The full list, with the honest stage of each one. Drafts sit for months. Data collection runs longer than planned. Three of these have been through more than one round of review. I keep this list current rather than showing only the finished work, because the stage is what decides whether a student can join at all.
Twelve of the fifteen are closed to students. Eleven because they are already written — drafted, submitted, or in revision — and joining a finished paper means running errands, not doing research, with no honest path to authorship. One is led by a colleague, so it is not mine to offer. The three that are open are open because they are genuinely unfinished.
| Project | Stage | Students |
|---|---|---|
| Enhancing Efficiency and Workflow in Oncology Outpatient Services by Simulation-Based Optimization | Forthcoming, Annals of Operations Research Accepted | Closed Written |
| Balancing Workforce Fissuring and Service Quality | Second round of revision, Journal of Operations Management; resubmitted Sep 2025 R&R 2 | Closed Written |
| Exploring the Triple Aim Through Clinician Empowerment | First round of revision, Journal of Operations Management; revision submitted R&R 1 | Closed Written |
| Enhancing Healthcare Operation in Disadvantaged Communities | Under review, Health Care Management Science; submitted Feb 2025 Under review | Closed Written |
| The Misplaced Intelligence: AI Position and Output in Sequential Care Lines | Under review, Journal of Operations Management; submitted Jul 2026 Under review | Closed Written |
| Engagement Without Responsibility: Ownership Transitions and Stakeholder Welfare in U.S. Nursing Homes | Preparing for submission Draft | Closed Written |
| Private Equity in Healthcare: Ownership Transitions in Global South Health Systems | Draft complete, awaiting a special issue deadline Draft | Closed Written |
| Chain-Level Quality Competition Under Yardstick Incentives (U.S. dialysis) | Draft complete Draft | Closed Written |
| When Stars Fall: Quality Instability, Consumer Choice, and the Demand for Long-Term Care | Draft complete Draft | Closed Written |
| When the Scorecard Changes: Quality Measurement Redesign and Resource Reallocation | Draft complete Draft | Closed Written |
| From Triple Aim to Quadruple Aim | Draft nearing completion Draft | Closed Written |
| Extension Services, Market Remoteness, and Agricultural Productivity in Ethiopia | In progress, targeting a mid-2026 deadline In progress | Open Track 3 |
| Deep Reinforcement Learning for Forward-Looking Chemotherapy Appointment Scheduling | In progress, targeting a mid-2026 deadline In progress | Open Track 2 |
| Do Cross-Level Staffing Patterns Influence Clinician Turnover and Care Operations | Data collection and early analysis Data | Open Track 1 |
| Improving Emergency Department Flow | IRB approved; data collection in progress Data | Closed Colleague-led |
A stage is not a promise. Papers get rejected, drafts get restructured, and collection timelines slip. A project that closes to students as it gets written up is the normal path, not a door slammed — and one that reopens is possible too, if a rejection sends it back to the data.
Ten research streams, and the teaching material each one produced. Everything in the right-hand column is already open on this site — you can run it before you ever contact me.
