📈 economics

International Linkage and Democratic Velocity: Nonlinear Attribution Evidence from a Neural Additive Model for Panel Data

This study employs a novel interpretable Neural Additive Model for Panel Data (NAM-PD) to demonstrate that institutional treaty depth is the dominant predictor of democratic backsliding with a universally negative impact, while diplomatic engagement uniquely promotes democratic gains, and other linkage factors like migration and foreign direct investment exhibit complex nonlinear or counterintuitive effects across 176 countries from 1995 to 2025.

Original authors: Hakan Mehmetcik

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hakan Mehmetcik

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world's countries as a giant, chaotic video game where the goal is to keep the "Democracy Score" moving forward. Sometimes the score goes up (democracy gets stronger), sometimes it stays flat, and sometimes it crashes down (democracy slips backward). For years, researchers have been trying to figure out what "power-ups" or "traps" in the game make the score move.

Most previous players used a very simple tool: a straight-line ruler. They assumed that if you added more of a specific ingredient—like money, treaties, or diplomats—the democracy score would just go up or down in a perfectly straight line. But this study, led by Hakan Mehmetcik, says, "Wait a minute. The real world isn't a straight line; it's a rollercoaster with loops, drops, and sudden stops."

To fix this, the team built a super-smart, flexible computer brain called a Neural Additive Model (NAM-PD). Think of it as a detective that doesn't just look at the whole picture but zooms in on every single clue to see exactly how it bends and twists the outcome. They fed this detective data from 5,433 different years and 176 countries, spanning from 1995 to 2025.

Here is what the detective found, and what it definitely didn't find:

The Big Surprise: The "Deep Treaty" Trap

The biggest clue the detective found was Institutional Treaty Depth (how deeply a country is locked into international agreements).

The Old Theory: People thought signing deep treaties was like putting a seatbelt on democracy—it would keep it safe and steady.
The Reality: The data shows the opposite. The more deeply a country is locked into these treaties, the more likely its democracy score is to slow down or drop.
The Numbers: This factor is the heavyweight champion, accounting for 37.28% of the prediction power. The relationship is negative across the board. Even when the researchers removed all European Union countries from the data to make sure they weren't just seeing a "Europe problem," the result stayed the same (still 37.03%).
The Shape: It's not a gentle slope. Once a country crosses a certain threshold (around a "Free Trade Agreement" level), the negative effect gets sharp and then stays flat at a low level. It's like a trapdoor that opens once you step too far in.

The Only Good News: Diplomatic Visits

If treaties are the trapdoor, Diplomatic Engagement (foreign leaders visiting each other) is the only true "power-up."

The Finding: More visits mean a higher chance of democracy moving forward.
The Numbers: This is the second most important factor, making up 19.38% of the prediction.
The Catch: It's not a magic wand that works forever. The benefit grows as visits increase, but it hits a "saturation point" around 21 visits per year. After that, sending more diplomats doesn't help much more. It's like drinking water: the first few glasses save you, but the tenth glass doesn't make you twice as hydrated.

The Weird Twist: The Migration Reversal

Migration is the most confusing character in this story. It acts like a switch that flips.

The Finding: When a country has a small to medium number of migrants, it actually helps democracy move forward. But once the migrant population hits a specific tipping point, the effect flips and starts hurting democracy.
The Numbers: That tipping point is exactly 47.3% of the country's total population.
The Context: The countries above this line are mostly places like the UAE, Kuwait, and Qatar, where huge numbers of temporary workers live under strict rules (the kafala system) that don't let them vote or join society. The model figured this out on its own without being told to look for it.

The Money Myth: Foreign Investment

For a long time, economists believed that Foreign Direct Investment (FDI)—money coming in from other countries—would naturally build democracy because rich people want fair rules.

The Finding: The data completely rejects this. In this study, more foreign money is actually associated with slower democratic progress.
The Numbers: This factor accounts for 13.15% of the prediction, and the line goes strictly down.
The Mystery: The study doesn't know why this happens. It could be that rich investors prefer stable dictatorships, or that governments suppress rights to keep investors happy. The model just sees the pattern; it doesn't solve the mystery.

What the Model Can't Do

It's important to know what this detective didn't do. The study is a predictive map, not a causal proof.

The model shows strong associations (patterns that happen together), but it doesn't prove that one thing causes the other.
The researchers admit that their model's ability to predict the exact future score of a country is no better than a simple guess. The "accuracy" is basically the same as a coin flip because democracy is messy and hard to predict.
The real value isn't in guessing the future; it's in revealing the shape of the relationship. It showed us that the "straight line" theories were wrong and that the world operates with thresholds, reversals, and saturation points.

