Nancy L. Geller, National Heart, Lung, and Blood Institute
A wrench is a tool you pick up when you need it. Many investigators at medical centers regard statisticians as wrenches.
Why would anyone write this titillating sentence? An investigator group designs a study and has a hypothesis—an “interesting question”—they think they can answer. They have no idea what the sample size should be and little idea what success will be, but the question is really, really interesting. So, these investigators go and perform the experiment(s) and collect data. Then, they realize the usual t-test they have always employed might not work. Or maybe they try it and know the experiment worked, but the t-test doesn’t show it.
They’ve heard there’s a statistics group at their institution, so this is a great time to call on them. Plain and simple: they need a wrench, so they try to find one! (Actually, they may try to buy one.)
A statistician is assigned to meet with this investigational group and very politely questions what kind of difference the investigators think would be meaningful. The investigators give a tentative answer, and the statistician politely inquires about the variability of the responses (and gets another tentative answer). The statistician does a calculation on a little piece of paper and points out that the sample size was inadequate to detect the meaningful difference that was anticipated. For example, they experimented on six mice and they really needed 12 to detect the anticipated difference (at the 0.05 significance level) with adequate power (say 80%). The investigators go back to the lab, do another six mice and analyze the data (on all 12 mice) using a t-test. Success! They write the paper (not including the statistician as a coauthor) and submit it.
The review comes back with the reviewer complimenting them on their positive result and asking for justification of the sample size and their analysis methods.
Oh dear! They need the wrench (again). So, they contact the same statistician who undoubtedly gets more than a little angry that she has spent a couple of hours on an applied project and gotten no credit whatsoever. Not even an acknowledgment. (Who acknowledges a wrench??) There are three possible actions.
The statistician can take a deep breath, sigh, write a brief paragraph addressing that particular referee’s comment, and hand it over passively. Everything is peaceful and the next time these investigators need a wrench, they will inevitably call on this same statistician. This possibility does no good for the contributing statistician nor any good for the statistics profession, as these investigators will continue to regard statisticians as wrenches.
The second possible action is for the statistician to mention that the investigators couldn’t have succeeded with this experiment without her and her contribution is worth at least an acknowledgment. This might be polite and charming but very difficult to pull off in such a manner. It is quite possible at this point that the investigators discover what it feels like to be hit by a wrench. This, too, will not do much good for the statistician’s career nor the statistics profession, for certainly the investigators will go away and may never seek statistical help again. Of course they might request another wrench, saying the current one was simply incompetent. An improvement on this method is for the statistician to ask her supervisor to intervene on her behalf. That might work, but it might not, and might well depend on whether the supervisor herself is regarded as a wrench.
But there is a third possibility, and it needs to be initiated from the first contact. First, the statistician should set up a meeting with the investigators and make sure the head of the investigative team can attend. Then, she should ask for a description of the research; for example, ask for the protocol in a human or animal study, read it in advance, and then ask questions.
So, the first step is to become involved in the research and make sure you (the statistician) understand what the specific aims are and what the primary hypothesis is. Then, the statistician can appreciate why the study is being done.
If the criteria for “success” seem unrealistic, the statistician can ask questions such as, “Why do you think your intervention will be that good?” It is scientifically useful to reset expectations if they are unrealistic. As for variability, it’s legitimate to ask about where the estimate comes from. Overall, the statistician proceeding this way engages herself in the project.
At that time, it might be useful to suggest a modification of the experimental plan, for example, a protocol amendment if the biostatistician decides an increase in sample size is appropriate and the investigators agree. That might be a time when the biostatistician asks to be added as an investigator on the protocol. (If a biostatistician is going to do this, she should “have a feeling” the suggestion will be well received.) She also should suggest that she should undertake the data analysis (or at least oversee it) at the end of the study, because the t-test just isn’t the best approach.
Now the biostatistician is on the right path. She is no longer a consultant, but she is now a collaborator. She should be sure to mention that when the investigators are writing up the results, she will be happy to write a brief biostatistics section for the paper. And everyone will realize (since they are already on good terms) that the biostatistician should be one of the coauthors.
To some biostatisticians, this third path is a natural one, but to others it must be taught. One biostatistician remembers going to an American Society of Clinical Oncology meeting in her early days and standing at the poster she helped create with the principal investigator. Someone he knew asked who she was. He put his arm around her and said, “She’s my biostatistician. I never go anywhere without her!” It was, perhaps, her very first professional compliment.
He went on to be chair of a prominent medical center, and his career is lauded on the internet. On that site, his many medical collaborators are interviewed about his career, but she who wrote many papers with him was never contacted. However, she went on to become head of a biostatistics group and has come to realize that if she had wanted to be a real star, she would have gone to medical school!
Yet all she wants for her chosen profession is to be respected and appreciated by those “real stars”!
Leave a Reply