Working in Interdisciplinary Teams
Many scientists are drawn to statistics and biostatistics by their passion to make a difference using their skills in mathematics to tackle applied problems. We discuss a variety of aspects of working in interdisciplinary teams, including expectations and roles of statisticians and collaborators, authorship and publication, funding expectations and alternatives, and combining methodology with collaboration. This article summarizes a panel discussion on interdisciplinary teams that took place at the inaugural Women in Statistics Conference held in Cary, North Carolina, May 15–17, 2014, in which the panelists were asked to address several topics relevant to working in interdisciplinary teams.
Expectations and Roles of Statisticians and Collaborators
Alyson Wilson: I work as a professor in a university statistics department. My research interests include engineering statistics and the application of statistics to solve problems in defense and security. I do very little “consulting”—at least if we define consulting as a short-term project in which the statistician meets with the client, performs analysis, and then meets again with a report. However, almost my whole research portfolio is comprised of collaborations, where I have an ongoing scientific relationship with other researchers. In the collaboration model, each researcher brings distinct skills to the project, and then we work together to develop a solution.
One feature of working collaboratively is that your research partners bring different levels of scientific, statistical, and data analytic sophistication to the relationship. I have worked across the spectrum. Some of my collaborators have had almost no statistical knowledge; some of my collaborations have been with scientists and engineers who have sophisticated measurements and experimental designs, but who need help developing methodology to analyze their data. A key feature of each of these collaborations is that we literally have to learn to talk to each other. We have different (or no!) language to talk with each other about the questions of interest, data collection, experimental design, uncertainty, and analytic techniques. Successful collaborations involve continually working to improve communication between the researchers.
Sandra Stinnett: When statisticians and physicians work together for long periods, work habits and roles become familiar. However, when working with a new investigator, it is helpful to be able to answer their questions. Many do not understand that we think of data in rows (observations) and columns (variables), so they are not aware of how to present data to us. Further, they sometimes have a general sense of what they want to analyze, but have difficulty articulating it.
Statisticians need to know the “story” of the project before they can begin to formulate an analysis plan. To help the collaborators relay the most critical aspects of the project, I ask them to complete a “Request for Analysis Form.” This often does not take the place of an in-person meeting, but gives enough information to get started. It also helps the client think through the types of questions a statistician needs to know to be able to help them. I obtain their contact information, the title and purpose of the study (such as an article or presentation), and the deadline for completion of the analysis. I ask for the study design (clinical trial, observational study), primary questions, and any other general information they can provide. Sometimes, these items have to be explained further when meeting with the client. During the meeting, often previously unknown contingencies come to light. These can change the course of the analysis, but the analysis request at least helps to start the conversation with the key points at the forefront.
Joanne Wendelberger: Establishing roles is critical to the success of interdisciplinary collaborations, and often the statistician holds the knowledge and experience to make this happen. When a new effort starts, the collaborators may not know what their true statistical needs are. In the past, advice was often sought for determining a sample size. In recent years, a frequent request has been for help with uncertainty quantification. The true need may actually be somewhat different from the stated need, and skillful discussion and probing may be required to establish goals, appropriate methods, and a path forward for a new problem.
Brenda Gaydos: Statisticians are employed in many areas of the pharmaceutical industry. Some areas of specialization include business analytics, product development, discovery, pre-clinical, toxicology, pharmacometrics, epidemiology, and clinical development. Statisticians are most often assigned to projects, and there is not an opportunity to agree on expectations with team members in advance. Collaboration and influence need to be earned along the way. For example, in clinical development, a statistician is assigned to a large cross-disciplinary team of physicians, scientists, and clinical trial operational experts. The statistician has a pre-defined set of deliverables they are responsible for. But to truly influence the direction of the project, strong communication and interpersonal skills are required. It is important to have high learning agility. To be successful, the statistician needs to be able to understand the science, the role of each team member, and the business of drug development.
Responsibility for Data Management
Alyson Wilson: Historically, much of the responsibility for data management within my collaborations has resided with me (or my graduate students). Statisticians frequently have more tools and expertise in data cleaning and data wrangling than our collaborators. As we enter the era of Big Data, this will likely have to change and our interdisciplinary teams will increasingly need to involve the partners who have specific expertise working with data and algorithms at scale.
Sandra Stinnett: Collaborators need to provide data in a form the statistician can use easily. To help them know what is expected, I provide them with a Word file called “Guidelines for Excel Data and Coding Document.” This document gives specifications for the format, variable names (such that they can be used in SAS), and a coding format for the variables. I ask that codes be laid out so it is easy to cut and paste them into a SAS program as formats. If the data are not in the desired form when I receive them, I ask for revisions. If the collaborator is aware of what a statistician needs to know to be able to perform a correct analysis and is able to provide data in a usable format, the in-person meeting between the two can be used for delving into more details. Since this information is something a statistician needs to relate to each client, it is time-saving and useful to have documents at the ready to address these common issues.
