Month: March 2014

what exactly is it that 53% of neuroscience articles fail to do?

[UPDATE: Jake Westfall points out in the comments that the paper discussed here appears to have made a pretty fundamental mistake that I then carried over to my post. I’ve updated the post accordingly.]

[UPDATE 2: the lead author has now responded and answered my initial question and some follow-up concerns.]

A new paper in Nature Neuroscience by Emmeke Aarts and colleagues argues that neuroscientists should start using hierarchical  (or multilevel) models in their work in order to account for the nested structure of their data. From the abstract:

In neuroscience, experimental designs in which multiple observations are collected from a single research object (for example, multiple neurons from one animal) are common: 53% of 314 reviewed papers from five renowned journals included this type of data. These so-called ‘nested designs’ yield data that cannot be considered to be independent, and so violate the independency assumption of conventional statistical methods such as the t test. Ignoring this dependency results in a probability of incorrectly concluding that an effect is statistically significant that is far higher (up to 80%) than the nominal α level (usually set at 5%). We discuss the factors affecting the type I error rate and the statistical power in nested data, methods that accommodate dependency between observations and ways to determine the optimal study design when data are nested. Notably, optimization of experimental designs nearly always concerns collection of more truly independent observations, rather than more observations from one research object.

I don’t have any objection to the advocacy for hierarchical models; that much seems perfectly reasonable. If you have nested data, where each subject (or petrie dish or animal or whatever) provides multiple samples, it’s sensible to try to account for as many systematic sources of variance as you can. That point may have been made many times before,  but it never hurts to make it again.

What I do find surprising though–and frankly, have a hard time believing–is the idea that 53% of neuroscience articles are at serious risk of Type I error inflation because they fail to account for nesting. This seems to me to be what the abstract implies, yet it’s a much stronger claim that doesn’t actually follow just from the observation that virtually no studies that have reported nested data have used hierarchical models for analysis. What it also requires is for all of those studies that use “conventional” (i.e., non-hierarchical) analyses to have actively ignored the nesting structure and treated repeated measurements as if they in fact came from entirely different subjects or clusters.

To make this concrete, suppose we have a dataset made up of 400 observations, consisting of 20 subjects who each provided 10 trials in 2 different experimental conditions (i.e., 20 x 2 x 10 = 400). And suppose the thing we ultimately want to know is whether or not there’s a statistical difference in outcome between the two conditions. There are three at least three ways we could set up our comparison:

  1. Ignore the grouping variable (i.e., subject) entirely, effectively giving us 200 observations in each condition. We then conduct the test as if we have 200 independent observations in each condition.
  2. Average the 10 trials in each condition within each subject first, then conduct the test on the subject means. In this case, we effectively have 20 observations in each condition (1 per subject).
  3. Explicitly include the effects of both subject and trial in our model. In this case we have 400 observations, but we’re explictly accounting for the correlation between trials within a given subject, so that the statistical comparison of conditions effectively has somewhere between 20 and 400 “observations” (or degrees of freedom).

Now, none of these approaches is strictly “wrong”, in that there could be specific situations in which any one of them would be called for. But as a general rule, the first approach is almost never appropriate. The reason is that we typically want to draw conclusions that generalize across the cases in the higher level of the hierarchy, and don’t have any intrinsic interest in the individual trials themselves. In the above example, we’re asking whether people on average, behave differently in the two conditions. If we treat our data as if we had 200 subjects in each condition, effectively concatenating trials across all subjects, we’re ignoring the fact that the responses acquired from each subject will tend to be correlated (i.e., Jane Doe’s behavior on Trial 2 will tend to be more similar to her own behavior on Trial 1 than to another subject’s behavior on Trial 1). So we’re pretending that we know something about 200 different individuals sampled at random from the population, when in fact we only know something about 20 different  individuals. The upshot, if we use approach (1), is that we do indeed run a high risk of producing false positives we’re going to end up answering a question quite different from the one we think we’re answering. [Update: Jake Westfall points out in the comments below that we won’t necessarily inflate Type I error rate. Rather, the net effect of failing to model the nesting structure properly will depend on the relative amount of within-cluster vs. between-cluster variance. The answer we get will, however, usually deviate considerably from the answer we would get using approaches (2) or (3).]

