“Disappointed with your results? Boost your scientific paper”

This, sent in by Ben Bolker, is just tooooo funny. Click on the above image to see more clearly. In addition to the quote I used in the above title, there’s also this:

+10.000 correlations/min

Sooner than later, your future discovery will pop up.

and this:

The most relevant conclusions in your scientific paper are concealed under the experimental data but you simply cannot see them.

All they need is to pipe in a Mechanical Turk request form on one end and a Psychological Science submission form on the other, and they’ll have the complete package!

29 thoughts on ““Disappointed with your results? Boost your scientific paper”

  1. It looks like someone monetized Kanazawa’s approach to empirical research. I wonder if they can autogenerate a waffly evo psych explanation for each correlation for an extra fee?

  2. Oh no! I can only do 10 correlations per minute. I’ll be out of a job.

    also, what kind of slow system can only do 10,000 correlations per minute. Seems 3 or 4 orders of magnitude too slow. I require faster spurious results.

  3. So who’s got enough time on their hands to sign up for the free trial?

    I find the “seamless data consolidation” to be the most unbelievable hype on the ad. That’s pretty much an oxymoron.

  4. Laugh all you will but the tragedy is they only seem to be automating what already goes on. If we are going to fish, why not trawl?

    The reductio ad absurdum can be a wake up call.

    • Fernando, I think that doing this is appropriate to science, however it is incompatible with NHST. As you say, if your going to do something why not do it right? The problems arise in the interpretation of the result. The same people who click around in SPSS and/or keep running an experiment over and over until they get p<0.05 will clearly misuse this tool. I mentioned in a comment below that doing this myself led me just to realize how little information a "significant" correlation provides. In my experience it is not possible to convey this realization to others until they see the results of fishing at such a large scale for themselves on their own data.

  5. I’ve written programs to do something very similar. It basically plotted and calculated pearson’s R for everything that was measured, then created barplots of each relationship ranked by pearson’s R. In the end I saw so many correlations it became meaningless to find them. I think it is not useless but people using it blindly will be easily led astray.

    39 euros seems like a decent price for people who would have no idea how to do this on their own. It took me about a weekend. It would be more useful to use something like Eureqa (Or Nutonian/Formalize whatever they call it now) to look for the form of relationship between different aspects of the data then go check the literature (or think on it) to see if there is theoretical justification for the function. Even then you end up playing with the complexity penalties if you don’t like the results, but it is better than only fitting lines.

  6. Actually, JG Gardin was forced to publish in protest a similar system for archeology (called SNARK) back in the 1960/70s – many in academic circles did not see the problem and more than a few graduate students got Archeology Phd,s by running it.

    Wonder if there might be a way to find out how often AutDiscovery has been referenced as a method in a grant application and been funded (by I guess a European funding agency)?

    The “You keep control” originally made me skeptical it was a hoax, perhaps given NIH (UK) recent comment re: NIH to Researchers: Credibility Counts – “The truth is we don’t really know what the full scope of the problem is” and their announcement they are working on “not be fooled anymore”
    http://www.medpagetoday.com/PublicHealthPolicy/ClinicalTrials/43995?isalert=1&uun=g646841d638R5778811u&utm_source=breaking-news&utm_medium=email&utm_campaign=breaking-news&xid=NL_breakingnews_2014-01-27

    Notice the link to a paper by authors known to many of us – http://www.thelancet.com/journals/lancet/article/PIIS0140-6736%2813%2962227-8/abstract

    Notice the link to a enlightening paper by authors known to many of us – http://www.thelancet.com/journals/lancet/article/PIIS0140-6736%2813%2962227-8/abstract

      • When I first saw the site, I was about 30/70 on whether it was a parody or not. But once I noticed that all the examples came from Spanish universities, I decided it must be real. If it was a parody, I think they’d either use super-famous universities (Harvard, etc.) or else go the other way and use funny fakes (Miskatonic U., etc.) or notorious examples (Diederik Stapel, Dr. Anil Potti, etc.). Real but obscure places, that would seem to imply it’s not a fake.

  7. Ooops! I hope I am not too late to the party!

    My name is Ray G. Butler and I am the one responsible for the development of AutoDiscovery. Thus way, I guess I am also the one that ultimately has given the kick off to this discussion.

    If I may, I would appreciate you giving me opportunity to explain my views:

    – AutoDiscovery is no statistical software, and DOES NOT pretend to, by the way.

    Products such as R, SPSS or similar do count with a large number of functionalities which, at the end of the day, provide the user with numeric results of high precision. Our aim has never been to compete against those software packages, which are already extremely good at that.

    AutoDiscovery does simply use a most common statistical tool (correlation) as a mean to an end, not as end in itself. The ultimate purpose is to help understand a little bit better the complex relationships there may be amongst variables in a given experiment.

    No more. No less.

    This way of “discovering stuff” rather than “data mining” may be useful at different stages of a given experiment, as it would at the beginning by helping the user designing the experiment, or at the end of it by helping the user reaching to more sound and meaning conclusions.

    This is not a rhetoric purpose or a bold statement, but a fact (autodiscovery.butlerscientifics.com/realsuccesscases.html).

    – AutoDiscovery does not intend to replace scientists.

    A “discovery” process is a very complex task itself, which requires a great deal of intellectual effort. No software could replace that human capacity.

    Needless to say, neither could AutoDiscovery.

    – AutoDiscovery has been built with the support of scientists, statisticians and software engineers, with the aim in mind at reducing the time and effort that interpreting multiple sources data files requires.

    There are many ways to do that and AutoDiscovery does only and humbly intend to be one more approach.

    We are also well aware that science does not circumscribe itself to correlations, that correlations do not imply causation, that analyzing thousands of correlations may turn out in spurious results and that thousands of spurious results is worse than any complex data file. We are well aware of all these facts, while we are also aware there may be another thousand things we may be overlooking, too.

    Because of that, we have always tried to minimize the impact those inconveniences imply by adding some functionalities to AutoDiscovery, such as:

    • Filtering by means of higher than regular significance thresholds.
    • Prioritizing detected correlations according to its exclusivity degree.
    • A visual and easy way to explore results.

    – But we intend to improve. Always.

    And your comments, far from being ignored, mean also a source and an opportunity to improve.

    To end with, if at any time we sounded offensive or arrogant with language subtleties we are not skilled at (English is not our mother tongue), we very sincerely apologize. We simply try to keep up with the challenge: there must be other ways to help speed up the research process in order to discover more, faster and better.

    For those who would be interested in learning what we have to say through AutoDiscovery, we would be glad and willing to demonstrate the capabilities of the software, and do it through a real case.

    For the rest, we can only wish you have fun with the dilapidation! ;)))

    We would really appreciate an opportunity to talk with you at your convenience. Please find below links to our contact channels:

    – Contact form (autodiscovery.butlerscientifics.com/wesupportyou.html)
    – Live chat (autodiscovery.butlerscientifics.com/wesupportyou.html)
    – Personalized online demonstrations (autodiscovery.butlerscientifics.com/bookademo.html)
    – Our e-mail ([email protected])
    – Our twitter (@butlersci)
    – My ResearchGate profile (https://www.researchgate.net/profile/Ray_G_Butler).
    – My LinkedIn profile (http://es.linkedin.com/in/raygbutler).

    Thank you,

    Ray

  8. Pingback: Friday links: 7-year postdoc, every major’s terrible, citation Matthew Effect, KILL BLOGS WITH FIRE, and more | Dynamic Ecology

  9. Pingback: Links 2/4/14 | Mike the Mad Biologist

Comments are closed.