Statistics and Biostatistics Graduate Admissions Policy

Spring 2013
Return to Statistics Policies

Introduction

The reason for this new policy is to articulate how the UW-Madison Statistics graduate programs—MS and PhD with options in statistics and biostatistics—conduct graduate admissions. This is important for several reasons: 1) to be transparent about the process to ourselves and to others; 2) to address the ongoing needs for high quality students; 3) to clearly communicate to the Graduate School the basis of our decisions, as this has implications on the level of funding for recruitment; and 4) to address any issues related to inclusive excellence that might have legal or other implications.

Up front, it is important to recognize that the field of statistics draws extensively on both technical expertise, grounded in advanced mathematical skills, and communications talent, grounded in broad, liberal arts-style education. As such, there is not one metric for excellence. Further, our field is rapidly expanding in this “Age of Big Data” (NY times, 12 feb 2012), and the range of expertise is evolving quickly. Thus we have a compelling interest to attract students with diverse backgrounds who can excel at technical tasks and can communicate well with other researchers about appropriate design and analytics needs in our data-rich society. See in particular, “Building the Statistical Work Force: Increasing the Pipeline of U.S. Students” (Amstat News, Feb 2013, magazine.amstat.org/blog/ 2013/02/01/prescornerfeb2013). It is also worth noting that the National Association of Colleges and Employers (www.naceweb.org) values these same qualities of communication and quantitative skills.

The UW-Madison Graduate School verbally raised concerns in Fall 2012 about (1) low enrollment rate (proportion of students accepting offers of guaranteed support) and (2) a procedure that appears to divide candidates up based on country of origin. Below, we address these concerns by providing more information about our discipline needs and our criteria for selection. We believe our procedure has merit and is fair, achieving a balanced weight of mathematical and communication skills that reflect the needs of modern statistics and heighten the likelihood of future career success. We explain our procedure through a series of questions.

What are the minimum math courses required for an MS or PhD?

Applicants need a minimum of three semesters of calculus and a course in linear algebra (similar to UW's Math 340) to be considered. In addition, a course in real analysis/advanced calculus that covers calculus of several variables (similar to UW's Math 521) is highly recommended for both MS and PhD. A widely used textbook that covers Math 521 material is Rudin (Principles of Mathematical Analysis). [A lower level course such as UW's Math 421 is a good start for MS applicants, but Math 521 should be taken here to build technical skills.]
PhD students are expected to have had an introduction to measure and integration theory with proofs (similar to UW's Math 629). Books that cover Measure & Integration Proofs are Royden (Real Analysis) and Ash (Real Analysis and Probability).

This is clearly stated in our admissions FAQ. Some students may be admitted without all these prerequisites, should they have other dimensions in their background that are particularly attractive, such as experience in scientific teams. These are considered risks worth the taking to broaden our horizons.

What minimum level of communication skills are required?

We are ideally looking for communication skills that are commonly developed through a “liberal arts” education. From wikipedia: "liberal arts ... are those subjects or skills ... considered essential ... to know in order to take an active part in civic life.... The aim of these studies was to produce a virtuous, knowledgeable, and articulate person.... Mathematics, science, arts, and language can all be considered part of the liberal arts.... Analyzing and interpreting information is also studied." UW-Madison L&S majors requires 120 credits, with at least 80 outside of any one department. Math requires at least 15 credits of advanced math while in residence. The option has 7 courses (21 credits) numbered above 306 with some particulars. Thus, less than 20% is required to be in math, and 75% is outside of that department. These relative weight of focus versus breadth in liberal arts degrees is typical of US undergraduate training, and has been relatively uncommon outside of the US, although this appears to be changing with globalization.

English language skills are quite important in statistics. However, international applications to statistics degree programs across the US typically have strong mathematical skills, but often weak English language skills. Applicants must have at least an 18 on the TOEFL speak portion, and ideally a 20 or higher. We have only had one international student apply with a TOEFL speak of 26; this person later won a teaching fellowship. We pioneered the L&S summer ITA training program, and work closely with English instructors to ramp up English and other communication skills. Still, most of our international students struggle with communications, and must refine these skills within our program (see next question). Those that complete our graduate programs, particularly those going into industry, speak highly of the communications skills they gained while here.

How do we evaluate excellence and rank-order graduate candidates?

