with the most powerful free social media ammunition. A TikTok gas station video costs nothing to produce, but its electoral influence may exceed millions of dollars in television advertising.
IX. Conclusion and Discussion
9.1 Core Findings
First, the weights of factors determining global election outcomes after 2010 have undergone a fundamental restructuring. A candidate’s personal social media influence is the single highest-weight variable: approximately 40–50% in midterm/local elections and approximately 25–35% in presidential elections.
Second, campaign funding must be efficiency-weighted by source. The effect of a candidate’s direct spending is approximately 2.8 times that of super PAC spending.
Third, platform algorithm control is an entirely new variable. The Nature experiment confirmed that X platform algorithmic recommendations can systematically shift political attitudes within seven weeks.
Fourth, social influence is the decisive variable; money is merely an amplifier. The Hungarian case pushes this conclusion to its extreme: even possessing an entire state apparatus and foreign government support cannot compensate for a candidate’s deficit in social influence.
Fifth, fake social influence is ineffective. AI-generated content, bot networks, and state-level information manipulation were all proven to fail in Hungary’s 2026 election. Only a candidate’s authentic native social influence produces electoral effects.
9.2 Theoretical Contributions
The core contributions of this paper include: (1) proposing a stratified four-dimensional election influence formula that distinguishes weight differences between presidential and non-presidential elections, responding to the challenge of the PNAS deactivation experiment; (2) operationalizing social influence into three measurable sub-dimensions—S₁ native influence, S₂ platform algorithm control, and S₃ earned media value; (3) rigorously distinguishing authentic social influence from fake social influence and demonstrating the systematic failure of the latter; (4) applying campaign finance efficiency-weighting methodology to actual election data analysis.
9.3 Limitations
Weight estimates are based on a comparative analysis of six cases rather than large-sample regression. While a framework for operationalizing social influence has been proposed, a standardized cross-platform indicator system still requires refinement. The moderating effects of different electoral systems (majoritarian vs. proportional representation) on weights have not been fully explored. The formula’s linear weighting assumption may underestimate interaction effects between variables—for example, high oil prices may amplify the effect of social influence multiplicatively rather than additively.
9.4 Future Research Directions
(1) Conduct regression testing using large-sample data from 435 House districts in the 2026 midterm elections; (2) study platform-specific electoral influence differences across X, TikTok, Instagram, and YouTube; (3) construct nonlinear models incorporating interaction terms; (4) explore regulatory frameworks for platform algorithm control—whether platform algorithm adjustments should be included in campaign finance disclosure, and whether algorithm parameters should be frozen during election periods.