Author: lifeofbai

The field of Statistics has evolved dramatically in the past decade. It has been very exciting but also a bit overwhelming. In this post, I attempt to synthesize seven of the major directions and trends that I have observed in Statistics. I foresee these areas continuing to play a prominent role in the field for the foreseeable future. A few caveats:

1. In what follows, I do not cover the analysis of particular data types, e.g., time-to-event data, spatial/spatiotemporal data, time series data, functional data, network data, etc. Analyses of these specific data types (among others) certainly remain very active areas of research.
   
   Rather than focusing on any particular type of data, I focus on “broad” modeling frameworks and paradigms. Indeed, all of the aforementioned types of data can be studied through the lens of the frameworks that I describe below (e.g., causal inference). It should also be noted that the below frameworks can be studied through the lens of both frequentist and Bayesian statistics.


2. This list below is by no means exhaustive. Indeed, there are many active areas of research in Statistics, some which I may have missed. If you think that I have missed something important, please feel free to describe it in the comments.


3. It is impossible to definitively predict what will be “popular” or “state-of-the-art” in the next few years. In particular, the widespread adoption of artificial intelligence (AI) in the past several years was not something that I could have foreseen years ago. Something new and unpredictable could always take off. However, I believe that the below areas are of broad current interest and will continue to be of enduring interest for the forseeable future.

So without further ado, let’s get to the list of seven major directions and trends in Statistics!

1. Generative Modeling and General Transport Problems

The advent of generative models has become a game-changer in many fields, e.g., medical imaging and electronic health records (EHRs). For example, it can be very costly and time-consuming to procure real brain scans and EHR data to study and diagnose disorders such as Alzheimer’s. However, with deep generative models (i.e., generative models parameterized by deep neural networks), it is now possible to generate very high-fidelity synthetic brain images and synthetic patient data using limited training data.

On a surface level, it may not be clear what generative models have to do with statistics. However, most deep generative models are implemented by mapping a reference probability distribution (such as a standard Gaussian) to a target probability distribution (for example, a distribution over images). In statistics, we encounter intractable distributions all the time, e.g., posterior distributions, conditional distributions, latent distributions, bootstrap sampling distributions, etc. It is of interest to adopt the idea of generative modeling to generate novel instances from these complex distributions so that we can learn them implicitly. This problem also lies within the broader framework of general transport problems where one aims to transform one probability distribution to another. To this end, mathematical tools such as optimal transport have gained considerable attention in the statistics and machine learning literature.

Using synthetic data as a surrogate for real data is also not a new idea. For example, in the analysis and design of many real-world physical systems, it is preferred to mimic these systems using simulated data and surrogate/computer models rather than performing actual experiments (e.g., it may not be feasible to expose a real aircraft to very dangerous weather conditions). Generative models represent a new frontier for generating high-fidelity synthetic data. However, in order for synthetic data created by deep generative models to be useful for clinical decision-making and downstream analyses, statistical inference methods are critical for ensuring the trustworthiness of these synthetic data. Thus, a very active area of current research is fidelity quantification, where statistical methods play a crucial role in assessing the fidelity of outputs from deep generative models.

2. Learning from Decentralized and/or Non-Static Datasets

Many “classical” machine learning methods were originally designed for static datasets on a single server. However, nowadays, it is increasingly common to encounter non-static datasets and decentralized data (i.e., data that is distributed across many nodes or clients).

A canonical example of non-static data is streaming data, i.e., data that arrives in real-time such as stock market data or clickstream data. The streaming nature of such data demands that these datasets be processed quickly and immediately rather than in batches. This requires the development of novel online algorithms and specialized statistical inference methods to capture dependencies in the data where the i.i.d. assumption is typically inappropriate.

Decentralized data, or data that is not stored on a single server but across multiple nodes, is also very common, especially in light of data privacy concerns. For example, individual hospitals often seek to collaborate on issues like disease detection but do not want to share sensitive patient data. This has led to significant interest in federated learning, where a global statistical model is built by only training a model on local servers and then communicating the updated model parameters (rather than transferring private, local data) back-and-forth with a central server until the global model converges.

3. Learning from Pre-Trained Models

In the early 2010s, there was a huge explosion in compute power. Thus, around the time when I was doing my PhD, there was a very strong emphasis on creating statistical methods and algorithms that could scale to massive datasets. While the design of scalable algorithms is still practically relevant, a new paradigm has taken hold in recent years: instead of having to train a new model from scratch — which could be very time-consuming, can we leverage a model that was already trained on a large dataset?

This is the premise behind all of the Generative Pre-trained Transformer (GPT) models and foundation models. The idea is that by fine-tuning a previously trained model, an agent such as a large language model or a self-driving car can rapidly perform a downstream task such as generating responses to a new prompt or navigating new traffic situations. The idea of pre-training also has relevance in statistics where a pre-trained model can often help to accelerate convergence of an algorithm. This is accomplished, for example, by using the pre-trained model as an intialization for the algorithm or by reducing the data requirements needed for learning a particular task (i.e., you may not need to use a massive amount of data to reliably perform this new task if you can simply tweak an existing fitted model).

Beyond the computational gains, pre-trained models also have relevance in knowledge and domain transfer. In recent years, there has been a considerable interest in transfer learning. Here, a model trained on one task (i.e., the source task) can be reused and tweaked for a second related task (i.e., the target task). This potentially allows the knowledge gained from the source task to boost performance in the target task, especially if data is limited or scarce in the target domain. Effectively addressing major challenges in transfer learning, e.g., negative transfer, will be an impotant issue to tackle in the years to come.

4. Learning Under Data Heterogeneity

Heterogeneity is frequently observed in real datasets and violates standard i.i.d. assumptions, which poses significant challenges for “traditional” statistical analyses. Current efforts for handling heterogeneous data include two major research directions: 1) ensuring reliable predictions and statistical inferences when heterogeneity is present, and 2) integrating heterogeneous data sources together to improve estimation and prediction.

Many statistical and machine learning models rely on fitting a model to a training dataset and then using this model to predict on unseen test data. However, if data shift occurs in the test data, then the model performance degrades and may produce biased predictions for the test data. For example, a model trained on younger adults may not work as well for predicting a certain health outcome in senior citizens. Handling the various types of data shifts, including covariate shift, label shift, and concept shift, is a major direction of current research. These issues are especially pertinent in transfer learning and federated learning, where data heterogeneity can significantly degrade the performance of the target model or global model.

Besides accounting for data shifts, there is also a great amount of interest in combining heterogeneous data sources and types together to improve model performance. This is also known as data integration, or data fusion. For example, in evidence-based medicine, a major trend is combining data from randomized clinical trials (RCTs) with real-world data (RWD) to both improve statistical efficiency and account for treatment effect heterogeneity. More generally, multimodal learning aims to integrate data from diverse sources (e.g. combining medical images, text data from clinical notes, and genomic information to predict a disease status) in order to enhance model generalizability, robustness, and predictive accuracy. In many cases, multimodal learning models reduce uncertainty and provide more compehensive, context-aware understanding than single-mode models which only use a single data source.

5. Causal Inference

Whenever we teach simple linear regression, one of the first things we caution students is that “correlation does not equal causation.” However, in many practical scenarios (e.g., evidence-based medicine), a cause-and-effect relationship is of greater interest than an association, especially if we aim to implement an intervention or public policy. We would typically be more interested in whether such an action (or treatment) actually works, rather than whether it is merely associated with a better outcome. Classical regression models can infer associations but not causal effects. Therefore, many statisticians have been working on moving beyond association to causality. In many adjacent fields to statistics such as econometrics and epidemiology, there has also been a noticeable long-term shift towards using causal inference to understand disease or economic etiologies and to evaluate the effectiveness of interventions and policies.

Randomized controlled trials (RCTs) are the gold standard for determining causal relationships. However, RCTs are frequently infeasible or unethical. As a result, causal inference often needs to be performed using observational data. Causal inference methods attempt to mimic an RCT by using statistical methods to create treatment and control groups that are comparable with respect to all relevant confounders and then inferring the average treatment effect (ATE).

