Most Network Science Studies Can't Be Rechecked — And That Affects You

Here's the problem in plain terms. Scientists who study networks — think how a post goes viral, how a rumor spreads through a town, or how a disease jumps between people — publish papers describing their findings. But according to a new paper by researcher Akrati Saxena, they usually don't share the two things you'd need to check their work: the computer code they wrote and the data they used. It's like publishing a cake recipe that says "mix ingredients, bake" and never telling anyone the amounts.

The author looked at four different kinds of network research, from classic math-based methods to newer AI-driven ones. The results were consistent across all four: code and data were usually missing. Saxena points to several causes. Scientists get little credit or reward for sharing their work, some data is legally or commercially restricted, and the steps involved are so complicated that writing them down fully is genuinely hard.

Why should you care? These aren't abstract studies. The same techniques power recommendation engines, fraud detection, public health modeling and the algorithms that decide what you see online. If the underlying research can't be verified, errors and exaggerations can quietly pile up — and companies may build products on results that simply don't hold up. It also means money spent on research may not translate into real-world benefit, because nobody can build on findings they can't reproduce.

Saxena's recommendations are straightforward: require researchers to share their code and data, create standard test sets everyone uses, and demand thorough write-ups of how experiments were run. Some journals and conferences already do this. The wider lesson is that transparency isn't bureaucracy — it's what separates a real discovery from a confident guess.