By contrast, approaches (2) and (3) will, in most cases, produce pretty similar results. It’s true that the hierarchical approach is generally a more sensible thing to do, and will tend to provide a better estimate of the true population difference between the two conditions. However, it’s probably better to describe approach (2) as suboptimal, and not as wrong. So long as the subjects in our toy example above are in fact sampled at random, it’s pretty reasonable to assume that we have exactly 20 independent observations, and analyze our data accordingly. Our resulting estimates might not be quite as good as they could have been, but we’re unlikely to miss the mark by much.

To return to the Aarts et al paper, the key question is what exactly the authors mean when they say in their abstract that:

In neuroscience, experimental designs in which multiple observations are collected from a single research object (for example, multiple neurons from one animal) are common: 53% of 314 reviewed papers from five renowned journals included this type of data. These so-called ‘nested designs’ yield data that cannot be considered to be independent, and so violate the independency assumption of conventional statistical methods such as the t test. Ignoring this dependency results in a probability of incorrectly concluding that an effect is statistically significant that is far higher (up to 80%) than the nominal α level (usually set at 5%).

I’ve underlined the key phrases here. It seems to me that the implication the reader is supposed to draw from this is that roughly 53% of the neuroscience literature is at high risk of reporting spurious results. But in reality this depends entirely on whether the authors mean that 53% of studies are modeling trial-level data but ignoring the nesting structure (as in approach 1 above), or that 53% of studies in the literature aren’t using hierarchical models, even though they may be doing nothing terribly wrong otherwise (e.g., because they’re using approach (2) above).

Unfortunately, the rest of the manuscript doesn’t really clarify the matter. Here’s the section in which the authors report how they obtained that 53% number:

To assess the prevalence of nested data and the ensuing problem of inflated type I error rate in neuroscience, we scrutinized all molecular, cellular and developmental neuroscience research articles published in five renowned journals (Science, Nature, Cell, Nature Neuroscience and every month’s first issue of Neuron) in 2012 and the first six months of 2013. Unfortunately, precise evaluation of the prevalence of nesting in the literature is hampered by incomplete reporting: not all studies report whether multiple measurements were taken from each research object and, if so, how many. Still, at least 53% of the 314 examined articles clearly concerned nested data, of which 44% specifically reported the number of observations per cluster with a minimum of five observations per cluster (that is, for robust multilevel analysis a minimum of five observations per cluster is required11, 12). The median number of observations per cluster, as reported in literature, was 13 (Fig. 1a), yet conventional analysis methods were used in all of these reports.

This is, as far as I can see, still ambiguous. The only additional information provided here is that 44% of studies specifically reported the number of observations per cluster. Unfortunately this still doesn’t tell us whether the effective degrees of freedom used in the statistical tests in those papers included nested observations, or instead averaged over nested observations within each group or subject prior to analysis.

Lest this seem like a rather pedantic statistical point, I hasten to emphasize that a lot hangs on it. The potential implications for the neuroscience literature are very different under each of these two scenarios. If it is in fact true that 53% of studies are inappropriately using a “fixed-effects” model (approach 1)–which seems to me to be what the Aarts et al abstract implies–the upshot is that a good deal of neuroscience research is very bad statistical shape, and the authors will have done the community a great service by drawing attention to the problem. On the other hand, if the vast majority of the studies in that 53% are actually doing their analyses in a perfectly reasonable–if perhaps suboptimal–way, then the Aarts et al article seems rather alarmist. It would, of course, still be true that hierarchical models should be used more widely, but the cost of failing to switch would be much lower than seems to be implied.

I’ve emailed the corresponding author to ask for a clarification. I’ll update this post if I get a reply. In the meantime, I’m interested in others’ thoughts as to the likelihood that around half of the neuroscience literature involves inappropriate reporting of fixed-effects analyses. I guess personally I would be very surprised if this were the case, though it wouldn’t be unprecedented–e.g., I gather that in the early days of neuroimaging, the SPM analysis package used a fixed-effects model by default, resulting in quite a few publications reporting grossly inflated t/z/F statistics. But that was many years ago, and in the literatures I read regularly (in psychology and cognitive neuroscience), this problem rarely arises any more. A priori, I would have expected the same to be true in cellular and molecular neuroscience.