Driven by the fact that treatment effects often vary significantly across subgroups in a population, there has been a greater emphasis in the past decade on estimating heterogeneous treatment effects (HTEs). This allows practitioners to identify which subpopulations benefit (or are potentially harmed) the most from a treatment. The increased emphasis on HTEs in areas such as precision medicine also coincides with the greater integration of machine learning methods (e.g., random forests or Bayesian additive regression trees) with causal inference. Advanced causal inference methods that address challenges such as treatment non-compliance, interference/spillover (where the treatment of one unit affects the outcome of other units), and unmeasured or hidden confounding are also being developed. Finally, there is also interest in causal discovery, i.e. finding the structure of causal relationships rather than estimating the magnitudes of causal effects.

6. Dynamic Decision-Making and Reinforcement Learning

Many real-world decision-making problems are made in a dynamic environment which requires constant adaptation to the current environment or state. Reinforcement learning provides a framework for optimizing such sequential decision-making processes. In particular, reinforcement learning enables agents to learn optimal behaviors (or an optimal policy) through trial-and-error and interaction with a dynamic environment to maximize long-term cumulative rewards. Given the explosion of precision/personalized medicine, a very active area of research at the intersection of reinforcement learning and causal inference is dynamic treatment regimes (DTRs) where individual treatment decisions are adaptively learned based on the patient’s changing health status over time.

Traditional reinforcement learning has focused on online algorithms, where the agent learns and updates its policy through real-time interaction with the environment. However, in many safety-critical applications such as healthcare and autonomous vehicles, it is very costly and dangerous to allow real-time environmental interactions. Therefore, a significant area of current research is offline reinforcement learning, where RL decision-making agents can be trained using pre-collected, static (offline) datasets. However, data shift and data heterogeneity are foundational challenges with offline reinforcement. Data shift causes overestimation of Q-values due to out-of-distribution actions, while data heterogeneity makes learning a single optimal policy difficult.

Due to the gap between fixed offline data and the dynamic data encountered from online exploration, there have been current attempts to bridge these two learning paradigms under the framework of offline-to-online reinforcement learning (O2O RL). O2O RL often improves sample efficiency and mitigates the safety risks of pure online training. By using a pre-trained policy learned from offline data as a starting point, the decision-making agent often requires significantly fewer online interactions to learn an optimal policy when deployed in an online environment. At the same time, online fine-tuning on dynamic data also helps to overcome practical issues such as suboptimal offline data.

7. Extracting Interpretable Insights from Black-Box Models

Black-box models such as tree ensemble methods (e.g., boosting, random forests, and Bayesian additive regression trees), neural networks, and Gaussian processes are considered the current state-of-the-art for prediction and classification tasks. This is because these methods are very flexible and capable of capturing highly complex and nonlinear higher-order interactions, while being less prone to the curse of dimensionality than more traditional nonparametric methods. However, they remain a bit opaque. As a result, there has been a great deal of effort devoted to making these models more interpretable.

One avenue for making these models more interpretable is through trustworthy uncertainty quantification. Conformal inference is a very popular method for calibrating black-box models. Conformal inference transforms point predictions into distribution-free prediction sets/intervals with guaranteed finite-sample coverage. Another avenue is through feature selection and feature extraction, e.g. representation learning. Feature selection aims to identify the most relevant individual features. Meanwhile, representation learning (e.g., disentanglement in variational autoencoder models) extracts hidden structure from the data in the form of compact, lower-dimensional, and semantically meaningful latent factors.

Finally, merging black-box models with ideas from causal inference can improve the interpretability of black boxes. One popular avenue right now is through counterfactual explanations, or identifying the minimal input changes that would fundamentally alter a model’s prediction. This can provide practitioners with practical insights like providing clear guidance on how to change a future outcome. Counterfactual explanations can also unveil when and whether decisions made by black boxes are due to spurious correlations vs. true causal relationships.

What’s the role of “classical” statistics, particularly mathematical statistics and probability theory?

Given the vast changes in the field of Statistics, one may wonder what the role of “classical” statistics is, particularly with regard to mathematical statistics and probability theory.

First, tools from “traditional” probability theory and mathematical statistics are still useful in these new paradigms and frameworks. Statistical theory can be used to analyze not just things like large-sample and finite-sample behavior or generalization error bounds, but theory can also help to explain certain phenomena for many popular models in surprising, and perhaps, unexpected ways. For example, I recently worked on a project where we used functional inequalities from advanced probabilty and functional analysis to characterize the fundamental behavior of transport maps in deep generative models. It is also important to build upon classical theory to develop new theoretical tools and frameworks.

Secondly, statisticians tend to be well-versed in statistical inference (e.g., uncertainty quantification and hypothesis testing) and regression models, e.g. linear regression and generalized linear models. These “classical” models — as well as other “classical” machine learning methods — can also be (and indeed have been) adapted to frameworks such as transfer learning, federated learning, and causal inference. We do not need to throw out the old or start completely from scratch. We can adapt existing tools to these new paradigms.

Finally, statisticians are experts at drawing interpretable insights from data. As black-box models and AI models have become much more prominent in decision-making, it will be essential for statisticians to use their expertise in inference to make these models more interpretable, robust, and trustworthy. This in turn can guide more informed decision-making.

My previous blog posts (such as my post on navigating the tenure-track job market) have reportedly helped many individuals. I do love helping people and sharing insights about what I have learned. So in this post, I figured that I would help another segment of people: aspiring PhD students in Statistics! This blog post attempts to walk prospective Statistics PhD students through the process, so they can know what to expect and how admissions works. However, aspiring graduate students in other disciplines may also find this post useful. That said, there are likely a few things that are unique to the Statistics discipline. In addition, this post focuses on Statistics PhD programs in the United States (the process could be very different outside the U.S.).

Things to Know Before Applying

I have served on the Graduate Admissions Committee for several years now at my current department. So I have been involved with graduate admissions for a long time. A few comments are in order first.

Graduate admissions in the field of Statistics are decided by an Admissions Committee made up of multiple faculty members. This committee almost always includes the Graduate Director or Graduate Coordinator, in addition to other current faculty members.

In other fields, it is often strongly recommended that prospective students reach out to individual Principal Investigators (PIs) if they are interested in joining their lab. In these other fields, the potential PhD supervisor’s recommendation carries a great deal of weight, since the applicant would be joining the PI’s lab directly if they are admitted. However, in Statistics, first-year graduate students typically do not join a lab or pick a supervisor right away. Instead, these students take a year of courses and then take qualifying exams. Only after a student passes the PhD Qualifying Exam(s) are they allowed to choose a PhD advisor.

Because it cannot be determined which students will pass the PhD Qualifier, most individual professors will not accept applicants directly as their PhD supervisees or agree to recommend applicants whom they do not know for admission. Instead, in most cases, it is solely the recommendation of the Graduate Admissions Committee whether or not to admit a PhD applicant.

Generally, the only exception to this is if a current professor in the department actually knows the applicant well (either through mentoring them on research or teaching the student in their class) and writes a recommendation letter for them. In this case, their recommendation will of course be taken into strong consideration.

You may want to dispel the notion of “safety schools.” A department’s selectivity is largely governed by the number of funded spots they have available. In recent years, my department has tried hard to raise the stipends for existing graduate students. Therefore, we currently target six to eight new graduate students every year (PhD and Masters students combined).

In addition, there may be other circumstances that determine the acceptance rate. In the past year, the University of South Carolina (USC) Board of Trustees approved waiving the application fee for all graduate programs at USC. Because of this, the number of applications we received to our graduate program skyrocketed to over 420 applications. Since our department typically admits around 20 applicants, our acceptance rate this past Spring was around 5%. (When we charged an application fee, the number of applications was much fewer, and the acceptance rate was in the high 20s/low 30s.)

