Overview

This document presents the methodological framework for analyzing the impact of targeted assassinations on Iranian nuclear scientists’ research output through bibliometric analysis.

Step 1: Initial Population

Starting Point

Metric Value
Total Scientists 17
Time Span 2007-2025 (18 years)
Event Types Confirmed & suspected assassinations
Source Table 7

Description

Scope: All Iranian nuclear scientists who were assassinated with definitive confirmation or strong international consensus attributing responsibility to foreign government agents.

The initial dataset includes:

  • Scientists targeted between 2007 and 2025
  • Both successful assassinations and attempted assassinations
  • Scientists with varying levels of seniority and expertise
  • Multiple institutions across Iran

Step 2: Filtering

Filtering Criteria

To ensure analyzable cases with sufficient data, we applied two key filters:

Filter 1: Temporal Constraint

  • Requirement: Assassination occurred before 2021
  • Rationale: Need 5+ years of post-assassination data for robust pre/post comparison
  • Result: Excludes 2025 strikes (too recent for analysis)

Filter 2: Bibliometric Footprint

  • Requirement: At least 1 OpenAlex author profile with ≥5 publications
  • Rationale: Insufficient publications = insufficient topic data for modeling
  • Result: Ensures adequate signal for statistical analysis

Final Sample

Scientist Year Expertise Institution
Ardeshir Hosseinpour 2007 Electromagnetism Shiraz University
Masoud Ali-Mohammadi 2010 Quantum Physics University of Tehran
Majid Shahriari 2010 Neutron Transport Shahid Beheshti University
Fereydoon Abbasi-Davani 2010 (survived) Nuclear Physics Shahid Beheshti University

Step 3: Author Identification

Challenge: Name Disambiguation

Problem: Persian names have multiple English transliterations, creating multiple potential OpenAlex profiles per scientist.

Example: “Fereydoon”, “Fereidon”, “Freidoon”, “Ferydoon” Abbasi-Davani

Process

  1. Exhaustive search of all name variants in OpenAlex
  2. Manual verification to eliminate false positives
  3. Physics expert cross-validation for high-profile cases (Abbasi-Davani & Shahriari)
  4. De-duplication of identical works across profiles
Table 9: Author Profile Summary
Scientist OpenAlex Profiles Total Works Citations H-index
A. Hosseinpour 2 11 474 7
M. Ali-Mohammadi 2 69 895 19
M. Shahriari 4 59 722 21
F. Abbasi-Davani 10 127 429 17

Detailed Profile Information

  • Tables 11-14 provide exhaustive lists of all OpenAlex author IDs per scientist
  • Each profile includes: alternate names, institutions, ORCID (if available), metrics
  • Profiles were carefully vetted to ensure accuracy

Step 4: Institution Mapping

Methodology

To identify the primary institutional home for each scientist:

  1. Examined institutional affiliations listed on all publications
  2. Aggregated affiliation frequency across each scientist’s body of work
  3. Selected institution with strongest/longest affiliation

Context

Table 10 shows the distribution of 403 Iranian institutions in OpenAlex: - Ranges from 106,779 works (University of Tehran) to smaller institutions - Top 30 institutions account for majority of Iranian scientific output

Institution Scientist Field Total_Works Note
Shiraz University A. Hosseinpour Electromagnetism 35,722
University of Tehran M. Ali-Mohammadi Quantum Physics 106,779
Shahid Beheshti University M. Shahriari + F. Abbasi-Davani Nuclear Physics 35,830 ⚠ ‘Double-Hit’ Scenario

Special Note: Shahid Beheshti University

Unique “Double-Hit” Scenario: - Two scientists (Shahriari and Abbasi-Davani) targeted at same institution on same day (Nov 29, 2010) - Shahriari was killed; Abbasi-Davani survived - Creates conditions that might simultaneously amplify and attenuate disruption


Step 5: Topic Extraction

OpenAlex Topic System

How it works: - Each work assigned up to 3 topics by algorithm - Topics determined from: title, abstract, citations, journal name - Hierarchical structure: Topic → Subfield → Field → Domain - Accuracy: ~53% for top-1 prediction; ~73% for top-10

Data Transformation

  1. Identified all publications by each scientist
  2. Extracted all assigned topics (up to 3 per work)
  3. Converted to long format (one row per work-topic pair)
  4. Compiled exhaustive list of unique topics per scientist

