Jan 15, 2014

The APA and Open Data: one step forward, two steps back?


Photo of Denny

I was pleasantly surprised when, last year, I was approached with the request to become Consulting Editor for a new APA journal called Archives of Scientific Psychology. The journal, as advertised on its website upon launch, had a distinct Open Science signature. As its motto said, it was an “Open Methodology, Open Data, Open Access journal”. That’s a lot of openness indeed.

When the journal started, the website not only boosted the Open Access feature of the journal, but went on to say that "[t]he authors have made available for use by others the data that underlie the analyses presented in the paper". This was an incredibly daunting move by APA - or so it seemed. Of course, I happily accepted the position.

After a few months, the first papers in Archives were published. Open Data enthusiast Jelte Wicherts of Tilburg University immediately tried to retrieve data for reanalysis. Then it turned out that the APA holds a quite ideosyncratic definition of the word “open”: upon his request, Wicherts was referred to a website that presented a daunting list of requirements for data-requests to fulfill. That was quite a bit more intimidating than the positive tone struck in the editorial that accompanied the launch of the journal.

This didn’t seem open to me at all. So: I approached the editors and said that I could not subscribe to this procedure, given the fact that the journal is supposed to have open data. The editors then informed me that their choice to implement these procedures was an entirely conscious one, and that they stood by it. Their point of view is articulated in their data sharing guidelines. For instance, "next-users of data must formally agree to offer co-authorship to the generator(s) of the data on any subsequent publications" since "[i]t is the opinion of the Archives editors that designing and conducting the original data collection is a scientific contribution that cannot be exhausted after one use of the data; it resides in the data permanently."

Well, that's not my opinion at all. In fact it's quite directly opposed to virtually everything I think is important about openness in scientific research. So I chose to resign my position.

In October 2013, I learned that Wicherts had taken the initiative of exposing the Archives’ policy in an open letter to the editorial board, in which he says:

“[…] I recently learned that data from empirical articles published in the Archives are not even close to being “open”.

In fact, a request for data published in the Archives involves not only a full-blown review committee but also the filling in and signing of an extensive form: http://www.apa.org/pubs/journals/features/arc-data-access-request-form.pdf

This 15-page form asks for the sending of professional resumes, descriptions of the policies concerning academic integrity at one’s institution, explicit research plans including hypotheses and societal relevance, specification of the types of analyses, full ethics approval of the reanalysis by the IRB, descriptions of the background of the research environment, an indication of the primary source of revenue of one’s institution, dissemination plans of the work to be done with the data, a justification for the data request, manners of storage, types of computers and storage media being used, ways of transmitting data between research team members, whether data will be encrypted, and signatures of institutional heads.

The requester of the data also has to sign that (s)he provides an “Offer [of] co-authorship to the data generators on any subsequent publications” and the (s)he will offer to the review committee an “annual data use report that outlines what has been done, that the investigator remains in compliance with the original research proposal, and provide references of any resulting publications.”

In case of non-compliance of any of these stipulations, the requester can face up to a $10,000 fine as well a future prohibition of data access from work published in the Archives.”

A fine? Seriously? Kafkaesque!

Wicherts also notes that “the guidelines with respect to data sharing in the Archives considerably exceed APA’s Ethical Standard 8.14”. Ethical Standard 8.14 is a default that applies to all APA journals, and says:

“After research results are published, psychologists do not withhold the data on which their conclusions are based from other competent professionals who seek to verify the substantive claims through reanalysis and who intend to use such data only for that purpose, provided that the confidentiality of the participants can be protected and unless legal rights concerning proprietary data preclude their release.”

Since this guideline says nothing about fines and co-authorship requirements, we indeed have to conclude that it’s harder to get data from APA’s open science journal, than it is to get data from its regular journals. Picture that!

In response to my resignation and Wicherts' letter, the editors have taken an interesting course of action. Rather than change their policy such that their deeds match their name, they have changed their name to match their deeds. The journal is now no longer an "Open Methodology, Open Data, Open Access Journal" but an "Open Methodology, Collaborative Data Sharing, Open Access Journal".