Given this, it is hard to identify any particular program as a “safety school” for PhD admissions. In fact, is quite common for applicants to be accepted to “better” programs but rejected from “lower ranked” ones. This scenario happens all the time due to various factors (e.g. limits on the number of available funded PhD positions, policies that vastly increase the number of applications, etc.). So applicants should not be surprised if this happens to them.

Despite this, it is the case that applicants can make a list of “target” schools. The less ‘prestigious’ a program is, the more likely it is that the admissions committee will consider strong Masters performance as “compensating” for possibly weaker undergraduate performance. Whereas highly prestigious programs may demand academic excellence in all previous study, less prestigious ones may be more inclined to accept students who excelled in their Masters and/or who gained solid research experience but did not do as hot in their undergrad.

Another thing for international applicants to note: Unfortunately, if the admissions committee is not familiar with your undergraduate institution, then it may be more difficult to gain admission to a Statistics PhD program in the United States. Such applicants are sometimes viewed as riskier to admit because it is uncertain whether their mathematical preparation is rigorous enough or whether they can pass the Qualifying Exam. International applicants from universities that are not internationally recognized can improve their chances by first enrolling in a Masters degree in Statistics or Mathematics from a reputable school in the United States or Canada. It is also imperative to earn grades of A in first-year Masters courses.

The Initial Review of Applications

Shortly after the deadline, the Graduate Admissions Committee reviews the applications. In our department, the applications are scored on a scale of 0 to 10, with increments of 0.5. We mainly look for several things: 1) evidence of mathematical ability, and 2) research potential. The strongest applications we receive communicate these two aspects. It is helpful if the recommendation letters also discuss the applicant’s math/quantitative skills and their research potential.

Mathematical ability is normally assessed by academic transcripts, recommendation letters, and to a lesser extent, test scores. It is not necessary that the applicant have a degree in mathematics or statistics (although most applicants do), but they should have one in a quantitative subject such as Computer Science, Physics, or Engineering.

If the student has weak math grades in undergrad (too many B/C’s or even lower) and no Masters degree, then it might be difficult to gain admission. However, if the transcript shows significant improvement after freshman year or if the weak grades are from a single semester/year, then we do take that into account. In this case, it may be helpful if the applicant gives a brief explanation for this (e.g. dealing with medical/family problems, disruption from the COVID-19 pandemic, etc.), while also highlighting their improvement in later semesters in their application.

In addition, my department does take into account strong Masters performance. It is possible to be admitted to our PhD program if the applicant did very well in a Masters program in Mathematics, Applied Mathematics, or Statistics. Our bar for applicants who only have a Bachelor’s is higher (i.e. it is very difficult to be admitted if your undergrad GPA is much lower than a 3.5). However, if the Masters performance is very strong, then that can make up for a lower undergrad GPA. This is not the case at every PhD program, but my program does give consideration to this. So if your undergraduate performance is not so great, then you may want to consider getting a Masters first in order to demonstrate mathematical maturity.

There are of course some exceptions to this. Some institutions are known to have grade deflation (e.g. a strict policy that only a certain percentage of A’s can be given out, the top ranked student in his/her class “only” has an 83 out of 100, etc.). Some of these schools are known to the members of the admissions committee, but in case they may not be aware of such policies, it is strongly recommended that the applicant clarify this point in their application or have a recommendation letter writer explain these policies. In addition, if the GPA is difficult to interpret (for grade deflation reasons or otherwise), it can be helpful if the applicant lists their class rank, First Class Honors, etc. on their CV. Anything that can help to put the GPA in context is highly appreciated. Having a recommendation letter writer explain these things also adds credibility.

The second thing that we look for is research potential. Therefore, it is very important to convey interest in research and understanding that PhD study is about research in your statement of purpose. We do not expect applicants to know exactly what field of statistics they will do their dissertation research on. However, we do expect them to show interest in academic research.

So how can an applicant’s research potential be conveyed? It helps if they actually have research experience! It does not have to be Statistics research or necessarily research that has already been published. Some of the applicants we end up admitting have done research in pure mathematics, engineering/physical sciences, or public health. However, substantive research experience (in any discipline) through a summer REU and/or by assisting a professor on a research project is definitely viewed very favorably. Being a third author on a manuscript or publishing a paper in an undergraduate journal is also viewed positively.

For the most part, we do not care so much about the venue in which the applicant’s research was published. However, we do expect the applicant to describe clearly their contribution to the research and what they learned in the process. In this case, it can also be very helpful to have a recommendation letter from a research supervisor who can explain the significance of applicant’s contribution and their potential for PhD research.

So what if a PhD applicant does not have formal research experience? In some cases, an otherwise strong applicant may not have research experience and has been out of school for awhile, making it difficult for them to accumulate this experience. In this scenario, it it still very important to convey research potential. In particular, it may be helpful to discuss notable class projects and how these projects informed your interest in statistics research in your application. It may also help to obtain recommendation letters from professors for whom you did these class projects and ask them to discuss your potential for PhD research based on the projects you did for them.

Finally, the GRE is also considered if the applicant submits their GRE scores. The GRE is mainly used as a “sanity check.” GRE Quantitative scores that are too low may give us some pause. Some departments may filter by GRE (so scores below a certain threshold are automatically rejected) — however, my department does not do this. It is also worth noting that high GRE scores on the Verbal section or the Analytical Writing section are viewed positively. Most of the applicants have GRE Quantitative scores above 160, but it is far less common for us to get applicants with GRE Verbal scores above 160 or Analytical Writing scores of 5 or 6. So when we see an applicant with these types of scores, this actually gets noted favorably by the admissions committee.

Zoom Interviews

After each member of the Graduate Admissions Committee has reviewed and scored the applications, they meet to discuss the applications. Typically only applications that had an average score around a certain threshold are discussed. First, the committee may decide which applicants to nominate for additional scholarships or topoff awards. Secondly, individual committee members may decide to change their score after discussion. Finally, borderline applications may be discussed to determine whether to advance the applicant to the interview stage. After all the initial scores are finalized, we then move to the interview phase.

Not all Statistics graduate programs conduct interviews. At some schools, the admissions committee simply decides which applicants to send first-round offers to after meeting to determine their shortlist. However, in my department, we do conduct interviews with a “long list” of both PhD and Masters applicants.

Typically, the 30-40 PhD and Masters applicants with the highest average score receive Zoom interviews. The overall rankings may not change drastically after the interview. However, a stellar Zoom interview may push a lower-ranked applicant into the “first-round offer” category, while a poor Zoom interview may push an applicant out of the “first-round offer” category. In very rare situations, the interview gives a major “red flag” that causes the applicant to be deemed completely unacceptable. However, this is very rare.

These Zoom interviews typically last for 15 to 20 minutes. We do not “grill” the applicants about their technical knowledge or quiz them on their knowledge about statistics. Instead, in the first half of the interview, we usually ask the applicants questions specific to their CV/application (e.g. their work or tutoring/teaching experience, their research experience or research interests, etc.). If the applicant is looking to switch to Statistics from an adjacent field (like pure math, economics, or engineering), we may ask them why they want to study statistics. Or if the applicant is doing a Capstone project or writing a Masters or Honors Thesis, we may ask them about that.

The second part of the interview is usually reserved for answering questions that the applicant may have. It is crucial for the applicant to ask questions — ideally, specific ones about our department (e.g. Qualifying Exam, program structure, etc.). If the applicant does not ask us any questions, that may be viewed unfavorably and cause their application’s score to decrease. Conversely, if the applicant asks very thoughtful questions, the interviewer will take note of this and likely increase their individual score.

Therefore, to “nail” the Zoom interview for graduate school, you should certainly be prepared to: a) discuss your background, and b) ask good questions.

A week or so after the Zoom interviews are completed, the Graduate Admissions Committee meets once again to discuss every single applicant that was interviewed. Individual committee members may once again change their scores based on feedback about the applicants’ interviews.

