Mary Kwasny is an associate professor in the department of preventive medicine and an active member of the Biostatistics Collaboration Center at Northwestern University, Feinberg School of Medicine. She has been enjoying the art of statistical consulting and collaboration for more than 20 years in academic medical centers and external non-profits.
As relatively new members of the statistics profession, you might be considering a career as a statistical consultant. You may also be wondering what it is exactly that statistical consultants do. (As a more established statistician, I hear that question a lot. I also want to reply that we don’t do anything “exactly”; we allow for error, unlike mathematicians.)
Well, as statisticians, we know how to deal with data—collect it, clean it, analyze it, and interpret the findings. We are not, however, experts in the subject matter in which the data was collected. But, the subject matter experts, or clients, who collect data are—guess what!—not experts in statistics. Working in a particular industry or research field, or on a specific project, the statistical consultant needs to not only understand, but be able to translate, the subject matter into a statistical problem and then translate findings back to the client.
Anyone who has read a translation of a translation, or played Telestrations or Telephone, knows this is not always easy. As statistical consultants, we have to be able to understand how the data are generated and what measures are being used. We have to translate research questions into problems that can be analyzed with data, determine the appropriate type of statistical analysis, and sometimes be the “honest broker” when interpreting the results to avoid bias.
Apart from knowing statistics and learning about the research field you work in, you will also need communication skills to be able to talk to clients, understand their language, and communicate statistics effectively. You will need business skills, business acumen, time-management skills, networking talents, and a host of other skills. Putting yourself out there as a new professional is hard, isn’t it?
Consulting Comes in Many Colors
There are different types of consulting and different types of consulting jobs available. On one end of the spectrum, there are short-term consultations, which may just be one or two meetings. These typically are for quantitatively skilled individuals who are comfortable running their own analyses, but appreciate an expert looking over their work. On the other end of the spectrum, there are long-term consultations; these are occasionally called collaborations in an academic setting. These are projects in which you are an integral part of the research team from start to finish, and they may span many years. And there are projects that could be anywhere in between. Consulting may include being an expert witness in a trial or helping someone with a master’s thesis in physical therapy.
As for settings, you could work on your own as an independent contractor or in a small group. You could also work in a large group of consultants, in a consulting firm, or be the “in-house” statistician in a large company. I personally enjoy working in an academic environment; I am a member of a collaboration center in which part of my job is consulting. Consulting may not be part of your job description at all, but you may want to do some consulting “on the side” if your current employer allows it.
Consulting Takes You to the Unknown
Almost every consulting project is a giant step into the unknown. You meet with a client, discuss their research questions who knows how many times, learn more about their problem, and then you do the statistics part—possibly including, but not limited to, providing assistance for study design, data collection, proposing an analysis plan, conducting an analysis, and writing up the results. But it doesn’t stop here. You need to estimate the time and cost for your work and later invoice for your work—and all this is for the straightforward projects.
It’s definitely not like coursework, in which you can expect to fit longitudinal models for your Longitudinal Analysis class or survival models for your Survival Analysis class. Any consulting project has the potential to be … anything!
My colleague and friend Masha Kocherginsky said that in more than 15 years of consulting, she has never encountered a real-world consulting project that’s straight out of a textbook, and it’s never the same analysis twice. As a result, it’s never boring!
So Why Would You Want to Do This?
For me, it was precisely because of all the above. John Tukey once said, “The best thing about being a statistician is that you get to play in everyone’s backyard.” I love learning and problem solving. I love seeing what other people are excited about. I love helping them learn more about their data, which, in turn, helps them learn more about their subject matter. Topics change, people are different, and the one thing you know will happen is you will learn something and help someone. This alone can be addictive.
There are times I call consulting a mix between statistical triage and statistical improvisation. In triage, the key is to quickly assess, diagnose, and assist a patient. Any actor will tell you their craft is honed when they are forced to improvise. Not that you are “making things up” as you go, but rather you are attempting to give unscripted advice to an unknown question.