| Research stream | What it teaches once it reaches a classroom | Open artifact |
|---|---|---|
| Oncology outpatient flow Forthcoming · Annals of OR |
Why an infusion center with spare capacity still makes patients wait: variability, not average load, produces queues. Students tune arrival and service distributions and watch the waiting time move nonlinearly. | Infusion center model · M/M/c simulator |
| Emergency department congestion Data collection · IRB approved |
Little's Law as a diagnostic instrument rather than a formula. Where boarding delay actually originates, and why adding beds sometimes does nothing at all. | Queueing simulator · Port congestion analogue |
| Workforce fissuring & service quality R&R · Journal of Operations Management |
Labor treated as a capacity decision with a quality price tag. Contract versus permanent staffing, and what the substitution costs downstream. | Demand Planning & Fulfillment · Causal Inference Handbook |
| Ownership transitions & private equity Working papers |
Who absorbs the cost of a capital-structure decision. Stakeholder analysis with an actual balance sheet attached instead of a values discussion. | Supply Chain Finance course · Trade credit NPV |
| Quality scorecards & star ratings Working papers |
Measurement systems as operating decisions: what a rating redesign does to where managers put staff, and how a metric quietly rewrites behavior. | Data-Driven Analysis Handbook · Six Sigma measurement module |
| Clinician empowerment & the Quadruple Aim R&R · Journal of Operations Management |
Service operations where the server is a person with discretion, fatigue, and a resignation option. Empowerment as a scarce operational resource. | Service operations module · construct measurement notes |
| Care in disadvantaged communities Under review · Health Care Mgmt Science |
Access as an operations outcome. Who is inside the service region and who is merely near it, and what the distinction does to measured performance. | Facility location simulator · ML course segmentation labs |
| Extension services & smallholder productivity Working paper · M&SOM SI |
Operations outside the firm: market remoteness, information access, and productivity frontiers on farms with no ERP system and no inventory software. | Routing simulator · Facility location |
| AI position in sequential care lines Under review · JOM |
Where a tool sits in a process decides what it is worth. Putting AI at the front of a line raises what arrives; it raises output only if the binding constraint is there too. A clean bottleneck lesson with a real adoption dataset behind it. | Queueing simulator · Bottleneck strategies · ML course |
| Deep RL for chemotherapy scheduling Working paper · POM SI |
Sequential decisions under uncertainty, and the precise reason a sensible myopic rule loses to a forward-looking policy over a full schedule horizon. | ML course · Newsvendor game |
Three ways to enter, listed with the hours they actually take. The estimates are deliberately low — a semester with two midterms does not have ten spare hours in it, and a commitment you cannot keep helps neither of us. Past four hours a week, co-authorship is on the table; at six to eight, with a question you brought yourself, so is shared first authorship. Treat the hours as a planning guide rather than a rule — credit follows what you actually contributed, not the level you signed up at. Every level fits inside a single term — nothing here asks you to sign up for a year — and each ends in something finished, so stopping after Level 1 is a normal ending rather than a dropout.
Explorer
- Read one paper from the track and one paper it argues with
- Run the matching simulator or open the public dataset
- Write one page: what you would test next, and what would change your mind
Contributor
- Own a defined, checkable task: cleaning, a codebook, a literature table, one constructed measure
- Keep a reproducible script that runs top to bottom
- Take on an analytical piece of your own if the first task goes well — a robustness table, a scenario set, a mapping module
- Biweekly 30-minute check-in with a written agenda
- If what you actually contribute reaches the co-authorship bar, you get it — entering here does not cap you
Co-author
- Four hours a week across a term is the co-authorship threshold — a semester, not a career
- Contribute to design, data, or analysis, and write part of the paper
- At six to eight hours, with a question you brought yourself, we share the first-author position
- Respond to referee comments alongside me
Most of the projects listed above are closed to student participation, and the reason is the same reason the authorship terms below are worth anything: you cannot earn a place on a paper that is already written. Anything drafted, submitted, or in revision is finished work — a student joining it would be doing errands, not research. Two more are led by a colleague, so they are not mine to staff. What is left is the work that is genuinely unfinished, where a student's judgment still changes the result.
Staffing Patterns and Why Clinicians Leave
What you would actually do
- Assemble panel datasets from CMS Payroll-Based Journal files — millions of daily records, merged across quarters and facility identifiers that do not always agree
- Construct the core measures: agency staffing share, turnover, hours per resident day, staffing volatility
- Run the descriptive work first, then fixed-effects specifications, and learn why the clustering choice changes the answer
- Write the codebook — the document that lets a stranger rebuild your dataset without asking you a single question
Prerequisites before you start
- Required: one regression course. You should be able to interpret a coefficient, a standard error, and a dummy variable interaction without looking them up
- Required: Stata or R at a working level — or a serious start on the Stata guide or R guide on this site
- Strongly recommended: read the fixed effects and difference-in-differences chapters of the Causal Inference Handbook before our first meeting
- Not required: healthcare knowledge, labor economics coursework, prior panel data experience
Does this cross your interests?