First Round Offers and Waiting List

After the scores are finalized for a second time for the applicants who received a Zoom interview, the committee decides on whom to send first round-offers to. In all cases, the department has a target number of incoming students. And in most cases, the Graduate Committee expects that only a fraction of these first-round offers will accept the offer and matriculate to our program in the coming fall. This requires a bit of “guess work” based on historical yield. Based on past years, our department expects about 30 to 35 percent of the individuals whom we admit to accept our offer. Therefore, if we are aiming for an incoming cohort of six to eight students, we will likely send out around 20 first-round offers. This usually happens in mid- or late February.

The first-round offers are made to the highest ranking (say) 20 applicants. Meanwhile, everyone else that was interviewed remains on the “long list” (or the waiting list).

In late March and early April, we may have heard back from many of the first-round offers, and then there could be some movement from the waitlist. If a higher percentage of the 20 first-round offers have declined than we would like, then we will move down the “long list” in the order that they are ranked and send out more offers to other applicants on the waitlist. Often times, even many of those on our waitlist have already received a competing offer from a different program that they prefer, in which case, we would then move to the next available applicant.

April 15 is the “official” deadline that PhD and Masters admittees are supposed to decide where they will be attending in the fall. Some first-round offers hold onto their offers and do not decline us until April 15. As a result, there may be some last minute offers of admission sent out to waitlisted individuals around April 15.

I would say that if you are extremely interested in a particular PhD program but are on the waitlist, you should wait until April 15 to see if there is any waitlist movement. If it is April 15 (or one of the days leading up to it), the Graduate Coordinator may reach out to you to ask your interest. But if you have not heard back and you remain interested, then you should certainly reach out to the Graduate Director and ask for a status on your application. If the target class size has been met, then the Graduate Director may choose not to admit any more PhD students. In this case, all the remaining applicants will be rejected. However, it does not hurt to ask.

Furthermore, if you remain very interested in a PhD program but are waitlisted, you should periodically keep in touch with the Graduate Director. Consistently conveying this enthusiasm creates a positive impression and has the potential to result in an offer of PhD admission if there are still available funded PhD slots.

What If I Didn’t Get In?

Hopefully, you will have received a PhD offer that is acceptable to you by April 15. However, in the unfortunate event that you do not receive any offers, you should keep a few things in mind.

First, there is a lot of noise in the processs. While the Graduate Admissions Committee does its best to identify the best applicants, there is invariably some degree of subjectiveness by the committee members. And it is possible that we overlooked some applicants.

Secondly, it is competitive. As I discussed above, we interview 30-40 PhD and Masters applicants for 6-8 available slots. If we granted you an interview, then that means we found your application strong. The overwhelming majority of those whom we interview would be perfectly acceptable graduate students (unless there was a major red flag in the interview, but again, this is very rare). In most cases, the primary reasons for rejection are just “The applicant did not rank high enough” and “There aren’t enough spots left.” You should keep in mind that your chances of admissions are not based on how your application looks in isolation, but rather, how it compared to all of the other applications that we received.

Finally, it is worth figuring out any potential shortcomings in your PhD application. Maybe the application didn’t mention research enough. Or if you only have a Bachelor’s and your GPA is a bit on the lower side and your grades in math classes are subpar (with too many grades of B or C), then it might be worthwhile to consider doing a Masters degree first and then reapplying to PhD programs.

Strong performance in a Masters program can certainly boost an applicant’s chances, at least in my program and programs at peer institutions. There is no surefire guarantee that this will get you admitted, but it does increase your chances of getting admitted somewhere. If possible, obtaining a Masters degree from a reputable university in the USA or Canada is also recommended for international applicants from undergraduate institutions that we may be unfamiliar with.

This post is a continuation of my previous post on the lessons that I have learned from being a tenure-track Assistant Professor for two years at an R1 research university. In this post, I discuss research, service, and external grant funding. My hope is that this post may be useful for new faculty who are just starting out on the tenure-track or for aspiring tenure-track faculty at research-oriented universities.

Research

1.  Make sure to work on research and/or write nearly every day. If you are at a research university, then it is important to recognize that research and teaching will likely be the most important parts of your tenure dossier. In addition, exceptional teaching and service will generally not be able to compensate for a “fair” or “poor” research record. Different universities will have different expectations about what constitutes a good research record, but it is definitely safer to try to exceed the minimum expectations when it comes to research. To this end, you have to make sure that you are devoting enough time to research.
   
   
2. Try to strike the right balance between quality and quantity. Most departments do weigh quality. So if you are able to publish a few articles in the most prestigious outlets, then it is very likely that these papers get weighted more heavily in your tenure and promotion dossier. Having a few papers in the most prestigious outlets will universally be viewed more favorably than having a prolific publication record in mid-tier or low-tier outlets.
   
   At the same time, junior faculty should also keep quantity in mind (though certainly not at the expense of quality). I know people who have been denied tenure because they only had a single Science/Nature/Annals of Statistics/etc. publication in the span of four years. Although there is definitely a premium for very prestigious journals, you typically cannot get away with only one or two publications, even if they appear in the most top-tier journals. On the flip side, I also know people who have been denied tenure because they had too many publications in outlets that were considered to be mid-tier or low-tier or because their publications were mainly from interdisciplinary collaborations, rather than first-author or last-author.
   
   If you are a tenure-track faculty at an R1 university, then you need to regularly have first-author publications and/or last-author publications where one of your supervisees is the first author (as you progress further in your career, most or all of your papers will fall in the latter category). Some of these papers can be review papers, and not all of the papers need to be in “high-ranking” outlets. However, you might need to have a few that are in “high-ranking” places. Note that “high-ranking” does not necessarily mean the most prestigious journals or conferences in your field, but it does mean that the outlet has to be considered a “premier” outlet in some sense.
   
   
3. Find out what your department considers “top-tier” journals and conferences. As I mentioned in the previous point, your department might like you to have at least a few articles in venues that are considered “top-tier” or “high-ranking” in your specific research area. I am not necessarily talking about the three or four most prestigious journals in your field, but a longer list of journals and conferences that are considered to be “high-ranking.” For example, the London School of Economics (LSE) maintains departmental journal lists (list from 2020-2021) about what LSE judges to be “high-ranking” journals.
   
   Note that your department may not maintain an ‘official’ list. However, you can probably still get a rough idea or an “unofficial” list of what the department considers to be “high-ranking” venues if you ask someone.
   
   
4. Say “yes” to as many invitations to speak at colloquia and conferences as possible. Unless you have extenuating circumstances that prevent you from traveling (there are certainly legitimate reasons not to!), I would generally say that it is a good idea to accept invitations to give talks at department seminars and conferences. In your tenure evaluation, most research universities will solicit external letters of recommendation from full professors who are considered experts in your research area. So it’s generally in your best interest to make yourself as widely known outside your department as possible. Plus, this is a great way to get to meet other researchers and foster potential collaborations.

Service

Service is also a very important part of the job, but for junior faculty at research universities, I would definitely encourage them to “selfishly” guard their time. I don’t mean that you should ‘skimp’ on service, but definitely make research and teaching a higher priority. At research universities, most departments do try to shield their junior faculty from taking on too much service. But you should still be cautious to make sure that research and teaching are prioritized above all else. As you advance in your academic career, you will have more service and administrative responsibilities. But very early on, service should be the “cherry on top” of a high-quality research and teaching record.

1. Keep in mind that outstanding service generally can not make up for a poor teaching record or a middling research record.  Just because your department tries to keep your service load “light” does not mean that you won’t often get service requests from different individuals. Even if it is “only one” request, if you get many of these requests from multiple individuals, then the time investment might add up. For this reason, I do advise junior faculty to be cautious about their time and try not to put too much extra stuff on their plate.
   
   
2. It is a very good idea for junior faculty to do one major piece of service to their department each year. This could be serving on a departmental committee, such as the Graduate Admissions Committee, the Qualifying Exam committee, or the Hiring Committee. It could also be organizing the departmental colloquium.
   