The APA and open data. One step forward, two steps back.

Nov 20, 2013

Theoretical Amnesia


Photo of Denny

In the past few months, the Center for Open Science and its associated enterprises have gathered enormous support in the community of psychological scientists. While these developments are happy ones, in my view, they also cast a shadow over the field of psychology: clearly, many people think that the activities of the Center for Open Science, like organizing massive replication work and promoting preregistration, are necessary. That, in turn, implies that something in the current scientific order is seriously broken. I think that, apart from working towards improvements, it is useful to investigate what that something is. In this post, I want to point towards a factor that I think has received too little attention in the public debate; namely, the near absence of unambiguously formalized scientific theory in psychology.

Scientific theories are perhaps the most bizarre entities that the scientific imagination has produced. They have incredible properties that, if we weren’t so familiar with them, would do pretty well in a Harry Potter novel. For instance, scientific theories allow you to work out, on a piece of paper, what would happen to stuff in conditions that aren’t actually realized. So you can figure out whether an imaginary bridge will stand or collapse in imaginary conditions. You can do this by simply just feeding some imaginary quantities that your imaginary bridge would have (like its mass and dimensions) to a scientific theory (say, Newton’s) and out comes a prediction on what will happen. In the more impressive cases, the predictions are so good that you can actually design the entire bridge on paper, then build it according to specifications (by systematically mapping empirical objects to theoretical terms), and then the bridge will do precisely what the theory says it should do. No surprises.

That’s how they put a man on the moon and that’s how they make the computer screen you’re now looking at. It’s all done in theory before it’s done for real, and that’s what makes it possible to construct complicated but functional pieces of equipment. This is, in effect, why scientific theory makes technology possible, and therefore this is an absolutely central ingredient of the scientific enterprise which, without technology, would be much less impressive than it is.

It’s useful to take stock here, and marvel. A good scientific theory allows you infer what would happen to things in certain situations without creating the situations. Thus, scientific theories are crystal balls that actually work. For this reason, some philosophers of science have suggested that scientific theories should be interpreted as inference tickets. Once you’ve got the ticket, you get to sidestep all the tedious empirical work. Which is great, because empirical work is, well, tedious. Scientific theories are thus exquisitely suited to the needs of lazy people.

My field – psychology – unfortunately does not afford much of a lazy life. We don’t have theories that can offer predictions sufficiently precise to intervene in the world with appreciable certainty. That’s why there exists no such thing as a psychological engineer. And that’s why there are fields of theoretical physics, theoretical biology, and even theoretical economics, while there is no parallel field of theoretical psychology. It is a sad but, in my view, inescapable conclusion: we don’t have much in the way of scientific theory in psychology. For this reason, we have very few inference tickets – let alone inference tickets that work.

And that’s why psychology is so hyper-ultra-mega empirical. We never know how our interventions will pan out, because we have no theory that says how they will pan out (incidentally, that’s also why we need preregistration: in psychology, predictions are made by individual researchers rather than by standing theory, and you can’t trust people the way you can trust theory). The upshot is that, if we want to know what would happen if we did X, we have to actually do X. Because we don’t have inference tickets, we never get to take the shortcut. We always have to wade through the empirical morass. Always.

This has important consequences. For instance, as a field has less theory, it has to leave more to the data. Since you can’t learn anything from data without the armature of statistical analysis, a field without theory tends to grow a thriving statistical community. Thus, the role of statistics grows as soon as the presence of scientific theory wanes. In extreme cases, when statistics has entirely taken over, fields of inquiry can actually develop a kind of philosophical disorder: theoretical amnesia. In fields with this disorder, researchers no longer know what a theory is, which means that they can neither recognize its presence nor its absence. In such fields, for instance, a statistical model – like a factor model – can come to occupy the vacuum created by the absence of theory. I am often afraid that this is precisely what has happened with the advent of “theories” like those of general intelligence (a single factor model) and the so-called “Big Five” of personality (a five-factor model). In fact, I am afraid that this happened in many fields in psychology, where statistical models (which, in their barest null-hypothesis testing form, are misleadingly called “effects”) rule the day.