Due to the wide range of skills required in statistics, we must largely conduct a holistic review. As indicated above, there is a floor level of mathematics and English/communications skills that is required to survive in our programs. Beyond this, there is a compelling interest to balance technical skills with communications ability. Therefore we invoke a narrow tailoring of the admissions process to achieve this goal.

The attention to liberal arts, and indirectly to domestic vs. international origin, has tremendous impact on our educational goals. The unique demand for domestic students is detailed below in a separate question. Many employers recognize that our emphasis on communication skills—embodied in our required course, “Statistical Consulting”, Stat 998, and in the rigorous, week-long MS exam—prepares both MS and PhD students for employment without the need for extensive retraining on the job.

Our policy is flexible, in that we take top-quality applicants with excellent mathematical skills and minimal communication skills, and we take risks with well-rounded applicants who have minimal mathematical skills. Sometimes, rarely, we make mistakes, but those better inform us for the next round.

We have only a limited number of positions, and cannot hope to admit everyone. We believe we often must make choices between two equally but differently qualified individuals. In these cases, there is never a single yardstick that would favor one over the other uniformly, and we often have extensive debates about these “border” cases.

Each year, after the admissions process is completed and all students have made their decisions (after 15 April), we have a brief recap of events. This includes a brief report to the faculty, as well as careful assessment of the workings of the admissions team itself. Later reviews include the assessment of communications skills in TA training in early Fall, collection of statistics to report to the Graduate School in late Fall, and broad periodic review of progress of students in each cohort.

Why are acceptance rates so low for guaranteed students?

Statistics typically gets 500-600 applications and offers 40-45 guarantees with an acceptance rate of 20-30%, although the rate was closer to 40% in 2012-13. Our acceptance rate is not atypical for other graduate statistics programs, with the notable exception of UC-Berkeley and Stanford U, the two top programs. Our rates are low because competition is steep, particularly for the small pool of domestic students. Further, the demand for training in statistics has risen, notably in the past decade, without a concomitant increase in room at universities for these students. At UW-Madison, Statistics has managed to keep graduate enrollment fairly steady at ~100-120 even as faculty numbers have declined; current graduate student-faculty ratio is 6-7, while the campus average is 4.4.

It used to be that acceptance rates differed markedly by region of the world. For historical reasons, having to do with early outreach by George Box and George Tiao, we have always had a large number of applicants from China. Until ~2000, the acceptance rate for guaranteed Chinese applicants was much higher than guaranteed domestic applicants (~35% vs. ~20%). Today, with globalization, the rates of acceptance are more similar across student origin, but are highly variable year to year. However, we still have 70-80% of our applicants from China, and only 10-15% domestic. Incidentally, if we only considered mathematical talent, our graduate program today would likely be >95% Chinese.

Why are domestic students viewed differently?

There are three reasons domestic students are viewed differently: 1) the unique nature of liberal arts education in the United States; 2) the pool of domestic students in STEM, and particularly statistics, is small; and 3) the restrictions on federal funding for training grants. Statistics is a collaborative field, where communications skills are highly valued; hence liberal education prepares students well for bridging across disciplines. STEM disciplines are now popular, but the numbers of domestic applicants for higher degrees in statistics is still quite low. This is a pipeline issue.

Graduate student training grants stipulate that only domestic students (nationals and green card holders) may be funded. This is particularly important in biostatistics. Since statistics and biostatistics students are trained together, we all work to increase the pool of top domestic candidates. Without a healthy proportion of domestic graduates, we literally cannot fill the training grant slots. Without those students in those slots, moving forward to complete PhDs, we cannot compete effectively for the training grants themselves. Thus, our whole program suffers when the number of domestic students is too low.

Why are so few URMs in statistics?

Please note that the field of statistics has done quite well in attracting and training women. The Department of Statistics has 25% women faculty, and has regularly a strong set of women graduate students. However, there are extremely few under-represented minorities (URMs) in the field of statistics (http://www.amstat.org/committees/cowis/pdfs/RepresentationReport.pdf). Those individuals who apply are courted heavily by all top-ranked programs. Our advising staff is working with the L&S Center for Academic Excellence on increasing interest in and awareness of statistics and mathematics among URM undergraduates, noting in particular the very strong job prospects at this time. As noted above, the low profile of statistics domestically, until recently, has contributed to the dearth of URMs in our field.

Updated 8 April 2016