Where it leads afterward
- Jobs: economic consulting (Analysis Group, Cornerstone, Charles River), people and workforce analytics, policy research organizations such as Mathematica, RAND, or Urban, and healthcare strategy groups
- Research: the most direct preparation on this page for OM, applied economics, or health policy doctoral study — panel data fluency is the price of admission
- Portfolio: a public replication package on GitHub that an interviewer can open and read
Scheduling Chemotherapy When Today's Choice Constrains Tomorrow
What you would actually do
- Build the simulation environment: arrivals, chair and nurse constraints, regimen durations, no-shows
- Implement the baselines honestly — a myopic rule tuned as well as you can tune it is the comparison that matters
- Run and diagnose training: reward shaping, and finding out why a policy that looks good on average is terrible on Mondays
- Turn one scenario set into a classroom-ready demo students in my analytics course can actually use
Prerequisites before you start
- Required: Python beyond the basics — you write functions and classes, not just scripts
- Required: one introductory statistics course; you should know what a distribution is and why an exponential tail differs from a normal one
- Helpful: any exposure to reinforcement learning, or willingness to work through the RL sections of the ML course; try the M/M/c simulator and see whether the behavior surprises you
- Not required: healthcare background, prior simulation coursework, deep learning experience
Does this cross your interests?
Where it leads afterward
- Jobs: hospital operations and capacity planning, industrial engineering inside health systems, healthcare consulting, and any scheduling-heavy operations role — airlines, logistics networks, cloud capacity
- Research: the standard entry path into OR/OM doctoral programs; simulation-optimization, stochastic scheduling, sequential decision-making under uncertainty
- Transferable well beyond healthcare: every appointment, dock, runway, and call center is the same mathematics with different vocabulary
Operations Without a Firm: Extension Services and Smallholder Productivity
What you would actually do
- Work with World Bank LSMS-ISA household microdata — multi-wave, multi-module, with household identifiers that must be tracked across rounds
- Construct market access and remoteness measures from geographic and infrastructure variables
- Estimate production functions and efficiency scores using stochastic frontier analysis or DEA, and interpret what a frontier actually claims
- Handle survey weights correctly — the difference between a descriptive statistic that is right and one that is quietly wrong
Prerequisites before you start
- Required: econometrics through multiple regression, and Stata at a working level
- Required: tolerance for documentation. Survey microdata comes with hundreds of pages of manuals, and the answers are in there
- Helpful: any development economics or global health coursework, or lived familiarity with an agricultural economy
- Not required: frontier analysis experience — SFA and DEA are learnable inside this project
Does this cross your interests?
Where it leads afterward
- Jobs: World Bank, IFC, and development finance analytics, international NGO monitoring and evaluation, development consulting, agricultural supply chain and ag-tech roles
- Research: development economics, agricultural economics, and the growing socially-responsible operations field — where M&SOM and POM now run dedicated special issues
- Also: operations methods combined with development microdata is an uncommon pairing, and it shows on an application
Closing the Loop: Turn a Finished Finding Into the Next Teaching Artifact
What you would actually do
- Take one finding from the finished work above and identify the single mechanism a student must feel to understand it
- Design the interaction: which parameters are exposed, what the default case shows, where the surprise happens
- Build it as a self-contained web simulator or notebook, in the visual language of the existing set
- Test it on real students, watch where they get confused, and rebuild the part that failed
Prerequisites before you start
- Required: ability to explain something clearly. This is the actual prerequisite and it is not negotiable
- Required: HTML and JavaScript, or Python with a notebook interface — intermediate is plenty; existing simulators are readable templates
- Helpful: having tutored, TA'd, or taught anything at all, including a club workshop
- Not required: research experience, statistics beyond the finding you are translating, design training
Does this cross your interests?
Where it leads afterward
- Jobs: educational technology, corporate learning and development, technical writing, data journalism, developer relations, product roles where explaining is the product
- Research: pedagogy and simulation-based learning are publishable in their own right; teaching-focused academic careers start with exactly this portfolio
- Also: a public, linkable artifact used by an actual class — something concrete to point an employer at
This part runs whether or not you ever email me. Everything it asks for is already open on this site, and it is roughly the same ground anyone joining a track has to cover anyway. Five weeks at one to two hours each. If you finish week five, send me the output — a one-page proposal tells me far more than a transcript does.
