Demonstration article

This sample was written to show how Northstar Briefing presents journalism. It does not report real events and contains no real quotations, data or named sources.

Science advances by building on earlier work. That only works if the earlier work holds up. Over the past two decades, efforts to repeat published experiments in several fields have found that a worrying share of results are harder to reproduce than their original papers suggested.

The causes are usually mundane rather than fraudulent. Small sample sizes make chance findings more likely. Flexible analysis — trying several statistical approaches and reporting the one that works — inflates apparent effects. And journals have historically been more interested in novel, positive results than in careful replications or null findings.

Changing the incentives

Reformers have proposed practical fixes. Pre-registration asks researchers to publish their hypotheses and analysis plans before collecting data. Registered reports go further: journals agree to publish a study based on its design, regardless of the outcome. Data and code sharing let others check the work directly.

A surprising result is a reason to look again, not a reason to stop looking.

None of these practices is free. They take time, and they can feel like bureaucracy to researchers under pressure to publish. Supporters argue the cost is small compared with the waste of building years of work on findings that later dissolve.

How to read a single study

Look for the sample size and whether the result was pre-registered.

Check whether other teams have found the same thing.

Be cautious when a small study reports a very large effect.