   I think that most junior faculty find these types of service to be quite positive experiences, as they do help Assistant Professors to integrate into the department and learn about its culture and inner workings. Doing one major piece of service each year also fulfills the requirement of acceptable service to your home department. At the same time, if you feel you are being given too much service that it is taking time away from research and teaching, then don’t be afraid to speak to the Department Chair. He or she will try to remedy this situation and reduce the service load.
   
   
3. You do need to do service to the university and the greater community, but don’t be afraid to be “choosy” about the service tasks you decide to undertake. For me personally, I typically always agree to serve as a committee member for a PhD dissertation. This type of service is valuable to both me and to others at the university.
   
   In addition to serving on PhD thesis committees, I have done a few other service projects here and there, such as reviewing applications and interviewing prospective students for the Top Scholars program at my university. I also served as a lunch ambassador for the New Faculty Orientation this year. It’s important to do some service, but you can be selective about what you do, and if you really are too busy, you can politely decline. If you do decline, it’s always best to suggest alternatives, e.g. people who may be able to take your place in completing the task.

Grant Funding

First, I would make sure to find out exactly what the requirements for tenure are with respect to external grants. In some departments (e.g. lab sciences), it is essential to actually get grants in order to keep your lab running. In other fields, the requirement may just be that you regularly seek (i.e. apply for) external grants. Even if your department does not require you to get external funding, there are many benefits to having grant money, not least of all: summer salary, paying for your graduate students to be full-time Research Assistants, scholarly visits to other institutions, more computer equipment, etc. All of these things are beneficial to your research, so if you are at an R1, then it is a good idea to at least regularly apply for external grants.

With regard to grant funding, I have a few suggestions as well:

1. Apply for grants and start learning the grant process early. One of the mistakes I made was not hitting the ground early enough. My first year was a “pandemic” year (2020-2021), so I was a bit ‘disoriented’ to say the least (i.e. I was not on campus at all during this year, so I did not get to meet with anyone in-person who could walk me through the process). There was a lot to navigate, to say the least. Nevertheless, I still probably could have found the resources that I needed to put together a grant proposal during the height of the pandemic. Even if my grant submissions had not been successful, I could have learned a lot from the experience, so I could hopefully succeed the second (or third) time around.
   
   The Chair of my department has also indicated that it can take at least six months to write a competitive grant. Given the amount of effort required, it is definitely a good idea to start early.
   
   
2. Volunteer to serve on a grant panel.  This past spring, I served on the NSF’s Launching Early-Career Academic Pathways in the Mathematical and Physical Sciences (LEAPS-MPS) panel, where I reviewed grant proposals and deliberated on them with other panel members. I learned a lot about the entire process from the “the other side,” i.e. how a grant panel determines which grants are “highly competitive” vs. “competitive” or “not competitive.”
   
   By serving on a panel (as an aside, you do get compensated for this from the funding agency), you can learn how to better consider things from the reviewers’ and panelists’ point of view and tailor your grants accordingly. For example, on the panel that I sat on, a great deal of emphasis was placed on how convincingly the grant proposal addressed the specific LEAPS-MPS solicitation criteria. In addition, I was able to gain crucial insight into what exactly “stood out” about grants that were rated as “Highly competitive” by a panel (e.g. in terms of novelty of the proposed research, clarity of the specific aims and structure of the proposal, broader impacts, etc.).
   
   
3. If your university offers a grant-writing mentorship program for junior faculty, then sign up to participate in it. This year, I am taking part in a nine-month mentorship program offered by the Office of the Vice President for Research. The aim of this mentorship program (which consists of workshops, intensive mentoring, and presentation coaching) is to equip junior and early-career faculty with the skills needed to develop competitive grant proposals. At the end of the nine months, each participant will have put together a proposal to submit to either the NSF or the National Institutes of Health (NIH).
   
   Putting together competitive grant proposals is a skill that takes a lot of time to develop, but these sorts of mentorship programs can help you to round the learning curve a bit faster.
   
   
4. Get examples of successful grant proposals from your colleagues and ask them to help you with your own proposals. I have several examples of grant proposals that were successfully funded by the NIH or the NSF. These sample grants have been extremely helpful in knowing how I should structure the project description and what things make up a “competitive” grant.
   
   Senior faculty are also likely to be very experienced with the grant proposal process, and they would be happy to offer their guidance. They can help to review your project summary, suggest ways to enhance your overall proposal, how to write an effective rebuttal (if resubmission is allowed), etc. So definitely ask your colleagues to help guide you through the process.
   
   
5. Make sure to budget carefully. I received an internal ASPIRE-I grant from my university, and I am currently the Principal Investigator (PI) on an NSF grant. My entire ASPIRE-I grant was devoted solely to supporting one PhD student as a Research Assistant (RA) for one semester, and a good chunk of the NSF grant is spent supporting another PhD student as an RA for two semesters. When you support a Research Assistant, you need to not only pay their graduate student stipend but also their tuition, health insurance, fringe benefits, etc. This is only one aspect of the entire budget too, so plan accordingly.
   
   Fortunately, you don’t need to do the budget planning entirely by yourself! Your college should have a Grants Operations or a Grants Administration Office that will help you plan everything (compensation for yourself and RA’s, equipment, travel, etc.) meticulously.
   
   

Tomorrow will be my two-year work anniversary as an Assistant Professor of Statistics at the University of South Carolina. I decided to write a two-part series of posts about what I have learned over the past two years about teaching, supervising PhD students, research, grant writing, and service. In this first part, I will focus on the first two aspects: teaching and PhD student supervision (read Part II here). My hope is that this post may be useful for new faculty who are just starting out on the tenure-track or for aspiring tenure-track faculty at research-oriented universities.

Teaching

First of all, I strongly recommend that all new Assistant Professors become familiar with their department’s tenure and promotion unit criteria. These unit criteria should clearly outline what is expected in terms of research, teaching, and service. Knowing these expectations will allow you to plan your teaching activities.

Roughly speaking, to be rated as “good” or “excellent” in teaching for my department, the teaching expectations are: average/above average student evaluations, good/excellent peer review evaluations, and at least one additional contribution in the area of teaching. For my department, the additional contributions could include things like developing a new topics course, supervising Masters and undergraduate senior theses, and participation in teaching workshops organized by my university’s Center for Teaching Excellence. In order to satisfy the additional contribution aspect, I developed my own graduate topics course and attended a series of New Faculty Academy workshops so that I could get a certificate of completion.

Every faculty member has their own teaching style, but I would say that there are a few good practices to follow for teaching.

1. Familiarize yourself with the classroom and classroom technology before the first day of classes. During the pandemic, due to mandatory COVID-19 testing and quarantining, instructors were strongly recommended to livestream and record their in-person lectures. I asked a colleague to show me how to do this through Blackboard. I also found that some students really liked being able to rewatch the lectures anytime they wanted (e.g. to prepare for exams or because they were particularly interested in mastering certain course topics). Thus, I decided to record all my lectures going forward, pandemic notwithstanding.
   
   Before the first day of classes, I also always make sure to familiarize myself with the classroom in which I am going to teach. This allows me to get acquainted with the technology in that room and ask the IT team to fix any issues if there are any (e.g. equipment needing to be replaced).
   
   
2. Give students frequent reminders and spell out your expectations of them explicitly. I found that students respond positively if you frequently remind them about upcoming deadlines or exams, both in and out of class.
   
   In addition, try to keep in mind that some things that you may think are obvious might not be to the students, unless you spell it out explicitly. For example, if you think that using sites like Chegg constitutes cheating, then you may need to be very explicit about forbidding their use. If you are teaching a class that includes a coding component and you want all students to submit their own individual code, you may need to specify clearly at the top of the assignments that while students are allowed to discuss “general concepts” with their peers, the code they submit must be their own work and that copying any snippets of code word-for-word is not allowed.
   
   In short, it is perfectly fine – and even appreciated – to repeat important reminders to students and to state/reiterate expectations to ensure that your students are on exactly the same page as you.
   
   
3. Be receptive to feedback from peer evaluations. Your department may have senior faculty sit in on one or a few of your classes to give you feedback about your teaching. These peer evaluations are meant to help you improve. I think that every educator could always stand to find ways to improve their teaching practices and effectiveness. So be sure to thoughtfully reflect on the feedback that you receive.