UPDATE 04/01 (no, not an April Fool’s joke)

The lead author, Emmeke Aarts, responded to my email. Here’s her reply in full:

Thank you for your interest in our paper. As the first author of the paper, I will answer the question you send to Sophie van der Sluis. Indeed we report that 53% of the papers include nested data using conventional statistics, meaning that they did not use multilevel analysis but an analysis method that assumes independent observations like a students t-test or ANOVA.

As you also note, the data can be analyzed at two levels, at the level of the individual observations, or at the subject/animal level. Unfortunately, with the information the papers provided us, we could not extract this information for all papers. However, as described in the section ‘The prevalence of nesting in neuroscience studies’, 44% of these 53% of papers including nested data, used conventional statistics on the individual observations, with at least a mean of 5 observations per subject/animal. Another 7% of these 53% of papers including nested data used conventional statistics at the subject/animal level. So this leaves 49% unknown. Of this 49%, there is a small percentage of papers which analyzed their data at the level of individual observations, but had a mean less than 5 observations per subject/animal (I would say 10 to 20% out of the top of my head), the remaining percentage is truly unknown. Note that with a high level of dependency, using conventional statistics on nested data with 2 observations per subject/animal is already undesirable. Also note that not only analyzing nested data at the individual level is undesirable, analyzing nested data at the subject/animal level is unattractive as well, as it reduces the statistical power to detect the experimental effect of interest (see fig. 1b in the paper), in a field in which a decent level of power is already hard to achieve (e.g., Button 2013).

I think this definitively answers my original question: according to Aarts, of the 53% of studies that used nested data, at least 44% performed conventional (i.e., non-hierarchical) statistical analyses on the individual observations. (I would dispute the suggestion that this was already stated in the paper; the key phrase is “on the individual observations”, and the wording in the manuscript was much more ambiguous.) Aarts suggests that ~50% of the studies couldn’t be readily classified, so in reality that proportion could be much higher. But we can say that at least 23% of the literature surveyed committed what would, in most domains, constitute a fairly serious statistical error.

I then sent Aarts another email following up on Jake Westfall’s comment (i.e., how nested vs. crossed designs were handled. She replied:

As Jake Westfall points out, it indeed depends on the design if ignoring intercept variance (so variance in the mean observation per subject/animal) leads to an inflated type I error. There are two types of designs we need to distinguish here, design type I, where the experimental variable (for example control or experimental group) does not vary within the subjects/animals but only over the subjects/animals, and design Type II, where the experimental variable does vary within the subject/animal. Only in design type I, the type I error is increased by intercept variance. As pointed out in the discussion section of the paper, the paper only focuses on design Type I (“Here we focused on the most common design, that is, data that span two levels (for example, cells in mice) and an experimental variable that does not vary within clusters (for example, in comparing cell characteristic X between mutants and wild types, all cells from one mouse have the same genotype)”), to keep this already complicated matter accessible to a broad readership. Moreover, design type I is what is most frequently seen in biological neuroscience, taking multiple observations from one animal and subsequently comparing genotypes automatically results in a type I research design.

When dealing with a research design II, it is actually the variation in effect within subject/animals that increases the type I error rate (the so-called slope variance), but I will not elaborate too much on this since it is outside the scope of this paper and a completely different story.

Again, this all sounds very straightforward and sound to me. So after both of these emails, here’s my (hopefully?) final take on the paper:

