So Why AREN’T Most Published Research Findings False? The Rebuttal

Welcome to “So Why ARE Most Published Research Findings False?”, a step by step walk through of the John Ioannidis paper “Why Most Published Research Findings Are False”. It probably makes more sense if you read this in order, so check out the intro here , Part 1  here ,Part 2  here,  Part 3 here, Part 4 here, and Part 5 here.

Okay people, we made it! All the way through one of the most cited research papers of all time, and we’ve all lost our faith in everything in the process. So what do we do now? Well, let’s turn the lens around on Ioannidis. What, if anything, did he miss and how do we digest this paper? I poked around for a few critiques of him, just to give a flavor. This is obviously not a comprehensive list, but it hits the major criticisms I could find.

The Title While quite a few people had no problem with the contents of Ioannidis’s paper, some took real umbrage with the title, essentially accusing it of being clickbait before clickbait had really gotten going. Additionally, since many people never read anything more than the title of a paper, a title that blunt is easily used as a mallet by anyone trying to disprove any study they chose. Interestingly, there’s apparently some question regarding whether or not Ioannidis actually wrote the title or if it was the editors at Plos Medicine, but the point stands. Given that misleading headlines and reporting are hugely blamed by many (including yours truly) for popular misunderstanding of science, that would be a fascinating irony.

Failing to reject the null hypothesis does not mean accepting the null hypothesis This is not so much a criticism of Ioannidis as it is of those who use his work to promote their own causes. There is a rather strange line of thought out there that seems to believe that life, or science, is a courtroom. Under this way of thinking, when you undermine a scientist and their hypothesis, your client is de facto not guilty. This in not true. If you somehow prove that chemotherapy is less effective than previously stated, that doesn’t actually mean that crystals cure cancer. You never prove the null hypothesis, you only fail to reject it.

The definition of bias contained more nuance In a paper written in response to the Ioannidis paper, some researchers from Johns Hopkins took umbrage with the presentation of “bias” in the paper. Their primary grouse seemed to be intent vs consequence. Ioannidis presents bias as a factor based on consequence, i.e. the way it skews the final results. They disliked this and believed bias should be based on intent, pointing out numerous ways in which things Ioannidis calls “bias” could creep in innocently. For example, if you are looking for a drug that reduces cardiac symptoms but you also find that mortality goes down for patients who take the medication, are you really not going to report that because it’s not what you were originally looking for? By the strictest definition this is “data dredging”, but is it really? Humans aren’t robots. They’re going to report interesting findings where they see them.

The effect of multiple teams This is one of the more interesting quibbles with the initial paper. Mathematically, Ioannidis proved that having multiple teams working on the same research question would increase the chances of a false result. In the same Hopkins paper, the researchers question the math behind the “multiple teams lead to more false positives” assertion. They mention that for any one study, the odds stay the same as they always have been. Ioannidis counters with an argument that boils down to “yes, if you assume those in competition don’t get more biased”.  Interestingly, later research has shown that this effect does exist and is much worse in fields where the R factor (pre-study odds) is low.

So overall, what would I say are the major criticisms or cautions around this paper that I personally will employ?

  1. If you’re citing science, use scientific terms precisely. Don’t get sloppy with verbage just to make your life easier.
  2. Remember, scientific best practices all feed off each other Getting a good sample size and promoting caution can reduce both overall bias and the effect of bias that does exist. The effect of multiple team testing can be partially negated by high pre-study odds. If a team or researcher employs most best practices but misses one, that may not be a death blow to their research. Look at the whole picture before dismissing the research.
  3. New is exciting, but not always reliable We all like new and quirky findings, but we need to let that go. New findings are the least likely to play out later, and that’s okay. We want to cast a broad net, but for real progress we need a longer attention span.
  4. Bias takes many forms When we mention “bias” we often jump right to financial motivations. But intellectual and social pressure can be bias, competing for tenure can cause bias, and confirming ones own findings can cause bias.
  5. There are more ways of being wrong than there are ways of being right Every researcher wants a true finding. They really do. No one wants their life’s work undone. While some researchers may be motivated by results they like, I do truly believe that the majority of problems are caused by the whole “needle in a haystack” thing more than the “convenient truth” thing.

Alright, that wraps us up! I enjoyed this series, and may do more going forward. If you see a paper that piques your interest, let me know and I’ll look in to it.  Happy holidays everyone!