Read one paper properly
Pick a track, read its lead paper, then read one paper it argues with. Do not skim the methods — that is where the argument lives.
Get the tool working
Install and use whatever your track requires — Stata, R, Python, or a simulator. Use the guides on this site; they were written for exactly this week.
Touch the real data
Download the public dataset your track uses. Load it, count rows, find the missing values, and locate the first thing that does not make sense.
Reproduce something
Rebuild a published descriptive statistic or figure from that data. It will not match on your first attempt. Finding out why is the entire exercise.
Propose one extension
Name a question the paper left open that your data could actually address. Be specific about the comparison you would make and what would falsify you.
These are my own teaching materials, written for students in exactly this position. No enrollment, no login, no cost. Work through the ones your track names in its prerequisites.
Python for Data Science
Tracks 2 and 4 · start with chapters 1–5
Stata for Econometrics
Tracks 1 and 3 · panel data and IV chapters
R for Statistical Computing
Track 1 · tidyverse and modeling
SQL for Data Analytics
Any track once a dataset outgrows memory
Causal Inference Handbook
Track 1 · read before our first meeting
Data-Driven Analysis
All tracks · EDA and feature construction
Machine Learning Course
Track 2 · eleven modules, including the RL sections
Supply Chain Finance
Background reading · working capital and ownership
AI-Assisted Coding
All tracks · use it well, and disclose it
Research relationships go wrong when expectations stay implicit. These are mine, written down before you start rather than negotiated after the work is done.
Authorship, including shared first authorship
Co-authorship starts at roughly four hours a week, over about three months. Contribute to design, data, or analysis, write part of the paper, and be able to defend it — that is the whole bar. It is lower than students usually assume, and it is meant to be reachable within a single term by an undergraduate who shows up consistently, rather than only by someone who already knows how to do research.
Shared first authorship is genuinely available, at six to eight hours a week. If the question is yours — you brought it, you framed it, you drove it — we share the first-author position, with the standard footnote stating that both authors contributed equally. That is an established convention, not a courtesy title, and it is read as one. At that level the work does not have to stay inside my agenda: we start from what you are curious about and find the version of it the data can actually answer. I would rather supervise a question you care about than hand you one I do not have time for myself.
The level you entered at does not cap your credit. Hours are a planning guide, not the rule. A contributor who came in at Level 2 for a data task, and whose actual input ends up reaching the co-authorship bar, is a co-author — whatever the plan said in September. The call is made on what you contributed, judged the same way for everyone, and I would rather err toward including someone than toward a defensible-sounding exclusion.
Author order is discussed before the work starts, not after it is done. Contributions below the co-authorship bar are acknowledged by name in the paper, which is a real and citable credit. No one's work gets used without attribution.
Data rules are absolute, not aspirational
- Restricted data never leaves approved storage. No personal laptops, no cloud drives, no exceptions
- Human-subjects projects require completed CITI training before any access
- Public data is the default entry point precisely so you can start while clearances proceed
- If you are unsure whether something is permitted, ask first. Asking is never the mistake
Reproducibility is the deliverable
Every task ends in a script that runs from raw data to output without manual intervention. No undocumented spreadsheet edits. If you cannot rerun it next month, it is not finished. This is the habit that separates people who can be trusted with an analysis from people who cannot, and I would rather you learn it here than in your first job.
AI tools are allowed, and disclosed
Use them — I write about teaching with them. Two conditions: you disclose where they were used, and you can defend every line as if you wrote it, because in the ways that matter you did. Code you cannot explain is code we cannot ship. An AI-drafted paragraph you have not verified is a retraction waiting to happen.
Ways to get credit for this
- Independent study — arranged with your department, graded on the deliverables above
- Honors thesis or capstone — if your program runs one, a Level 3 term can seed it or sit inside it
- Paid research assistantship — when grant funding allows; I will tell you plainly whether it does
- Portfolio only — entirely legitimate, and what most Level 1 and 2 students choose
What I commit to
- A response to any email that names a specific track
- A scheduled check-in you can count on, with written feedback on what you submit
- An honest reference letter with specifics in it — or an early, direct conversation if I cannot write a strong one
- Telling you when a direction is not working, quickly, rather than letting you spend a semester on it
How to Apply
One email, under two hundred words. No CV required, no formal application, no deadline.