  • Work in molecular, cellular, and developmental neuroscience–or at least, the parts of those fields well-represented in five prominent journals–does indeed appear to suffer from some systemic statistical problems. While the proportion of studies at high risk of Type I error is smaller than the number Aarts et al’s abstract suggests (53%), the latter, more accurate, estimate (at least 23% of the literature) is still shockingly high. This doesn’t mean that a quarter or more of the literature can’t be trusted–as some of the commenters point out below, most conclusions aren’t based on just a single p value from a single analysis–but it does raise some very serious concerns. The Aarts et al paper is an important piece of work that will help improve statistical practice going forward.
  • The comments on this post, and on Twitter, have been interesting to read. There appear to be two broad camps of people who were sympathetic to my original concern about the paper. One camp consists of people who were similarly concerned about technical aspects of the paper, and in most cases were tripped up by the same confusion surrounding what the authors meant when they said 53% of studies used “conventional statistical analyses”. That point has now been addressed. The other camp consists of people who appear to work in the areas of neuroscience Aarts et al focused on, and were reacting not so much to the specific statistical concern raised by Aarts et al as to the broader suggestion that something might be deeply wrong with the neuroscience literature because of this. I confess that my initial knee-jerk impression to the Aarts et al paper was driven in large part by the intuition that surely it wasn’t possible for so large a fraction of the literature to be routinely modeling subjects/clusters/groups as fixed effects. But since it appears that that is in fact the case, I’m not sure what to say with respect to the broader question over whether it is or isn’t appropriate to ignore nesting in animal studies. I will say that in the domains I personally work in, it seems very clear that collapsing across all subjects for analysis purposes is nearly always (if not always) a bad idea. Beyond that, I don’t really have any further opinion other than what I said in this response to a comment below.
  • While the claims made in the paper appear to be fundamentally sound, the presentation leaves something to be desired. It’s unclear to me why the authors relegated some of the most important technical points to the Discussion, or didn’t explictly state them at all. The abstract also seems to me to be overly sensational–though, in hindsight, not nearly as much as I initially suspected. And it also seems questionable to tar all of neuroscience with a single brush when the analyses reported only applied to a few specific domains (and we know for a fact that in, say, neuroimaging, this problem is almost nonexistent). I guess to be charitable, one could pick the same bone with a very large proportion of published work, and this kind of thing is hardly unique to this study. Then again, the fact that a practice is widespread surely isn’t sufficient to justify that practice–or else there would be little point in Aarts et al criticizing a practice that so many people clearly engage in routinely.
  • Given my last post, I can’t help pointing out that this is a nice example of how mandatory data sharing (or failing that, a culture of strong expectations of preemptive sharing) could have made evaluation of scientific claims far easier. If the authors had attached the data file coding the 315 studies they reviewed as a supplement, I (and others) would have been able to clarify the ambiguity I originally raised much more quickly. I did send a follow up email to Aarts to ask if she and her colleagues would consider putting the data online, but haven’t heard back yet.

strong opinions about data sharing mandates–mine included

Apparently, many scientists have rather strong feelings about data sharing mandates. In the wake of PLOS’s recent announcement–which says that, effective now, all papers published in PLOS journals must deposit their data in a publicly accessible location–a veritable gaggle of scientists have taken to their blogs to voice their outrage and/or support for the policy. The nays have posts like DrugMonkey’s complaint that the inmates are running the asylum at PLOS (more choice posts are here, here, here, and here); the yays have Edmund Hart telling the nays to get over themselves and share their data (more posts here, here, and here). While I’m a bit late to the party (mostly because I’ve been traveling and otherwise indisposed), I guess I’ll go ahead and throw my hat into the ring in support of data sharing mandates. For a number of reasons outlined below, I think time will show the anti-PLOS folks to very clearly be on the wrong side of this issue.

Mandatory public deposition is like, totally way better than a “share-upon-request” approach

You might think that proactive data deposition has little incremental utility over a philosophy of sharing one’s data upon request, since emails are these wordy little things that only take a few minutes of a data-seeker’s time to write. But it’s not just the time and effort that matter. It’s also the psychology and technology. Psychology, because if you don’t know the person on the other end, or if the data is potentially useful but not essential to you, or if you’re the agreeable sort who doesn’t like to bother other people, it’s very easy to just say, “nah, I’ll just go do something else”. Scientists are busy people. If a dataset is a click away, many people will be happy to download that dataset and play with it who wouldn’t feel comfortable emailing the author to ask for it. Technology, because data that isn’t publicly available is data that isn’t publicly indexed. It’s all well and good to say that if someone really wants a dataset, they can email you to ask for it, but if someone doesn’t know about your dataset in the first place–because it isn’t in the first three pages of Google results–they’re going to have a hard time asking.