PhD Student Supervision

One of my favorite parts of this gig is being able to supervise my own graduate students. However, this is something that takes time to learn how to do well. Every faculty member has their own approach to this, with some being more hands-on and others being more hands-off. Some PhD mentors meet with all their advisees as a group, while others prefer one-on-one meetings (I personally do the latter). I am still learning how to be an effective PhD mentor, but I did find that there were a few general strategies that worked for me.

1. Be clear about your expectations with your advisees, including what milestones you want them to meet. For me personally, my expectation for my PhD advisees is that they write three papers, two of which should be submitted by the time that they defend their dissertation and the third one nearly ready to submit (or in the best case scenario, also submitted). Of these three projects, I expect at least one of them to be mostly/completely the student’s own idea (though I may suggest a general direction for them to explore and help them to refine their idea so that it can be feasibly completed in a reasonable timeframe).
   
   In addition, I expect my students to have submitted their first paper and have the second one nearly ready to submit before they do their PhD proposal.
   
   
2. Try to adapt your approach based on the needs and strengths of individual advisees. Every PhD student is different. Some students may be very independent from the start, while others may need more “hands-on” guidance and more structure in the beginning. I try to determine each student’s needs and adjust the amount of structure accordingly. For example, if needed, I may outline more specific directions and give more specific instructions (e.g. what papers to read, what algorithms to try to implement). Or I may write up a short note that explains key concepts, so that the student can gain a better understanding of the foundational material.
   
   
3. Supervising students takes a lot of time, so bear this in mind if you are still not tenured. I meet with each of my advisees one-on-one once a week. But beyond this weekly meeting, I also spend time reflecting on our meetings and planning the “next steps.” This includes thinking about what potential issues or challenges may arise, what other simulation studies we should do, and what other ideas we should test out as a proof of concept in case one of our initial ideas failed. Sometimes I may type up a short note for the student as a ‘quick’ introduction to an important concept needed for their research. A few of my students are just now writing their first papers, and it also takes a lot of time to come up with the basic paper structure and the contents to include in the paper, to proofread, etc.
   
   If you are not yet tenured, then you might still need to do your own research and get at least a few first-author publications. Or you may need to successfully get grant funding in order to sustain your research after your start-up funds run out (Note: the grant requirements differs widely by department and discipline – in math and statistics departments, many students are supported by department teaching assistantships). Writing/revising your own first-author papers and writing competitive grant proposals also takes a considerable amount of time, on top of teaching.
   
   This academic year, I am preparing to submit grant proposals to both the National Institutes of Health (NIH) and the National Science Foundation (NSF). It may take at least six months of preparation for these proposals to be considered “competitive.” I am also considering having my mid-tenure review. Therefore, I will need to revise several manuscripts and hopefully get them accepted.
   
   Given the enormous amount of effort required for the aforementioned things, I decided to tentatively not accept any new PhD advisees for this upcoming academic year. If I am very confident about the progress of my current PhD advisees and we are able to submit two new papers by Spring of 2023, then I will reconsider taking new advisees in spring 2023. However, at this time, I do feel that in order to guard my time and complete several other things that I hope to accomplish this year, I should put a pause on accepting new PhD advisees.
   
   I am somebody who has a hard time saying “no.” But depending on how much bandwidth you have, it may be prudent – or even necessary – to limit the number of PhD advisees at times when you are especially busy. You can always reconsider at a later time depending on how well things are going. But make sure that you are protecting your time for your own research and teaching.

It has been about six months since I made the transition from postdoc to Assistant Professor (more on my experience on the academic job market here). I figured that it would be a good idea to write a timely blog post about this transition. I should note that I did make this move during a global pandemic, so some of my experiences are probably not “typical” of a first-year Assistant Professor on the tenure track. For example, I will have spent my entire first year teaching graduate courses completely online rather than in-person. Nevertheless, there are perhaps still many things about this transition that are universal and broadly applicable, which I detail below.

 

• As a new Assistant Professor, you have a lot more responsibilities, and it can take some time to figure out the “optimal” way to balance priorities. In my third and fourth years of graduate school, I was supported by a fellowship so I had no responsibilities except for my dissertation research. After finishing graduate school, I did a postdoc where research was also my only major responsibility. However, after starting my new career, I suddenly also had to teach and advise PhD students, in addition to a few service requirements. Needless to say, it can be tough to juggle these additional responsibilities.
  
  It took me awhile to figure out how to go from spending most of my work hours on research to adding teaching and service on top of that. After a bit of trial and error, I realized that I am most productive at research-related tasks in the morning. That is, if I have to write code, conduct data analysis, write/revise manuscripts, or work through technical details and proofs, then I am most “focused” in the morning.
  
  Because of non-research responsibilities (and partly because of pandemic fatigue), I find it somewhat difficult to sustain this concerted effort towards the above research tasks past a certain hour of the day. Therefore, I requested to teach my classes in the afternoons. This way, I am usually able to get five to six hours of solid research done every weekday before lunch. Then in the afternoons, any “research” I do is not so much focused on the nitty-gritty aspects of research, but instead, it mainly consists of meetings with collaborators/PhD students and e-mail exchanges between me and research collaborators. However, outside of these meetings and e-mails, I focus the majority of my effort in afternoons on teaching — both on actual teaching (delivering lectures) and doing other course preparation, such as preparing lecture notes, grading, etc.
  
  Of course, this is not a “strict” schedule. If I have upcoming revisions or research deadlines, I might devote an entire day when I am not teaching to research. Or if I have a lot of grading or lecture notes to prepare, I might devote an entire day’s effort to teaching. But in general, I find that devoting my mornings to research and afternoons to teaching helps me strike the right balance. Depending on what time of day you are most productive on research, I would recommend that junior faculty schedule their teaching around those hours.
  
  
• Try to make teaching part of your research. Related to the first point, you can also help move your own research along by incorporating aspects of your research into your teaching. My first semester on the tenure track, I taught my department’s graduate-level course in Linear Models. I spent a good deal of time reviewing matrix algebra, which is a very useful tool for my own research. This semester, I am teaching a special topics class on modern high-dimensional data analysis. Since some of my recent research has focused on methodology and theory for non-Gaussian generalized linear models (GLMs), I included a unit in my topics class on classification and GLMs.
  
  Teaching these classes has really helped me to reinforce important concepts and make sure that I stay on top of the relevant work in these areas of statistics (which is of course helpful for my own research). In addition, this has also enabled me to get to know some of the graduate students in my new department. I would highly recommend that new Assistant Professors at research universities volunteer to teach graduate courses and topics courses related to their own research. This can be very helpful in getting to know some of the graduate students in your department and getting new PhD students to work under your supervision (if they really like your class).
  
  
• Be kind to yourself the first semester. Everything can seem a bit overwhelming the first semester. It definitely takes some time to get acclimated to a new environment. Moving to a new place, buying new furniture, getting a new driver’s license, and all that kind of stuff takes a lot of time that can really eat into research time. Moving during a pandemic also presented its own unique set of challenges. Therefore, I had to be realistic about how much work I was really going to get done the first semester, besides teaching. While I was successful in fulfilling my teaching duties, I did often feel as though I was falling behind on research during my first semester. You just have to be kind to yourself and acknowledge that it can take some time to adjust to your new environment and new responsibilities. After a couple of months, I was able to catch up on research and figure out the optimal balance for myself personally (as noted in my first bullet point).
  
  
• At the same time, don’t struggle by yourself in silence. Be sure to reach out to your colleagues for help if you need it. I have great colleagues whom I regularly consult for help on things such as university computing resources, using my startup funds, and other things that I have no prior experience with. For example, I plan to apply for my first research grant as a Principal Investigator (PI) this year. To this end, I asked my colleagues who had gotten their grant proposals successfully funded in the past to share their successful grant proposals with me. These examples will help me to craft my own grants this year. I also plan to ask my colleagues for feedback on my proposals in order to maximize their chances getting funded. I have found that everyone is very warm and willing to help out. So don’t be afraid to reach out to others if you need help.