- Name the track by number. "Track 1" tells me more than three paragraphs about your interest in healthcare.
- State your prerequisites honestly — which you meet, which you do not. A missing prerequisite is a plan, not a disqualification. Overstating one always surfaces in week two.
- Two sentences on why the question interests you. Your own words. A generic paragraph reads as generic.
- What you want to be able to do by the end of the term. This determines which level and which task I would put you on.
- If you already have your own question, lead with that instead. A question you brought is the route to shared first authorship, and it does not have to match my current list — only be close enough that I can supervise it usefully.
- Hours per week and which term you are describing.
- Optional but decisive: attach anything you have built — a course project, a script, a week-five proposal from the onboarding sequence above.
Realistic expectations
Openings depend on the term and on where each project sits — a paper in revision needs different help than one in data collection. If nothing fits when you write, I will say so directly and tell you when to check back, rather than leaving you waiting on a maybe.
Public-data tracks (3, 4, 5, 7, 8) rarely have a queue, because the work scales with however many careful people are doing it.
Not a Rutgers student? Public-data tracks work remotely. Clinical and IRB tracks require affiliation and institutional training.
I have never coded. Is there anything here for me?
Not right now, and I would rather say so than pretend otherwise. All four open tracks need either Stata/R or Python — that is a consequence of which projects happen to be unfinished this term, not a policy. What still works: the five-week onboarding sequence above takes you from nothing to a running script using free materials already on this site, and it costs you nothing but time. Do weeks one through three, then write to me — at that point you meet the prerequisites for Track 1 or Track 4. If a project reopens at the data stage, spreadsheet-level entry comes back with it, and this page will say so.
Do I need to be a supply chain major?
No. These tracks are written to be legible to students from economics, statistics, public health, computer science, nursing, psychology, and business generally. What matters is whether the question holds your attention when the work gets tedious — because it will get tedious, in every track, at some point.
Will I get paid?
Sometimes, when grant funding covers it, and I will tell you plainly whether it does at the moment you ask rather than leaving it vague. Most Level 1 and 2 positions are for credit or for the portfolio. I would rather you know that before you invest a semester than discover it afterward.
What if I start and discover I hate it?
Then Level 1 did its job. Three or four weeks is deliberately short so that leaving costs you almost nothing and teaches you something real about your own preferences. Tell me directly, and I will still write about what you did. The outcome I want to avoid is a student staying six months out of obligation.
Will this help my PhD or medical school application?
It helps in the ways admissions committees can actually verify: a letter that describes specific work rather than general enthusiasm, a writing sample, a reproducible analysis they can look at, and, from Level 3, a conference presentation. It does not substitute for grades or test scores, and I will not claim otherwise. What it reliably provides is evidence that you have done research rather than only taken courses.
How is this different from a class project?
There is no answer key, and no guarantee the question has an answer. Real data arrives broken. A measure you spend three weeks building sometimes has to be discarded. Findings come out null, and a null result still has to be written up honestly. Nothing about the work is arranged in advance to come out well.
Can I propose my own question instead?
Yes, and that is the route to shared first authorship rather than a detour around the tracks. The only real constraint is that I have to be able to supervise it competently — close enough to healthcare operations, workforce, measurement, or service systems that my feedback is worth something to you. Within that, we work from your interest rather than my list. The practical path is entering through a track, learning the setting and the data, then proposing your extension at week five of onboarding or at the end of Level 2; questions proposed from inside the data tend to be sharper than questions proposed from outside it. But if you already have the question when you write to me, lead with it.
How much time does this really take?
Everything here is scoped to a single term — three months at the outside, and Level 1 is three or four weeks. Nothing asks you to commit to a year. The hours are set low on purpose but unevenly distributed: some weeks are quiet, the week before a submission deadline is not. Tell me your exam schedule in advance and I will plan around it. If you are unsure between two levels, take the lower one; moving up later costs nothing.