People don’t actually share on request

Much of the criticism of the PLoS data sharing policy rests on the notion that the policy is unnecessary, because in practice most journals already mandate that authors must share their data upon request. One point that defenders of the PLOS mandate haven’t stressed enough is that such “soft” mandates are largely meaningless. Empirical studies have repeatedly demonstrated  that it’s actually very difficult  to get authors to share their data upon request —even when they’re obligated to do so by the contractual agreement they’ve signed with a publisher. And when researchers do fulfill data sharing requests, they often take inordinately long to do so, and the data often don’t line up properly with what was reported in the paper (as the PLOS editors noted in their explanation for introducing the policy), or reveal potentially serious errors.

Personally, I have to confess that I often haven’t fulfilled other researchers’ requests for my data–and in at least two cases, I never even responded to the request. These failures to share didn’t reflect my desire to hide anything; they occurred largely because I knew it would be a lot of work, and/or the data were no longer readily accessible to me, and/or I was too busy to take care of the request right when it came in. I think I’m sufficiently aware of my own character flaws to know that good intentions are no match for time pressure and divided attention–and that’s precisely why I’d rather submit my work to journals that force me to do the tedious curation work up front, when I have a strong incentive to do it, rather than later, when I don’t.

Comprehensive evaluation requires access to the data

It’s hard to escape the feeling that some of the push-back against the policy is actually rooted in the fear that other researchers will find mistakes in one’s work by going through one’s data. In some cases, this fear is made explicit. For example, DrugMonkey suggested that:

There will be efforts to say that the way lab X deals with their, e.g., fear conditioning trials, is not acceptable and they MUST do it the way lab Y does it. Keep in mind that this is never going to be single labs but rather clusters of lab methods traditions. So we’ll have PLoS inserting itself in the role of how experiments are to be conducted and interpreted!

This rather dire premonition prompted a commenter to ask if it’s possible that DM might ever be wrong about what his data means–necessitating other pairs of eyes and/or opinions. DM’s response was, in essence, “No.”. But clearly, this is wishful thinking: we have plenty of reasons to think that everyone in science–even the luminaries among us–make mistakes all the time. Science is hard. In the fields I’m most familiar with, I rarely read a paper that I don’t feel has some serious flaws–even though nearly all of these papers were written by people who have, in DM’s words, “been at this for a while”. By the same token, I’m certain that other people read each of my papers and feel exactly the same way. Of course, it’s not pleasant to confront our mistakes by putting everything out into the open, and I don’t doubt that one consequence of sharing data proactively is that error-finding will indeed become much more common. At least initially (i.e., until we develop an appreciation for the true rate of error in the average dataset, and become more tolerant of minor problems), this will probably cause everyone some discomfort. But temporary discomfort surely isn’t a good excuse to continue to support practices that clearly impede scientific progress.

Part of the problem, I suspect, is that scientists have collectively internalized as acceptable many practices that are on some level clearly not good for the community as a whole. To take just one example, it’s an open secret in biomedical science that so-called “representative figures” (of spiking neurons, Western blots, or whatever else you like) are rarely truly representative. Frequently, they’re actually among the best examples the authors of a paper were able to find. The communal wink-and-shake agreement to ignore this kind of problem is deeply problematic, in that it likely allows many claims to go unchallenged that are actually not strongly supported by the data. In a world where other researchers could easily go through my dataset and show that the “representative” raster plot I presented in Figure 2C was actually the best case rather than the norm, I would probably have to be more careful about making that kind of claim up front–and someone else might not waste a lot of their time chasing results that can’t possibly be as good as my figures make them look.

Figure 1.  A representative planet.

