This post is specifically about academic research, where peer review provides a form of external audit of scientific claims and methods. The dynamics discussed here may not apply in the same way to other professional environments.
When you are surrounded by brilliant researchers, following their lead is often the easiest thing to do. They are senior for a reason. They have seen more papers, more failed ideas, more experiments, and more research cycles than you have. And if you are the more junior person in the room, impostor syndrome does the rest. There is therefore a very natural temptation to confuse seniority with correctness.
Most of the time, listening to experienced researchers is a very good strategy.
Until one day, it isn’t.
At some point in research, you may encounter a particularly uncomfortable situation: you look at the experiments, the data, or the failure cases, and slowly come to the conclusion that your PI, mentor, or collaborator is technically wrong. Not incompetent. Not generally wrong. Wrong about this particular thing.
This should not be surprising. A senior researcher may have more scientific experience than you while having spent less time on the specific problem in front of you. You may have inspected examples they have never seen. You may know every strange corner case of the dataset while they interact with the project for an hour a day, at best, or an hour a month, at worse.
And yet disagreeing can still feel extraordinarily difficult. Perhaps they sound certain. Perhaps they have already constructed a coherent explanation of what is happening. Perhaps you have heard how they react when other people challenge them. Perhaps you would simply prefer to remain in their good books. Or perhaps, you are the one who is wrong.
So you hesitate.
How wrong are they?
Is the approximation acceptable?
Does the distinction even matter?
Is this really a battle worth fighting for?
The easiest solution is often to agree and move on. And sometimes that is perfectly reasonable.
Until you submit the paper.
In a double-blind review process, several researchers who have no reason to protect your internal hierarchy will inspect the work. They may miss the limitation, you should not assume reviewers are omniscient, but they are also free to ask the question that nobody inside the project wanted to ask. And if the disagreement concerns something scientifically important, “my collaborator was convinced” will not be a particularly useful rebuttal.
At that point, the question is no longer who had more authority. The question is who had the evidence.
But what counts as evidence?
In experimental research, the answer can itself become debatable.
Is a qualitative observation enough?
Can the phenomenon be quantified properly?
Is the metric improved globally but fails locally, precisely where the interesting behavior occurs?
How many examples are necessary before an observation becomes a pattern rather than a confirmation bias?
Research disagreement is not about winning an argument. It is about nuancing the parts of the argument that depend on status, intuition, eloquence, or confidence until something remains that can be inspected by other people. In the best-case scenario, the two interpretations converge into a more complete explanation.
But then there is another complication: there might be something else at stake for everyone involved. A deadline. A current grant proposal. A recommendation for a future position. An authorship decision. The question of who will continue this line of research after the project ends. None of this automatically means that anyone is acting in bad faith. But it does mean that changing one’s mind becomes more expensive.
There is also a reputational dimension. Publishing a strong claim without sufficiently investigating evidence that makes it fragile can damage scientific credibility.
But the opposite failure mode exists too.
If you remove every strong claim simply because you cannot prove it beyond any possible doubt, you may end up stripping the work of part of its scientific contribution. Research is not only about reporting measurements, it is also about interpreting them. A good framing can itself be part of the novelty, because it tells the reader what phenomenon matters, how seemingly disconnected observations relate to one another, and why the results should change the way we think about the problem.
This is why learning to disagree technically matters.
The useful response to “I think this interpretation is wrong” is not necessarily immediate agreement or disagreement. It can be: Show me.
Show me the failure cases. Show me the distribution. Show me whether the effect persists across seeds. Show me what happens when we isolate the variable. Show me a metric that captures what the current metric misses. Show me an experiment where my explanation and yours make different predictions.
And, crucially, be willing to stop the debate and accept the answer. Being proven wrong does not mean that you are incompetent. It means you are learning from an ambiguous situation, and sometimes that is precisely how you earn respect and build trust with your collaborators.
Being able to disagree is part of becoming a scientist.
Being able to be disagreed with is part of becoming a good scientific leader.
Both are worth a try :)
And later, you may realise something else: good PIs, mentors, collaborators and reviewers are the people who train you to disagree in the first place. They challenge your assumptions, push on weak arguments, ask for stronger evidence, and sometimes take the opposite side simply to see whether your reasoning holds. They may even pretend they don’t know a well-known fact, or play the bad cop.
And then the moment comes when you need to face a giant and look them straight in the eyes and tell them, with a bit of confidence: I believe this is incorrect.
Come with evidence, overcome your impostor syndrome, take a deep breath, and be strong!
Disagreement can have consequences. So can staying silent when you observe something is wrong.
It is all part of the training.