All in all, I would say that my first six months on the tenure track have been pretty successful. It has been quite a learning experience for sure, but I have gradually gotten the hang of it.

I am moving to Columbia, SC at the end of this week to begin my new career as an Assistant Professor at the University of South Carolina. I have been doing academic research for four years now. I began doing research after passing my PhD qualifying exams in the summer of 2016. Following completion of my PhD in 2018, I completed a two-year research postdoc. After four years, I can say that I have learned quite a bit. Before beginning the next phase of my academic career, I figured it would be a good idea to write down all that I have learned in the past four years.

• Be ready to start over or shift gears at any time. Research is not easy. Often times, things don’t quite go as planned. You might have had some idea in your head how something would turn out, but after working on it awhile, you just couldn’t solve it or it turned out not to work. Or your research might end up going in a totally different direction than you initially “planned.” That’s totally fine! There have been many times when a project that I initially conceived turned out totally differently — either because I found out the ultimate conclusion wasn’t what I was expecting, or because I had discovered that another researcher had already done something very similar (then I was kind of forced to change the scope of my project if I wanted to be able to publish my work). The unpredictability is partly what makes research so exciting for me.
  


• Rejection gets easier over time. The first paper I wrote as a PhD student was actually rejected four times before finally being accepted. I have also gotten several other rejections on manuscripts over the years. At first, it was quite crushing to have worked so hard for months only to have my work be rejected. But over time, I have learned to deal with it.
  
  If you are aiming for more prestigious journals or conferences where the acceptance rate hovers around 10-20 percent, then it will naturally be more difficult to get your work accepted. I do recommend “aiming high,” so I do often initially try the most prestigious venue I can think of that could be a good fit for my work. I’ve occasionally had success and gotten an invited revision from one of these venues, but other times, I haven’t been able to. So I just revised my work according to the suggestions from the reviewers and tried another decent journal.
  


• If you have a particularly good idea, then keep “milking” the idea for several papers. The first paper from my postdoc was finally accepted in a top field journal this past year. Since I was intimately familiar with the topic, I followed it up with two other papers (here and here) that were related but that still made novel contributions. The first paper was about a group regularization prior used in the contexts of Gaussian grouped linear regression and Gaussian additive models. In my follow-up works, I extended the use of this prior to varying coefficient models and to non-Gaussian data via the generalized linear framework. The contributions in my follow-up papers were still original in their own right, but they built upon tools that I was already familiar with.
  
  If you are hoping for an academic career at a research university, then quantity does matter. Of course, quality matters the most (so articles in “top” journals and conferences count for more). But you definitely need more than just one “major” paper in order to guard against potential claims of poor productivity for promotion (e.g. from Assistant to Associate Professor). Because the entire publication process could take well over a year from first submission to final acceptance in my field, time is of the essence. It is typically easier to conceive new papers that extend your existing work in new directions, so I recommend doing that as much as possible.
  


• At the same time, keep learning other topics so you can move onto something new in the future. While I have a lot of other ideas how to possibly extend my current work, I am hoping to have some of my future PhD advisees work on them instead of me “hogging” all the ideas. Plus, even the best ideas could eventually become so commonplace (or even obsolete) that it becomes much harder to publish papers on them in good journals. Given this fact, I know that I always have to be ready to “move on,” so to speak. In view of this, I try to spend a good chunk of my research time just reading about and fostering collaborations with other researchers on new topics, especially in emerging fields of interest where there isn’t as much work done.
  
  To this end, I have been currently exploring new approaches to uncertainty quantification in complex Bayesian models, especially with non-Gaussian data (e.g. sparse and/or correlated binary data). There is a lot of exciting work to be done with regard to both scalable Bayesian inference algorithms and the theoretical underpinnings.
  


• You’re never truly “prepared,” so try not to worry too much about it and just do your best. I have basically been “underprepared” for everything at every stage of my professional life. Whenever I have changed careers (before returning to academia, I worked in engineering and finance), I have had to learn an entirely new field and familiarize myself with a completely new set of tools. Now, whenever I decide to pursue a new research direction, I usually only have a vague idea about the topic. To become more familiar with the topic, I need to spend quite a bit of time teaching myself the subject, reading the relevant literature, and learning a new set of relevant tools.
  
  As I begin the next stage of my career, there is still so much that I don’t know. As an Assistant Professor at an R1, I will need to learn about obtaining external grant money, supervising PhD students, and teaching graduate students. These are not things that I have any prior experience with. It can all be very intimidating. But I just remind myself that I have been in this position of “unpreparedness” before, and that I can learn to excel at it if I work hard at it, ask questions, and gain a little bit more experience.

Update on 2/29/20: I am excited to be joining the University of South Carolina Department of Statistics as an Assistant Professor in August!

I figured it would be a good idea to write a post about what I learned from my experience on the tenure-track academic job market. This post pertains mainly to jobs at major research universities in the United States in the fields of Statistics and Biostatistics, but some of this could nevertheless generalize to other fields. Also, of course, this is just a sample size of n=1. In what follows, I describe the overall job application process, my application materials, and the interviews.

Deciding Where to Apply

In the fall of 2019, I applied to 53 tenure-track positions, all at research universities in the United States. I applied to positions in Statistics, Biostatistics, and Mathematics departments. For all of my efforts, I received 8 phone screening interviews, 3 of which turned into campus invitations. I also received 2 campus invitations without a phone interview, bringing my total up to 5 campus interviews. Most application deadlines were in early November through early December, and I heard back from most schools before Christmas (one I heard back from in early January).

Update on 2/29/20: After I had accepted the position at University of South Carolina in February, I also received several more e-mails from various schools asking if I was still available/interested. So if you haven’t heard back by January, do not panic! Some schools may decide to do a second round of interviews in late February or March, and if you were ranked right below the threshold for the first-round campus interviews, you could still be contacted for an on-campus interview in late February or March.

There are over 4000 institutions of higher learning in the U.S., but only around 400 of them grant doctoral degrees (research universities). Of these doctoral-granting schools, 259 of them are classified as “R1” (very high research activity) or “R2” (high research activity). So that means 90+ percent of colleges do not award PhDs, and the vast majority of academics teach at primarily undergraduate institutions (PUIs), regional comprehensive schools (which may award Masters degrees), or community colleges. At the R1 research universities, the teaching load will typically be 2-1 (two courses one semester, one course the other), sometimes 2-2 or even 1-1 at very elite universities. The primary criterion for tenure at R1s is research (i.e. publications, grants) and supervision of PhD students. Most R1’s will reduce the teaching load for Assistant Professors to one course a semester for their first few years in order to allow them to focus on their research. At PUIs and regional comprehensives, a typical teaching load is 3-3 or 4-4 (at a select few prestigious colleges, it is 2-2), and at these institutions, the primary criterion for tenure is teaching and service to the college. These schools also require some research to get tenure (with the elite colleges requiring more), but typically, the publishing requirements will not be as intensive as those at R1s, because they want you to focus most of your time on your teaching.

I decided to focus my job search on research universities. Due to the fact that there are relatively few jobs at research universities and not all of them will be hiring in your field every year, I cast a wide net to improve my chances of landing some interviews. That meant that I applied all over the country and was not particularly picky about geographic region or location. From what I am told, the most important things required to get interviews at R1s in my field are a strong publication record and strong letters of recommendation. For teaching-focused schools, your teaching experience (e.g. having taught a course as instructor of record) is vital to success in the job market and some publications are also helpful as evidence of scholarly output.

At the time that I sent out my job applications, I had six papers, five of which were first-author papers and at least one which was forthcoming in a top journal in my field. One of my PhD classmates who went on the job market the year before me also had the same number of papers as me when he applied, including one in a top journal. He also managed to land a tenure-track job at a research university. So in order to be a competitive enough candidate for tenure-track jobs in Statistics at an R1 or R2 research university, I would say that this is a good benchmark. Of course, there are job candidates who had even more papers than I did, including multiple ones in top journals, and I assume that they landed interviews at multiple top schools.

