Working Paper Series · No. 26-01 · This version: 18 November 2026
Messy Data, Causal Answers, and Software That Ships: Evidence from One Applied Economist
I am an applied economist and data scientist, finishing an M.Sc. in Applied Economics and Data Science in Oldenburg, Germany. I build panels from satellite and administrative data, replicate published results on independently constructed pipelines, and turn the boring parts into tools that run on their own: an R package on CRAN, a retrieval-augmented assistant in production, a monthly data pipeline to Kaggle. Using 1709 GitHub contributions as data, I find that I ship consistently, with a statistically significant drop on weekends (§3). Policy implication: hire me.
JEL codes C23, C55, C87, O18, R11
Keywords causal inference, panel data, GIS, reproducibility, data engineering, RAG
Cite this person
@misc{coban2026hire,
author = {\c{C}oban, Furkan},
title = {Messy Data, Causal Answers, and Software That Ships},
howpublished = {\url{https://ofurkan.co}},
year = {2026},
note = {Working Paper No. 26-01. Open to work.}
}* Carl von Ossietzky Universität Oldenburg. I thank R, Quarto and an unreasonably patient VPS for helpful comments. Correspondence: see the end of this page.
- [1]Regional Favoritism, RebuiltA from-scratch replication of Hodler & Raschky (2014, QJE): do regions glow brighter at night while their native son holds power? Extended with a harmonised DMSP/VIIRS panel through 2023.R · GIS · Nighttime lights · Panel econometrics · QuartoFig. A.1
- [2]inkaRAn R package on CRAN that turns the BBSR INKAR regional statistics API into tidy data frames and publication-ready maps of Germany.R · CRAN · sf · ggplot2 · REST APIFig. A.2
- [3]AEDS RAG AssistantA retrieval-augmented chat assistant that answers students' questions about a master's programme strictly from its official documents, with citations, a relevance gate, and staff review.Python · FastAPI · LangGraph · Chroma · BM25Fig. A.3
- [4]Urban Land-Use EfficiencyWhy the official SDG 11.3.1 sprawl indicator is statistically unstable, what a log-based replacement recovers, and how per-capita designs can manufacture the drivers they claim to find. 193 countries, 1985 to 2020.R · Google Earth Engine · GHSL · Dynamic panel GMM · Two-way FEFig. A.4
- [6]R Plot ProA VS Code extension that brings an RStudio-style plots pane to R and Julia: automatic capture, no focus-stealing windows.TypeScript · VS Code API · R · JuliaFig. A.6
- [7]German Power Price ForecastingForecasting hourly day-ahead electricity prices in Germany (2020 to 2026) around the merit-order effect, from automated ETL to ARIMAX vs Ridge vs XGBoost.Python · XGBoost · statsmodels · scikit-learn · SMARD APIFig. A.7
Claims about productivity should come with data. Below is every public contribution I made on GitHub between Nov 2025 and Nov 2026, pulled from the GitHub API when this page was built, and treated the way I would treat any other dataset.
| N | Mean | SD | Median | P90 | Max | Share 0 | |
|---|---|---|---|---|---|---|---|
| Daily contributions | 368 | 4.64 | 11.29 | 1 | 15 | 102 | 0.37 |
| Weekly contributions | 53 | 32.25 | 44.11 | 13 | 92 | 207 | 0.11 |
Main resultRelative to a Monday, a weekend day sees 7.1 fewer contributions on average (column 1). The effect survives a linear trend, month fixed effects and a lagged dependent variable. I code on weekdays and rest on weekends, and now there are standard errors to prove it.
| (1) | (2) | (3) | (4) | (5) Your spec | |
|---|---|---|---|---|---|
| Tuesday | -0.23 | -0.26 | -0.37 | -2.12 | |
| (2.87) | (2.73) | (2.63) | (2.45) | ||
| Wednesday | -1.58 | -1.65 | -1.64 | -3.36 | |
| (2.99) | (2.82) | (2.68) | (2.90) | ||
| Thursday | -2.06 | -2.05 | -2.06 | -3.46* | |
| (2.67) | (2.49) | (2.30) | (2.07) | ||
| Friday | -3.03 | -3.04 | -3.06 | -4.30* | |
| (2.42) | (2.32) | (2.23) | (2.26) | ||
| Saturday | -7.33*** | -7.38*** | -7.32*** | -8.31*** | |
| (2.21) | (2.15) | (2.04) | (2.05) | ||
| Sunday | -6.81*** | -6.78*** | -6.74*** | -6.68*** | |
| (2.29) | (2.21) | (2.08) | (2.07) | ||
| Trend (per 100 days) | 3.25*** | 9.07 | 6.38 | ||
| (0.59) | (6.81) | (6.76) | |||
| Contributions, t-1 | 0.26** | ||||
| (0.11) | |||||
| Constant | 7.64*** | 1.70 | 2.64 | 3.55* | |
| (2.20) | (2.10) | (1.92) | (1.94) | ||
| Month fixed effects | No | No | Yes | Yes | |
| Observations | 368 | 368 | 368 | 367 | |
| R² | 0.060 | 0.153 | 0.220 | 0.273 |
Monday is the omitted category. Heteroskedasticity-robust (HC1) standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01. Estimated with a 130-line OLS routine written for this page, cross-checked against R's lm().
Data: GitHub GraphQL API, retrieved 18 Nov 2026. Replication code: scripts/fetch-github.mjs.