Short consults can be difficult this way, but not everyone needs a full-time statistician. Other projects provide great opportunities to learn new topic domains, as well as to delve into potentially different statistical methods or skills. For these projects, I use the analogy that statistics are the scaffolding that helps construct or renovate the building. Statistical consulting provides a means to keep statistical and communication skills sharp. Additionally, like teaching an introductory statistics course, it provides the opportunity to discover different and potentially better ways to explain things. All told, statistical consulting is educational, exciting, and challenging.
When I was a graduate student, I did a summer internship and had my first “outside the classroom” statistical consult. A pediatric surgical fellow wanted to look at the importance of staging laparotomy in pediatric Hodgkin’s disease. As an intern, I had the good fortune to work on this consult under the supervision of a faculty member, who was there to facilitate the consult. It was the first time I used categorization and regression trees (in my defense, I don’t think random forests had been developed at that time). I learned something about medical procedures and treatments. The investigator was really nice and brought doughnuts to our meetings. And I got my first collaborative publication out of it. It was amazing. I was hooked. I never worked with that investigator again, but recently looked him up only to discover it was his first publication as well. He is now an endowed chair in surgical research at a prominent children’s hospital. Apparently, he was hooked, too.
Consulting Perks and Struggles
Not all projects have gone as smoothly or been as productive—and I quickly realized doughnuts were not the norm. There have been problems with investigators wanting to run inappropriate methods, times when I thought I understood a data set and was—in fact—wrong, and times when I sent an invoice only to have the recipient email back, “FOR WHAT!!!” I would hazard a guess that these problems are common and that every statistical consultant, anywhere in the world, has dealt with one or more of the same struggles. But, over the years, I have learned to be a better collaborator and to make it a priority to align expectations so those problems don’t occur quite as much. Granted, bad consulting experiences will happen, but, fortunately, there are many more positive experiences and enough enjoyment in the “random consult” to continue including collaboration in my job description.
Another perk of consulting is that, if you are interested in methodologic research, it is a hotbed for motivating examples and finding omissions and gaps in knowledge. Clients may seek statistical help after data collection, and it may be clear that traditional statistical methods are not appropriate. Other times, when investigators are collecting data, they change protocol out of necessity. These changes may not have gotten a statistical stamp of approval beforehand. For example, I once worked with an investigator who failed to tell me they changed urine collection from a 24-hour collection to a 12-hour collection until the end of the trial. While pragmatic and understandable, had they consulted me on the change, I would have requested they had a period of time where both were collected so we could properly adjust numbers taken under the different conditions.
In many of these situations, we can make assumptions, appropriately and clearly state those assumptions and potential limitations, and still provide a timely analysis. However, there may be more statistically efficient or better methods that could be developed over time to address these adaptations. The bonus with these motivational examples is that, provided you have the investigator’s consent, you automatically have data to use as an example!
Apart from the statistical, educational, and communication opportunities consulting provides, it is also a convenient way to expand your professional network. Being newer to the workforce, this may not seem like an important reason to become a consultant, but thanks to different collaborators over the years, I have developed a network of experts in various fields. These individuals have referred other investigators in need of statistical assistance to me and provide me with feedback if I’m looking for second opinions or clarifications when working on another project—or since I work in a medical school, when looking for a trusted medical specialist. Additionally, you will learn that not every statistician gets along with every potential collaborator. When that does happen, it is nice to know there are investigators out there who appreciate your unique talents and style.
Consulting for New Statisticians
As new statisticians, I would strongly recommend some consulting, be it on the job or as pro bono work. It is a great way to keep learning and honing your statistical craft. I would also suggest you don’t go it alone; find a mentor, colleague, or boss who is available for advice. This is especially helpful when you have one of those difficult clients.
Independent consulting can be a fabulous career choice, but first it is important to have a network of potential clients and a portfolio of work. As you develop as a consultant, you will be able to better estimate how long a project might take, more easily figure out the real question the investigator is asking, and develop an expanding toolbox of statistical methods to be used for other projects and new problems. While you learn and hone these skills, it is vital to have backup and advice.
The Statistical Consulting Section is active in the online ASA Community and caters to all types of consulting—within universities, pro-bono, on the side, sole proprietorships, or wherever and however consulting is done. This is a tremendous resource for those who are thinking of, or actively consulting in, any of those environments. So, go! Consult! Have fun!