The Industrial Mentor: Knowledge Spillovers from High-Complexity Sectors

Event study analysis of knowledge spillovers through mentor arrivals in Italian firms, using matched employer-employee data from the Veneto region (1975–2014). Tests whether mentors from high-complexity sectors (manufacturing, energy, mining) generate larger wage gains for incumbent workers than mentors from low-complexity sectors.

Research Question

Do firms benefit from knowledge spillovers when experienced high-ability workers (mentors) arrive, and does the complexity of their previous sector matter?

Methodology

Data

• Veneto Workers History (VWH): Matched employer-employee panel covering all private-sector workers in the Veneto region, Italy (1975–2014)
• Variables: Daily wages, firm identifiers, worker demographics, sector codes (ATECO81), contract details

Stage 1: AKM Wage Decomposition

Abowd-Kramarz-Margolis (1999) fixed effects model:

log(wage) = worker_FE + firm_FE + year_FE + f(age) + residual

Decomposes wages into worker ability, firm productivity, and time effects. Variance decomposition quantifies the relative contribution of each component.

Stage 2: Mentor Identification

• Source firms: Top 20% by both firm size and firm fixed effects
• Mentors: High-ability workers (top 25% worker FE) with 3+ years at source firms
• Complexity classification: Binary — high-complexity (ATECO81 divisions 2–3: manufacturing, energy, mining) vs. low-complexity (all other sectors)

Stage 3: Event Study (Staggered Difference-in-Differences)

• Treatment: First arrival of an activated mentor at a non-source firm
• Treated group: Incumbent workers present before mentor arrival
• Control group: Never-treated workers at non-source firms
• Estimator: Sun & Abraham (2021) heterogeneity-robust staggered DID (sunab)
• Specification: Firm + year fixed effects, weighted by number of incumbents

Robustness Checks

• Any-link mentor definition (relaxing main-job restriction)
• Firm-year level aggregation
• Burn-in sensitivity analysis (left-censoring at 1975 panel start)

Key Findings

1. Knowledge spillovers exist: Incumbent workers at firms receiving mentors experience wage gains relative to never-treated controls, consistent with knowledge transfer from experienced workers.

2. Sector complexity matters: Mentors from high-complexity sectors (manufacturing, energy, mining) generate larger spillover effects than those from low-complexity sectors, supporting the hypothesis that technical knowledge is more transferable.

3. Parallel trends validated: Pre-treatment coefficients (t=-3 to t=-1) are close to zero, confirming the parallel trends assumption required for causal interpretation.

4. Gradual knowledge transfer: Post-treatment effects show an increasing pattern, consistent with knowledge being absorbed and applied over time rather than instantaneously.

5. Results are robust: Findings hold across alternative mentor definitions (any-link), aggregation levels (firm-year), and burn-in specifications addressing left-censoring concerns.

Repository Structure

File	Description
knowledge_spillovers.Rmd	Full R Markdown analysis — AKM estimation, event study, robustness checks
final_paper.pdf	Final research paper with results and conclusions
supplementary_analysis.pdf	Supplementary analysis notebook

Tools & Packages

Language: R (RMarkdown)

Key packages: data.table · fixest · lfe · haven · ggplot2 · knitr

How to Reproduce

1. Install R (>= 4.0) and RStudio
2. Install required packages:

   install.packages(c("data.table", "fixest", "lfe", "haven", "ggplot2", "knitr", "tictoc"))

3. Place VWH data files (anagr.dta, contr.dta, azien.dta) in a VWH/ subdirectory
4. Open knowledge_spillovers.Rmd in RStudio
5. Knit the document (Ctrl+Shift+K)

Note: The VWH dataset is restricted-access administrative data. Contact the original data providers for access.

Author

David Stahl