Five Keys to Producing More and Better Scientific Papers

Scientific papers are the currency of academia. The dream paper tells a compelling story and answers a deep question conclusively. Because it fundamentally shifts the field, it gets cited hundreds of times and its authors get invited to present at authoritative conferences and work with experts from all over the world.

For most of us, however, scientific writing is a slow and painful process. We can’t all be nobel prize winners of course, but there certainly are methods we could adopt that will all but guarantee higher quality papers in less time. Academics are famously known for their ability - habit even - to burn the midnight oil. As romantic as this may be, it often comes at a high price of physical, mental and social wellness. Therefore, becoming a more efficient scientist could quite literally be life saving. Below are five keys to producing more and better scientific papers, and doing less overtime.

Pick a problem that the world cares about

Thanks to technological and ideological shifts - the fast-growing internet of things, cloud computing, FAIR data and a thriving community of analysts and data scientists producing open access R packages and Python modules - young scientists now have access to a universe of (big) data and a massive toolbox of analytics methods and techniques. Endless opportunities for scientific discovery, right? Perhaps. But not all discoveries are equally breathtaking. Before going down the trenches of analysis and scientific writing, make sure you have picked a battle worth fighting. Reading recently published review papers are often a good reality check for whether your research question is indeed right on top of the priority list. So is attending the top conferences in your field. Not only to be inspired or challenged by the keynote lectures, but also to ask the experts what they think of your research question during the tea breaks.

Be a (wo)man with a plan

Asking a meaningful research question is one thing, but designing an analysis plan for answering it is another. If your dataset contains dozens or even hundreds of variables, you can slice the data in as many ways. Broadly speaking you can take one of two approaches: either you see your paper as the output of a premeditated analysis plan. You have declared upfront that the answer to the research question equates to the calculation of a number of predefined metrics, effect sizes, trends and associations. Alternatively, you only have an exploratory data analysis plan, and you allow yourself to make additional analysis decisions on the fly, driven by what a previous analysis step reveals or suggests. Both approaches may be valid, depending on the context, but the latter carries several risks. Besides the risk of giving undue attention to spurious noise in the data, doing the analysis in an entirely "data-driven" way can lead to a never ending stream of new ideas, a steep learning curve to master additional analysis methods, and a myriad of piecemeal results. The recipe for a monster paper that fails to tell a compelling story and never sees the finish line.

Be a team player

How often can you say with confidence that you are the smartest or the most experienced person in the team? If you can, make your team bigger and ask again. The point is that we arrive at better analysis plans, produce more insightful graphics and write clearer paragraphs when we do it as a team. The downside of course is that collaborative work requires jointly agreed rules about roles and responsibilities, and modes of engagements. For example, code review by a more experienced colleague can expose and correct errors and boost the readability of the code, but if not adequately regulated it can also be the reason for being held hostage in a tug of war over irrelevant style preferences. Sharing data, code and manuscripts also means that you need a method to handle the madness that is version control. The last thing you want is to have an untold number of different versions of the dataset, scripts and manuscript drafts floating around on various local hard drives and cloud platforms. Git is the standard for version control of code and has become a mission-critical skill for productive scientists. DVC is a git-like tool for version control of data. With Google’s G Suite, Overleaf and Microsoft’s Office 365, you really don’t have an excuse anymore for emailing manuscripts around.

Become a visual artist

Humans are visual creatures. I might not remember the exact numbers, but I can still picture the figures of key papers that caused a breakthrough in my understanding. Graphics can be used to show the conceptual framework and logical reasoning behind the hypotheses under investigation, they can bring the raw data to life, and they can turn abstract model inferences into insightful visualisations of their meaning and implications. It is often possible and a good idea to free-hand sketch what you think the key figures of the paper might look like. It’s a guaranteed boost of your confidence when you guessed correctly, and an opportunity to learn and investigate deeper when you got it all wrong. Much like ballet is a visual spectacle, supported and enhanced by music, so you could see your paper: a visual storyboard accompanied by text that serves to provide context, and discuss meaning.

Be a writer first, and editor last

Scientists want to get it exactly right. From the first time. The problem with that, is that premature wordsmithing and editing can seriously hamper you ability to put down the essence of the story. A better approach is to mind map first. It allows you to engage in the creative process and share a first draft with your co-workers fast. The sooner you can discuss the draft paper with others, the sooner you can learn whether the flow of the story needs some rethinking, which sections need more (or less) explanation, and whether your take on what the study findings mean in the context of the original research question is objective and supported by the evidence or conjecture based on a cherry-picked selection of the study results.

None of the above five tips are never-heard-before epiphanies. Most scientists live by at least one or two of them, but few consistently embrace all. Perhaps it’s the nature of the beast. Academia is a fiercely competitive environment. Pressure to publish may drive us to go for projects that seem quick wins and low hanging fruit. That same publication pressure may lead us to favour novel methods over standard ones. When you feel like you’re in a career-long competition with each other, it’s not obvious to collaborate, share code, ask for help and encourage others to find mistakes in your work. Presenting results in a few tables or two-dimensional plots takes less time than crafting a storyboard of nuanced yet intelligible figures. And our innate tendency to seek novelty and perfection may stifle the efficiency and quality of our publication output. However, if we dare to take a few steps back and aim to master all five of these ideas, we’re on a path to success. In this context, defined as conducting great science without sacrificing our mental and social health.