In recent years, I have often written about the replication crisis in psychology. This development, which I see as a welcome correction, is difficult to separate from another longstanding feature of academic life: publish or perish. Universities still largely evaluate scientists based on their publication record. The unintended consequences of that system have is subject of debate for decades. I am not referring here to the occasional high-profile scandal. Rather, I am referring to a much more ordinary phenomenon. Studies with a higher chance of publication are more likely to be written up, submitted, and ultimately published.

Researchers call this the file drawer problem. Imagine ten research teams investigating the same intervention. Eight find little or no effect, while two find positive results. Those two studies will often have a much better chance of appearing in a journal. Journals may favour statistically significant findings, but researchers themselves may also be less inclined to write up and submit null results. Together, these decisions create a distorted picture of reality.

Anyone familiar with discussions about publication bias will recognise this story. Researchers have known for a long time that publication bias exists. The more interesting question is not whether studies end up in the drawer, but how large that drawer really is and what exactly it contains.

A new meta-analysis by Peter Bergman and Nat Chowanajin attempts to uncover exactly that. To do so, they examined research into a specific category of interventions: low-cost ways to involve parents more in their children’s education via text messages, phone calls, or apps. Interesting in itself, but what makes this study truly special is that the researchers did not just collect published studies. They also actively sought out studies that had been conducted but never written up or published.

That proved to be no easy task. They searched research registers, grant databases, reports from research organisations, and even contacted researchers. Ultimately, they found 82 randomised studies from more than twenty countries. Notably, 24 of them had never been published.

At first glance, that seems like a strong confirmation of the file drawer problem. But as is so often the case, the story becomes more interesting when you look deeper. The question is not only how many studies have been left in the drawer. But, more importantly, which studies are they?

To this end, the authors developed a statistical model to reconstruct what is in that drawer. Their conclusion is surprisingly nuanced. Yes, there is publication bias. Studies with statistically significant results are published more often than studies without significant results. But the hidden studies do not appear to contain massive negative effects. What is primarily missing are studies with smaller effects. And that is an important difference from the idea that sometimes exists regarding publication bias.

The caricature of publication bias is often that researchers hide a closet full of failed experiments while only the success stories see the light of day. This research suggests a different picture. The published literature appears to somewhat overestimate the effects, but does not completely overturn them. When the authors correct for the missing studies, the average effects remain positive. They only become slightly smaller.

That is actually good news for those who take science seriously. Not because publication bias isn’t a problem—it certainly remains one. But because it shows that the scientific system sometimes works better than both the biggest optimists and the biggest sceptics think. The truth turns out, once again, to lie somewhere in the middle.

But there is more. We often speak of “research” as if it were a finished product. In reality, research is a process. Some studies get published. Others remain unpublished because researchers change jobs, funding stops, there is no longer time to complete a report, or simply because no one gets around to it. That does not automatically mean that those results are worthless. It does mean, however, that a good overview of the literature requires more than just searching for published articles.

It may be stating the obvious. But science does not consist solely of what it publishes. Sometimes it is more than worth the effort to take a look at what has been left in the drawer. Apparently, you won’t necessarily find hidden revolutions or conspiracies there, but you will find a slightly more complete picture of reality.