There is a fair bit of luck involved in getting an interview because most programs receive many more applications from qualified applicants than they can offer interview spots to. It is not completely dumb luck (so applicants with no publications, PhDs in a completely unrelated area, etc. will be eliminated in the initial review of applications). But since most programs will only invite 3-5 job candidates for campus interviews for every open position advertised, making it to the campus interview often involves things beyond your control.

For example, if the only professor in the department who teaches/researches probability theory is about to retire, then the search committee may prioritize applications from probabilitists. If the department is looking to grow its faculty specifically in one area (e.g. I saw a job ad that stated that the department would give special consideration to applicants with research focused on environmental and spatial statistics), then you may be out of luck if your research is unrelated to that. Some programs may have these kinds of preferences but not explicitly state so in their job ads. For example, I applied to a math department at one school which had statistics under its umbrella, but I later heard from a colleague there that the search committee decided that they would focus on hiring an applied mathematician rather than people in other specialties. So I was out of luck there. Finally, if a department has recently hired a bunch of people with a very similar research focus as your own subfield, they may opt to pass on your application because they want to diversify the department’s expertise. These are all things beyond your control. This is partly why I recommend applying widely to maximize your chances.

My cover letters briefly outlined my education/experience and my research accomplishments. Each cover letter also included a paragraph or two customized specifically for each department I applied to, including: why I was excited about that department in particular, which professors in that department I could potentially collaborate with and/or whose research was highly complementary to my own, which courses I would be prepared to teach if hired and what special topics courses I could offer, and what (if any) departmental or university initiatives I would like to be involved with if hired.

My research statement was three pages long, excluding references. The first two pages described my general research interests and summarized my past papers and present research projects. The last page of my research statement expounded upon my future research plans and topics that I planned to pursue in the future. In general, I tried to keep things at a somewhat high level so that non-subject matter experts on the search committees could grasp the overarching theme of my research agenda. My teaching statement was two pages long and focused mainly on my past teaching experience as an instructor of record and my teaching philosophies. I did not customize my CV, research statement, or teaching statement for most schools that I applied to. Unlike the cover letters, I sent the same documents to most schools.

My letters of recommendation were from my PhD advisor, my postdoc mentors, and one was from a prominent professor in my field who was both familiar with my work and a research collaborator. To help them write the strongest letters possible, I sent them my CV, my research statement, and my teaching statement.

Skype/Zoom Interviews

As mentioned earlier, I had 8 Skype/Zoom interviews. The phone interviews were usually conducted with 10-20 semi-finalists, from which 3-5 would be selected to interview at campus. The most common questions I was asked were:

• “Why did you apply to our program?”
• “If you were to explain your research to a non-expert, what would you say?”
• “Could you describe your research plans for the next five years?”
• “Are there any research collaborations you could foresee if you came to our school?”
• “Could you describe your past teaching experience?”
• “What course(s) would you like to teach?”

So you really do need to be ready to succinctly describe your research and teaching. Namely, you should be able to articulate that you have a clear agenda that you can sustain for the pre-tenure years. Additionally, you really do need to have researched the program and be familiar with the research of its faculty. Finally, you should definitely go through the course catalog and course descriptions to see which classes offered by the department you could teach.

For these Skype/Zoom interviews, I would just do your best. Try to convey enthusiasm for the specific program and demonstrate that you have thoroughly researched the program and its school. Additionally, I would not try to gouge your chances of being invited for a campus interview from how well your phone interview went. Even if you “nailed” the phone interview, you have no idea how the other semi-finalists did. Other job candidates may have also interviewed very well. Maybe you had been ranked 9th before the interview and a stellar interview pushed you up to 5th on the list… but the department was still only going to invite four people to campus. So you may have fallen just below the threshold. At this point, I think you should pause a bit and be proud that you made it this far. Your application was picked as one of the top 10-20, so obviously, the search committee thought you were well-qualified for the job.

On-Campus Interviews

As mentioned earlier, I did five on-campus interviews. In my field, interviews at research universities usually take place from December through February. All these schools paid for the hotel and the meals, but a few of them asked me to pay for my plane ticket and ground transportation to/from the airport and reimbursed me for these costs later. As long as you send the receipts to the appropriate office administrator promptly after your interview, you should receive a reimbursement check within a matter of weeks. Still, if you have on-campus interviews, I would prepare to go into a bit of short-term debt during the interview season. To deal with this, I got a second credit card right before the interview season.

The interviews themselves typically lasted one or two days. Within a week before the interview date, I would receive a detailed itinerary with the schedule for my visit. In all cases, I flew in the afternoon or evening before the campus interview. If I arrived early enough the day before, I had dinner with a current faculty member. Then the next day, I would be on-campus interviewing from 8 or 9 am to 5 pm. These day-long interviews comprised of a number of one-on-one interviews with individual faculty, the search committee, the dean of the college, and current graduate students. After the day was done, a few faculty members would take me out to dinner at a nice restaurant before dropping me off at the hotel.

Overall, these meetings and meals were not that stressful. Obviously, you should come prepared to these meetings so you have good questions to ask (e.g. about the department, the school, expectations for tenure, etc.) and so you can ensure a nice conversation “flow” (so it may be a good idea to look at faculties’ webpages once you receive an itinerary of who you are meeting with). However, there is no reason to be very tense or nervous. I found that for the most part, people were just trying to get to know you and gouge if you would be a congenial colleague. Rarely was I “grilled” in these one-on-one meetings. Everyone I met with was super nice and friendly. The only time I was “grilled” was during the Q&A of my job talk, which brings me to my next point.

In addition to these one-on-one meetings, I had to give a 45-50 minute research seminar (or “job talk”), followed by a 10-15 minute Q&A session. I was usually given a 30-60 minute break before the actual talk, giving me enough time to “prep” and set up the presentation. I am told that if you apply for a job at a teaching-focused institution, you may also need to give a teaching demonstration as part of your campus interview, but I did not have to do this. For my job talk, I gave a talk on one of my recent research papers, but some people like to talk about two different projects in their seminar talk.

I had started preparing my job talk slides in November and December when I started to get interview invitations. Before my interviews, I practiced giving my job talk in front of three different audiences. Each time, I made changes to my presentation in response to the valuable feedback that I had received.

My job talk was structured so that I opened with about 10 minutes of relevant background information on the topic for non-experts. The next 30 minutes were about my specific contributions to this research area. Here, I still tried to keep it fairly accessible. Finally, I closed with 5-10 minutes of future research directions. This included not just extensions to the specific work in my job talk, but also a few slides on my overall future research agenda and my future plans to obtain grant funding. This was mainly to demonstrate that I could sustain an active research agenda for years beyond my PhD and postdoc research.

The Q&A session after the job talk can be just as important as the talk itself. So in addition to preparing the job talk itself, you also need to be prepared to address potential questions, challenges, and criticisms of your work. During the Q&A, some questions are simply asking for clarification, so you should be able to give cogent responses to such questions. During the Q&A, some faculty will also try to ask “tough” questions to try to press you to think deeper about the topic at hand. In this case, it is best not to be defensive, but to acknowledge these as legitimate issues and discuss how you might be able to address them. Keep in mind that no method or research is “perfect” and everything can always be improved. If an audience member suggests something, be amenable to that as well and thank them for their suggestions.

The day after flying back from my interviews, I sent a thank you note to the chair of the search committee and the administrative assistant who organized the visit. Then I simply waited to hear back. Schools will typically not contact you until all the interviews are finished. If you are lucky enough to get a job offer, then you can spend some period of time negotiating the terms of it. There are some great online resources for negotiation: see here, here, and here. Among the things that you can negotiate on are salary, start-up package (e.g. funds for computer equipment, conference travel, etc.), a one-year delay to your starting date to complete (or finish) a postdoc, and teaching releases (e.g. one semester off from teaching to jump-start your research).