The Data are a part of the Methods

If you still don’t find this convincing, consider that one could easily have applied nearly all of the arguments people having been making in the blogosphere these past two weeks to that dastardly scientific timesink that is the common Methods sections. Imagine that we lived in a culture where scientists always reported their Results telegraphically–that is, with the brevity of a typical Nature or Science paper, but without the accompanying novel’s worth of Supplementary Methods. Then, when someone first suggested that it might perhaps be a good idea to introduce a separate section that describes in dry, technical language how authors actually produced all those exciting results, we would presumably see many people in the community saying something like the following:

Why should I bother to tell you in excruciating detail what software, reagents, and stimuli I used in my study? The vast majority of readers will never try to directly replicate my experiment, and those who do want to can just email me to get the information they need–which of course I’m always happy to provide in a timely and completely disinterested fashion. Asking me to proactively lay out every little methodological step I took is really unreasonable; it would take a very long time to write a clear “Methods” section of the kind you propose, and the benefits seem very dubious. I mean, the only thing that will happen if I adopt this new policy is that half of my competitors will start going through this new section with a fine-toothed comb in order to find problems, and the other half will now be able to scoop me by repeating the exact procedures I used before I have a chance to follow them up myself! And for what? Why do I need to tell everyone exactly what I did? I’m an expert with many years of experience in this field! I know what I’m doing, and I don’t appreciate your casting aspersions on my work and implying that my conclusions might not always be 100% sound!

As far as I can see, there isn’t any qualitative difference between reporting detailed Methods and providing comprehensive Data. In point of fact, many decisions about which methods one should use depend entirely on the nature of the data, so it’s often actually impossible to evaluate the methodological choices the authors made without seeing their data. If DrugMonkey et al think it’s crazy for one researcher to want access to another researcher’s data in order to determine whether the distribution of some variable looks normal, they should also think it’s crazy for researchers to have to report their reasoning for choosing a particular transformation in the first place. Or for using a particular reagent. Or animal strain. Or learning algorithm, or… you get the idea. But as Bjorn Brembs succinctly put it, in the digital age, this is silly: for all intents and purposes, there’s no longer any difference between text and data.

The data are funded by the taxpayers, and (in some sense) belong to the taxpayers

People vary widely in the extent to which they feel the public deserves to have access to the products of the work it funds. I don’t think I hold a particularly extreme position in this regard, in the sense that I don’t think the mere fact that someone’s effort is funded by the public automatically means any of their products should be publicly available for anyone’s perusal or use. However, when we’re talking about scientific data–where the explicit rationale for funding the work is to produce new generalizable knowledge, and where the marginal cost of replicating digital data is close to zero–I really don’t see any reason not to push very strongly to force scientists to share their data. I’m sympathetic to claims about scooping and credit assignment, but as a number of other folks have pointed out in comment threads, these are fundamentally arguments in favor of better credit assignment, and not arguments against sharing data. The fear some people have of being scooped is not sufficient justification for impeding our collective scientific progress.

It’s also worth noting that, in principle, PLOS’s new data sharing policy shouldn’t actually make it any easier for someone else to scoop you. Remember that under PLOS’s current data sharing mandate–as well as the equivalent policies at most other scientific journals–authors are already required to provide their data to anyone else upon request. Critics who argue that the new public archiving mandate opens the door to being scooped are in effect admitting that the old mandate to share upon request doesn’t work, because in theory there already shouldn’t really be anything preventing me from scooping you with your data simply by asking you for it (other than social norms–but then, the people who are actively out to usurp others’ ideas are the least likely to abide by those norms anyway). It’s striking to see how many of the posts defending the “share-upon-request” approach have no compunction in saying that they’re currently only willing to share their data after determining what the person on the other end wants to use it for–in clear violation of most journals’ existing policy.

It’s really not that hard

Organizing one’s data or code in a form minimally suitable for public consumption isn’t much fun. I do it fairly regularly; I know it sucks. It takes some time out of your day, and requires you to allocate resources to the problem that could otherwise be directed elsewhere. That said, a lot of the posts complaining about how much effort the new policy requires seem absurdly overwrought. There seems to be a widespread belief–which, as far as I can tell, isn’t supported by a careful reading of the actual PLOS policy–that there’s some incredibly strict standard that datasets have to live up to before pulic release. I don’t really understand where this concern comes from. Personally, I spend much of my time analyzing data other people have collected. I’ve worked with many other people’s data, and rarely is it in exactly the form I would like. Often times it’s not even in the ballpark of what I’d like. And I’ve had to invest a considerable amount of my time understanding what columns and rows mean, and scrounging for morsels of (poor) documentation. My working assumption when I do this–and, I think, most other people’s–is that the onus is on me to expend some effort figuring out what’s in a dataset I wish to use, and not on the author to release that dataset in a form that a completely naive person could understand without any effort. Of course it would be nice if everyone put their data up on the web in a form that maximized accessibility, but it certainly isn’t expected*. In asking authors to deposit their data publicly, PLOS isn’t asserting that there’s a specific format or standard that all data must meet; they’re just saying data must meet accepted norms. Since those norms depend on one’s field, it stands to reason that expectations will be lower for a 10-TB fMRI dataset than for an 800-row spreadsheet of behavioral data.