The Bottom Line

This study is a high-fidelity map that tells us:

Deep treaties might be slowing down democracy, not saving it.
Diplomacy helps, but only up to a point.
Migration is a double-edged sword that flips at 47.3%.
Foreign money isn't the hero we thought it was.

The authors aren't claiming to have "solved" democracy. Instead, they've handed future researchers a much better set of tools to figure out why these patterns exist, so we can stop guessing and start testing the right ideas.

Technical Summary: International Linkage and Democratic Velocity

Problem Statement
The study addresses two structural limitations in the empirical literature on international linkage and democratic backsliding. First, prior comparative work typically isolates single linkage mechanisms (e.g., institutional membership, FDI, migration) without a common framework to compare their relative predictive weights. Second, macro-comparative studies predominantly rely on linear fixed-effects (FE) estimators. These linear models impose constant, additive marginal effects, rendering them incapable of recovering threshold dynamics, saturation points, or directional reversals (e.g., migration effects that flip at high stock levels). Consequently, nonlinear associations remain systematically invisible to standard estimation techniques.

Methodology
The authors introduce a Neural Additive Model for Panel Data (NAM-PD) to analyze a global panel of 5,433 country-year observations across 176 countries from 1995 to 2025.

Data: The dependent variable is annual democratic velocity ( $\Delta v2x\_polyarchy$ ), defined as the first difference of the V-Dem Electoral Democracy Index. Predictors include five linkage pillars: institutional treaty depth (DESTA), diplomatic engagement, migration stock, foreign direct investment (FDI), and conflict intensity.
Architecture: The NAM-PD routes each predictor through an isolated sub-network (Linear $\to$ ReLU $\to$ Linear $\to$ ReLU $\to$ Linear) and sums the outputs. This architectural design ensures exact additive attribution by construction, eliminating the need for post-hoc approximation methods like SHAP or LIME.
Validation Strategy: The study employs a chronological partition (Training: 1995–2015; Validation: 2016–2020; Test: 2021–2025) to prevent data leakage. It establishes frequentist baselines using Two-Way Fixed Effects and Pooled OLS with country-clustered standard errors. Crucially, the authors explicitly state that the model is a predictive instrument, not a causal one; it maps statistical associations and functional forms but does not resolve endogeneity or establish counterfactuals.

Key Results
The analysis yields four primary findings regarding the functional form and predictive weight of international linkages:

Institutional Treaty Depth (DESTA): This is the dominant predictor, accounting for 37.28% of the attribution weight. Contrary to the "anchoring" hypothesis, the shape function is universally negative across the entire ordinal range (depth 0–6). The negative association intensifies sharply at the transition from Preferential Trade Agreements (PTA) to Free Trade Agreements (FTA) and stabilizes at a negative plateau. This signal persists even when EU member states are excluded (37.03% attribution in the non-EU subsample), rejecting the "EU dummy" concern.
Diplomatic Engagement: The sole predictor with an unambiguously positive association (19.38% attribution). The relationship exhibits saturation, with marginal returns diminishing above approximately 21 inbound visits per year.
Migration: Exhibits a structural reversal at a migrant stock of 47.3%. Below this threshold, migration is positively associated with democratic velocity; above it, the association turns negative. This discontinuity corresponds to the kafala sponsorship model prevalent in Gulf Cooperation Council states.
Foreign Direct Investment (FDI): Carries a universally negative association (13.15% attribution), contradicting modernization theory predictions of a positive co-evolution between capital flows and democracy.
Predictive Performance: All models (NAM-PD, FE, Pooled OLS) achieve near-identical out-of-sample accuracy (RMSE $\approx$ 0.041–0.043, $R^2 \approx$ 0) on the 2021–2025 holdout. The authors note that annual democratic velocity is a "near-noise-floor" forecasting target, meaning the NAM-PD's value lies not in superior point forecasts but in functional form resolution.

Significance and Claims
The paper claims its primary contribution is the generation of a high-fidelity functional form map of international linkage associations. By demonstrating that linear estimators miss substantial predictive information (e.g., the 46.15% combined attribution of non-significant linear predictors in the NAM-PD), the study argues that the linkage hypothesis requires a differentiated reformulation.

The authors position this work as a systematic precursor to causal identification. They explicitly state that the model does not identify mechanisms or rule out endogeneity. Instead, it provides the necessary structural evidence—specifying where thresholds, saturation, and directional reversals exist—to guide future research designs and instrument selection. The findings suggest that deep institutional integration may suppress democratic velocity (consistent with democratic deficit or authoritarian international law theories), while diplomatic engagement remains the only robust positive linkage, and migration effects are highly contingent on stock levels.