Result: Complete map of each scientist’s research domains

Table 4: Topic Portfolio Summary
Scientist Unique Topics Example Topics
A. Hosseinpour 9 Magnetic Properties of Ferrites, Electromagnetic wave absorption
M. Shahriari 44 Radiation Effects and Dosimetry, Nuclear Physics and Applications
M. Ali-Mohammadi 46 Black Holes and Theoretical Physics, Quantum Information
F. Abbasi-Davani 109 Particle accelerators, Gyrotron Research, Nuclear reactor physics

Topic Portfolio Characteristics

Note: Lower topic counts (e.g., Hosseinpour’s 9) create greater variance and less precise model estimates. Larger portfolios (e.g., Abbasi-Davani’s 109) enable more robust statistical analysis.


Step 6: Treatment Definition

Operationalizing “Treated” vs “Control” Areas

For each case study, we defined which topic areas were “exposed” to assassination (treated) and which were not (control).

Case-by-Case Definitions

Case Institution Treated Definition Treated_Works Control_Works
Cases 1 & 2 Shahid Beheshti University Core Nuclear Physics Areas INTERSECTION of Shahriari ∩ Abbasi topics 1,240 33,750
Case 3 University of Tehran Quantum Physics Areas All Ali-Mohammadi topics 3,450 100,834
Case 4 Shiraz University Electromagnetism Physics Areas All Hosseinpour topics 217 34,807

Special Logic for Shahid Beheshti University

Since two scientists were targeted at Shahid Beheshti on the same day, we used a stricter definition:

  • Treated = Topics that BOTH Shahriari AND Abbasi-Davani published in
  • Logic: Intersection represents topics most relevant to nuclear program
  • Result: More conservative estimate of exposed areas

Summary (Table 5): All treated vs. control definitions established to enable within-institution comparisons that control for institutional variance.


Step 7: Statistical Analysis

Data Structure

Element Description
Unit of Analysis Topic-Month-Year
Time Window 10 years: 5 years pre + 5 years post assassination
Outcome Variable Publication counts per topic-month-year
Model Type Difference-in-Differences (Poisson regression)

Comparison Strategies

Two complementary approaches:

  1. Within-Institution (Cases 1, 3, 4)
    • Compare treated topics vs. control topics within same university
    • Controls for: Institutional factors, funding, infrastructure, publishing culture
  2. Between-Institution (Case 2)
    • Compare same treated topics across Shahid Beheshti vs. Shahid Beheshti Medical Sciences
    • Controls for: Topic-specific trends, field cadence, collaboration norms

Difference-in-Differences Model

Model Specification:

\[Y_{it} = \beta_0 + \beta_1 \cdot Treatment_i + \beta_2 \cdot Post_t + \beta_3 \cdot (Treatment \times Post)_{it} + \varepsilon_{it}\]

Where: - \(Y_{it}\) = publication count for topic \(i\) at time \(t\) - \(Treatment_i\) = indicator for treated topic areas - \(Post_t\) = indicator for post-assassination period - \(\beta_3\) = differential effect of assassination on treated vs. control areas

Key Features: - Accounts for baseline growth rates across all fields - Isolates effect of external shock (assassination) - Can detect subtle slowdowns or plateaus, not just absolute decreases - Standard errors clustered by topic


Summary

Complete Methodological Pipeline

Pipeline Summary Table

Step Input Output Key_Decision
  1. Initial Population
All assassinations 17 scientists Historical records
  1. Filtering
17 scientists 4 scientists Pre-2021 + ≥5 pubs
  1. Author ID
4 scientists 4-26 profiles Manual verification
  1. Institutions
Author profiles 3 institutions Frequency
  1. Topics
3 institutions 9-109 topics each OpenAlex algorithm
  1. Treatment
Topic portfolios Treated/Control Intersection for SBU
  1. Analysis
4 case studies DiD estimates 10-year windows

Key Methodological Strengths

  1. Systematic Selection
    Rigorous filters ensure analyzable cases with sufficient data for robust analysis

  2. Robust Disambiguation
    Manual verification + expert validation prevents false matches and ensures accuracy

  3. Multiple Comparisons
    Within and between-institution designs triangulate effects and control for confounds

  4. Controlled Analysis
    Accounts for institutional factors, topic-specific trends, and baseline growth rates

  5. Transparent & Reproducible
    All decisions documented, open-source data, replicable methods

:::