If your science thrives on experiment and statistics, but lacks the power of theory, you get peculiar problems. Most importantly, you get slow. To see why, it’s interesting to wonder how psychologists would build a bridge, if they were to use their typical methodological strategies. Probably, they would build a thousand bridges, record whether they stand or fall, and then fit a regression equation to figure out which properties are predictive of the outcome. Predictors would be chosen on the basis of statistical significance, which would introduce a multiple testing problem. In response, some of the regressors might be clustered through factor analysis, to handle the overload of predictive variables. Such analyses would probably indicate lots of structure in the data, and psychologists would likely find that the bridges’ weight, size, and elasticity loads on a single latent “strength factor”, producing the “theory” that bridges higher on the “strength factor” are less likely to fall down. Cross validation of the model would be attempted by reproducing the analysis in a new sample of a thousand bridges, to weed out chance findings. It’s likely that, after many years of empirical research, and under a great number of “context-dependent” conditions that would be poorly understood, psychologists would be able to predict a modest but significant proportion of the variance in the outcome variable. Without a doubt, it would ta ke a thousand years to establish empirically what Newton grasped in a split second, as he wrote down his F=m*a.

Because increased reliance on empirical data makes you so incredibly slow, it also makes you susceptible to fads and frauds. A good theory can be tested in an infinity of ways, many of which are directly available to the interested reader (this is what give classroom demonstrations such enormous evidential force). But if your science is entirely built on generalizations derived from specifics of tediously gathered experimental data, you can’t really test these generalizations without tediously gathering the same, or highly similar, experimental data. That’s not something that people typically like to do, and it’s certainly not what journals want to print. As a result, a field can become dominated by poorly tested generalizations. When that happens, you’re in very big trouble. The reason is that your scientific field becomes susceptible to the equivalent of what evolutionary theorists call free riders: people who capitalize on the invested honest work of others by consistently taking the moral shortcut. Free riders can come to rule a scientific field if two conditions are satisfied: (a) fame is bestowed on whoever dares to make the most adventurous claims (rather than the most defensible ones), and (b) it takes longer to falsify a bogus claim than it takes to become famous. If these conditions are satisfied, you can build your scientific career on a fad and get away with it. By the time they find out your work really doesn’t survive detailed scrutiny, you’re sitting warmly by the fire in the library of your National Academy of Sciences1.

Much of our standard methodological teachings in psychology rest on the implicit assumption that scientific fields are similar if not identical in their methodological setup. That simply isn’t true. Without theory, the scientific ball game has to be played by different rules. I think that these new rules are now being invented: without good theory, you need fast acting replication teams, you need a reproducibility project, and you need preregistered hypotheses. Thus, the current period of crisis may lead to extremely important methodological innovations – especially those that are crucial in fields that are low on theory.

Nevertheless, it would be extremely healthy if psychologists received more education in fields which do have some theories, even if they are empirically shaky ones, like you often see in economics or biology. In itself, it’s no shame that we have so little theory: psychology probably has the hardest subject matter ever studied, and to change that may very well take a scientific event of the order of Newton’s discoveries. I don’t know how to do it and I don’t think anybody else knows either. But what we can do is keep in contact with other fields, and at least try to remember what theory is and what it’s good for, so that we don’t fall into theoretical amnesia. As they say, it’s the unknown unknowns that hurt you most.

1 Caveat: I am not saying that people do this on purpose. I believe that free riders are typically unaware of the fact that they are free riders – people are very good at labeling their own actions positively, especially if the rest of the world says that they are brilliant. So, if you think this post isn’t about you, that could be entirely wrong. In fact, I cannot even be sure that this post isn’t about me.