There are some valid concerns, but…

I don’t want to sound too Pollyannaish about all this. I’m not suggesting that the PLOS policy is perfect, or that issues won’t arise in the course of its implementation and enforcement. It’s very clear that there are some domains in which data sharing is a hassle, and I sympathize with the people who’ve pointed out that it’s not really clear what “all” the data means–is it the raw data, which aren’t likely to be very useful to anyone, or the post-processed data, which may be too close to the results reported in the paper? But such domain- or case-specific concerns are grossly outweighed by the very general observation that it’s often impossible to evaluate previous findings adequately, or to build a truly replicable science, if you don’t have access to other scientists’ data. There’s no doubt that edge cases will arise in the course of enforcing the new policy. But they’ll be dealt with on a case-by-case basis, exactly as the PLOS policy indicates. In the meantime, our default assumption should be that editors at PLOS–who are, after all, also working scientists–will behave reasonably, since they face many of the same considerations in their own research. When a researcher tells an editor that she doesn’t have anywhere to put the 50 TB of raw data for her imaging study, I expect that that editor will typically respond by saying, “fine, but surely you can drag and drop a directory full of the first- and second-level beta images, along with a basic description, into NeuroVault, right?”, and not “Whut!? No raw DICOM images, no publication!”

As for the people who worry that by sharing their data, they’ll be giving away a competitive advantage… to be honest, I think many of these folks are mistaken about the dire consequences that would ensue if they shared their data publicly. I suspect that many of the researchers in question would be pleasantly surprised at the benefits of data sharing (increased citation rates, new offers of collaboration, etc.) Still, it’s clear enough that some of the people who’ve done very well for themselves in the current scientific system–typically by leveraging some incredibly difficult-to-acquire dataset into a cottage industry of derivative studies–would indeed do much less well in a world where open data sharing was mandatory. What I fail to see, though, is why PLOS, or the scientific community as a whole, should care very much about this latter group’s concerns. As far as I can tell, PLOS’s new policy is a significant net positive for the scientific community as a whole, even if it hurts one segment of that community in the short term. For the moment, scientists who harbor proprietary attitudes towards their data can vote with their feet by submitting their papers somewhere other than PLOS. Contrary to the dire premonitions floating around, I very much doubt any potential drop in submissions is going to deliver a terminal blow to PLOS (and the upside is that the articles that do get published in PLOS will arguably be of higher quality). In the medium-to-long term, I suspect that cultural norms surrounding who gets credit for acquiring and sharing data vs. analyzing and reporting new findings based on those data are are going to undergo a sea change–to the point where in the not-too-distant future, the scoopophobia that currently drives many people to privately hoard their data is a complete non-factor. At that point, it’ll be seen as just plain common sense that if you want your scientific assertions to be taken seriously, you need to make the data used to support those assertions available for public scrutiny, re-analysis, and re-use.

 

* As a case in point, just yesterday I came across a publicly accessible dataset I really wanted to use, but that was in SPSS format. I don’t own a copy of SPSS, so I spent about an hour trying to get various third-party libraries to extract the data appropriately, without any luck. So eventually I sent the file to a colleague who was helpful enough to convert it. My first thought when I received the tab-delimited file in my mailbox this morning was not “ugh, I can’t believe they released the file in SPSS”, it was “how amazing is it that I can download this gigantic dataset acquired half the world away instantly, and with just one minor hiccup, be able to test a novel hypothesis in a high-powered way without needing to spend months of time collecting data?”