Every year we invite you to ask us anything! Today’s question is actually two closely-related questions we’ve combined into one, from Fernanda Henrique and sagitanita (paraphrased; click that link for the originals):
How do you introduce and write a descriptive or exploratory paper when you don’t have a specific hypothesis, or a statistical answer to your question?
Let me start by saying I completely agree with the premise of this question–that it is very bad practice to dress up descriptive/exploratory papers if they’d been planned as hypothesis-testing papers all along. At worst, it leads to papers that are actively misleading–that look like severe hypothesis tests when in fact the hypothesis couldn’t possibly have failed the “test” because it was developed to fit the data being used to test it. At best, it leads to vague, pointless “pseudo-hypotheses” like “variable X will vary with variable Y”. And in a recent survey, our readers said that dressing up exploratory analyses as if they were pre-planned hypothesis testing analyses is perhaps the most serious problem in ecological research.
But at least when it comes to paper writing, you don’t have to pretend to be doing hypothesis testing in order to get published. Honestly, you don’t! There are various sorts of non-hypothesis-testing papers one can write. Like hypothesis-testing papers, their structures tend to follow some standard “templates”. Here are a few to get you started, hopefully commenters will chime in with others. I hope that one of these fits your next descriptive/exploratory paper!
- Here’s an unusual/weird natural history observation, that’s interesting just because it’s unusual/weird. Who doesn’t like hearing about the first ever observation of an albino jackalope, or a python that burst after eating an alligator, or etc.? 🙂
- Here’s a natural history observation that’s interesting because it suggests some new hypothesis/insight/broader implication that could be further pursued. That’s the structure of most of the “natural history notes” that Ecology and Am Nat started publishing recently.
- Here’s a descriptive statistical summary of a bunch of data, that identifies a new ubiquitous pattern in need of theoretical explanation. That’s the structure of Hatton et al. 2015 Science, one of the best and more important ecology papers of recent years.
- Here’s a descriptive statistical or graphical summary of a bunch of data, and what it reveals/suggests. This is a variant of the previous template, the difference being what’s revealed/suggested is something other than a ubiquitous pattern in need of theoretical explanation. Many high-profile papers about global change have this structure. Think of papers with titles like “A high-resolution global map of forest loss”. Many papers that report new phylogenies have this structure as well–here’s a new phylogeny of [taxonomic group] and what it reveals/suggests about the group’s evolutionary history. Note that the introduction to this sort of paper (and the sort described in the previous bullet) will need to offer some reason to compile and analyze these data. But having a good reason to describe/explore/summarize a dataset isn’t the same as having a hypothesis about the results of that description/summary/exploration. It’s common for authors to mistakenly think they have to have a hypothesis when all they need is a good reason.
- Here’s a bit of data, which doesn’t mean much on its own but would be useful as part of some much larger compilation of data from other places/times/species/study systems. This sort of paper is only publishable in an unselective venue, or perhaps best published not as a paper but merely by adding the data to a relevant database (e.g., uploading sightings of birds into eBird).
Related: Brian’s old post on exploratory statistics, which has a very good comment thread too.
The key is you have to convince the reader you are doing interesting science. A hypothesis test is one flavor of interesting science. But far from the only flavor. I like Jeremy’s list. I would add one more: estimating an important quantity. These range from papers in Science or Nature on, e.g. total carbon uptake of the world’s forests or the average trend in local biodiversity, to regional conservation journals on estimating the abundance of a specific species and population under management.
But the flipside is not every paper with data and analysis is interesting. And no amount of dressing it up is going to hide that.
I tend to like to focus on questions rather than hypotheses. Do you have a good question that motivated the data you collected and your analysis? Is it a question other scientists wonder about and want answered? If so you’re probably fine. Just write your paper with a good story arc.
If you just collected the data because you had to spend time in the field, didn’t really have time to get organized, and don’t know what question your data attacks, well, then, … I don’t know what to say.
So the real priority should be on identifying good questions before collecting and analyzing data. If you do that, writing the paper really is no different than a hypothesis paper. Anchor your question in a general challenge, state your specific question, describe how you answered it, and then broaden back out.