Oct 2, 2013

Smoking on an Airplane


Photo of Denny

People used to smoke on airplanes. It's hard to imagine, but it's true. In less than twenty years, smoking on airplanes has grown so unacceptable that it has become difficult to see how people ever condoned it in the first place. Psychological scientists used to refuse to share their data. It's not so hard to imagine, and it's still partly true. However, my guess is that a few years from now, data-secrecy will be as unimaginable as smoking on an airplane is today. We've already come a long way. When in 2005 Jelte Wicherts, Dylan Molenaar, Judith Kats, and I asked 141 psychological scientists to send us their raw data to verify their analyses, many of them told us to get lost - even though, at the time of publishing the research, they had signed an agreement to share their data upon request. "I don't have time for this," one famous psychologist said bluntly, as if living up to a written agreement is a hobby rather than a moral responsibility. Many psychologists responded in the same way. If they responded at all, that is.

Like Diederik Stapel.

I remember that Judith Kats, the student in our group who prepared the emails asking researchers to make data available, stood in my office. She explained to me how researchers had responded to our emails. Although many researchers had refused to share data, some of our Dutch colleagues had done so in an extremely colloquial, if not downright condescending way. Judith asked me how she should respond. Should she once again inform our colleagues that they had signed an APA agreement, and that they were in violation of a moral code?

I said no.

It's one of the very few things in my scientific career that I regret. Had we pushed our colleagues to the limit, perhaps we would have been able to identify Stapel's criminal practices years earlier. As his autobiography shows, Stapel counterfeited his data in an unbelievably clumsy way, and I am convinced that we would have easily identified his data as fake. I had many reasons for saying no, which seemed legitimate at the time, but in hindsight I think my behavior was a sign of adaptation to a defective research culture. I had simply grown accustomed to the fact that, when I entered an elevator, conversations regarding statistical analyses would fall silent. I took it as a fact of life that, after we methodologists had explained students how to analyze data in a responsible way, some of our colleagues would take it upon themselves to show students how scientific data analysis really worked (today, these practices are known as p-hacking). We all lived in a scientific version of The Matrix, in which the reality of research was hidden from all - except those who had the doubtful honor of being initiated. There was the science that people reported and there was the science that people did.

In Groningen University, where Stapel used to work, he was known as The Lord of the Data, because he never let anyone near his SPSS files. He pulled results out of thin air, throwing them around as presents to his co-workers, and when anybody asked him to show the underlying data files, he simply didn't respond. Very few people saw this as problematic, because, hey, these were his data, and why should Stapel share his data with outsiders?

That was the moral order of scientific psychology. Data are private property. Nosy colleagues asking for data? Just chase them away, like you chase coyotes from your farm. That is why researchers had no problem whatsoever denying access to their data, and that is why several people saw the data-sharing request itself as unethical. "Why don't you trust us?," I recall one researcher saying in a suspicious tone of voice.

It is unbelievable how quickly things have changed.

In the wake of the Stapel case, the community of psychological scientists committed to openness, data-sharing, and methodological transparency quickly reached a critical mass. The Open Science Framework allows researchers to archive all of their research materials, including stimuli, analysis code, and data, to make them public by simply pressing a button. The new Journal of Open Psychology Data offers an outlet specifically designed to publish datasets, thereby giving these the status of a publication. PsychDisclosure.org asks researchers to document decisions regarding, e.g., sample size determination and variable selection, that were left unmentioned in publications; most researchers provide this information without hesitation - some actually do so voluntarily. The journal Psychological Science will likely implement requirements for this type information in the submission process. Data-archiving possibilities are growing like crazy. Major funding institutes require data-archiving or are preparing regulations that do. In the Reproducibility Project, hundreds of studies are being replicated in a concerted effort. As a major symbol of these developments, we now have the Center for Open Science, which facilitates the massive grassroots effort to open up the scientific regime in psychology.

If you had told me that any of this would happen back in 2005, I would have laughed you away, just as I would have laughed you away in 1990, had you told me that the future would involve such bizarre elements as smoke-free airplanes.

The moral order of research in psychology has changed. It has changed for the better, and I hope it has changed for good.