Brenda Gaydos: It is always important for statisticians to thoroughly understand how data were collected and how they were translated into the analysis data set. This is necessary to ensure the appropriate analyses are performed and conclusions drawn. The pharmaceutical industry has rigorous standards and requirements to ensure data quality. In clinical development, many companies have individuals who specialize in data management. There are also statistical roles (e.g., computational analysts) that focus on data set and data analysis creation and validation. This does not mean, however, the team statistician should delegate all data management responsibilities. They need to maintain a hands-on approach, working closely with the computational statistician and operations team to have a complete understanding of the data. Some activities include developing the data set and analysis programming requirements; developing the data unblinding plan; developing the data collection system; defining edits and reports to help with data-cleaning activities; and performing a final review to declare the data, data sets, and analysis code correct prior to unblinding.
Authorship and Publication
Alyson Wilson: I have always tried to maintain a policy that I will talk to anyone about a problem for one hour. Over time, a surprising number of collaborations and graduate student projects have grown from this policy and, at the very least, I have learned about much interesting science. But if you would like me to work on your problem for more than an hour, we need to discuss “payment.” This can take the form of academic payment (e.g., writing a paper together, writing a grant together, supporting me on your grant, developing a project for one of my graduate students) or actual financial compensation directly to me or through the department consulting center.
Sandra Stinnett: In a successful collaboration, statistical contributions are honored and authorship is given without hassle.
Funding Alternatives and Expectations
Joanne Wendelberger: Identification of funding and/or other resources is important in any setting. Funding may come from a mutual employer, through a consulting agreement, or from proposals and grants. Tangible and intangible costs and benefits need to be considered in determining whether to participate in a collaborative relationship and in deciding how much time and energy to invest. In industry and government, there will likely be requirements in place concerning allocation and usage of resources, while academic settings may involve both formal and informal agreements. The funding model can affect the nature of the relationship and the value the different parties place on the work.
Alyson Wilson: For academics engaging in interdisciplinary work, there is always a tension between spending your time participating as a co-investigator/key personnel on another scientist’s grant and developing your own grant as principal investigator (PI). Increasingly, there are opportunities to work as co-PIs on interdisciplinary awards (e.g., the Joint National Science Foundation Division of Mathematical Sciences/National Institutes of Health National Institute of General Medical Sciences Initiative to Support Research at the Interface of the Biological and Mathematical Sciences, which supports research in mathematics and statistics on substantive questions in the biological and biomedical sciences).
Combining Methodology with Collaboration
Joanne Wendelberger: Interdisciplinary work often can motivate the need for new methodology, creating opportunities for research and creative discovery. Often, approaches from different fields can cross-fertilize, resulting in new techniques or innovative ways to apply existing methods in new settings.
Brenda Gaydos: Homework is always necessary, regardless of the level of experience. The statistical literature is vast and growing and there is a wealth of knowledge in other disciplines. The pharmaceutical industry is rich with complex problems, and solutions are needed in a timely manner. The goal in industry is to solve the problem, not to develop new methods. Therefore, it is important to first identify that an adequate solution does not readily exist. Where it does not exist, developing extensions to work in other disciplines (e.g., economics, engineering, social sciences) and collaborating with others can be the most expedient approach to solving a complex problem. Collaborations across statisticians and scientists within the company as well as external to the company are fairly common and necessary. Many of the problems have wide-reaching impact. This is well recognized. To address this, there are forums where industry, regulatory, and academic statisticians come together with other disciplines to advance scientific methodology.
Discussion
Interaction with the audience on these topics at the Women in Statistics Conference stimulated discussion on a variety of topics. Differences of opinion about the way to approach a problem can occur, in some cases challenging the statistician to assert the importance of using appropriate methods. Authorship practices vary across fields, so clarification of expectations can be important both in the determination of authorship and in the communication of accomplishments for performance evaluations and promotions. In addition to the statistical expertise required to solve problems, development of leadership skills can be an important part of building a career, creating opportunities for leading teams, and having impact.
About the Authors
Brenda L. Gaydos earned her PhD in mathematical statistics from The Pennsylvania State University. She is a senior research fellow and science-driven adaptive program leader at Eli Lilly and Company, as well as an adjunct professor of biostatistics at the Indiana School of Medicine. She is a Fellow of the American Statistical Association and an elected member of Quantitative Sciences in the Pharmaceutical Industry who has given more than 50 invited lectures on statistical methods.
Sandra Stinnett is associate professor of biostatistics and bioinformatics at Duke University Medical Center. She holds a bachelor’s degree in psychology from the University of Houston, master’s in biometry from the University of Texas School of Public Health, and a DrPH in biostatistics from The University of North Carolina. She has served the American Statistical Association as chair of the Committee on Women and Section on Statistical Consulting and president of the Caucus for Women in Statistics.
Joanne Wendelberger is the group leader of the Statistical Sciences Group within the computer, computational, and statistical sciences division at Los Alamos National Laboratory. She holds a bachelor’s degree in mathematics from Oberlin College and master’s and PhD degrees in statistics from the University of Wisconsin. She is a Fellow of the American Statistical Association. She also has served as chair and program chair of the ASA Section on Physical and Engineering Sciences and has been a member of both the editorial board and management committee for Technometrics.
Alyson Wilson is an associate professor in the department of statistics at North Carolina State University. She holds an MS in statistics from Carnegie Mellon and a PhD in statistics from Duke. Wilson is a Fellow of the American Statistical Association. She has worked collaboratively at the National Institutes of Health with clinical scientists, at a small defense contractor with military personnel, at a national laboratory with engineers, and in Washington, DC, with policy analysts.
