NYU · Biostatistics · New York City
Finding the human story inside the data.
Hi, I’m Tiffany. I study biostatistics at NYU and build careful, approachable ways to understand health data—especially when the answer matters beyond a spreadsheet.
- Focus
- Public health
- Methods
- Causal + spatial
- Tools
- R + Python
Selected projects
Questions worth measuring.
Coursework and independent explorations at the intersection of statistics, public health, and clear communication.
A cooler city, block by block
Mapping summer heat exposure and tree canopy across New York City to find where small green-infrastructure investments could have the biggest public-health impact.
- Research question
- Where do high summer surface temperatures, sparse tree canopy, and indicators of population vulnerability overlap at the census-tract level?
- Analysis
- I joined open geospatial layers in R, standardized exposure indicators, checked spatial autocorrelation, and fit a hierarchical model with borough-level partial pooling.
- Deliverable
- An interactive priority map and a two-page methods memo comparing current conditions with modest tree-canopy improvement scenarios.
Key takeaway A useful priority index should surface compounding exposure—not simply rank neighborhoods by temperature alone.
- R
- sf
- brms
- NYC Open Data
Technical notes
The model combines land-surface temperature, tree-canopy coverage, age, and housing indicators while retaining a separate uncertainty interval for every tract. Posterior predictive checks were used to catch areas where the model fit poorly instead of forcing a single citywide ranking.
Side effects, said clearly
A prototype that turns dense adverse-event tables into patient-readable risk comparisons without flattening the uncertainty behind the numbers.
- Research question
- Can someone distinguish baseline risk from treatment-associated risk without having to interpret a dense statistical table?
- Analysis
- I built a cleaning and validation pipeline, calculated absolute risks and risk differences, and tested several visual encodings for frequency and uncertainty.
- Deliverable
- A responsive prototype using natural frequencies per 1,000 people, small multiples, and plain-language “what we know” notes.
Key takeaway Lead with the absolute difference; use relative risk as context instead of the headline.
- Python
- pandas
- Altair
- FastAPI
Technical notes
The interface keeps denominators consistent across treatment and comparison groups, flags missing follow-up information, and pairs each estimate with its confidence interval. A reusable data schema separates the statistical calculations from the language shown to readers.
Who gets counted?
Auditing demographic reporting in metabolic-disease trials and comparing enrolled cohorts with the New York communities most affected by those conditions.
- Research question
- How consistently do metabolic-disease trials report race, ethnicity, age, and sex—and how well do recruitment locations reflect affected communities?
- Analysis plan
- I pull public registry records through an API, normalize reporting categories, score completeness separately from representation, and log every missing field.
- Current output
- A reproducible Quarto audit with a searchable study table, data dictionary, recruitment-site map, and explicit limitations statement.
Current status Data validation and missingness sensitivity checks are underway before any representation comparisons are interpreted.
- ClinicalTrials.gov
- Python
- Quarto
Technical notes
Completeness and representativeness are deliberately treated as different questions: a study can report demographic data clearly and still enroll an unrepresentative cohort. The pipeline preserves original registry values alongside standardized categories so the transformation remains auditable.
Current coursework
The toolkit behind the questions.
Classes shaping how I design studies, work with health data, and communicate results responsibly.
Biostatistical Methods
Regression, inference, model diagnostics, and translating estimates into useful health conclusions.
Epidemiologic Study Design
Building stronger observational studies by thinking carefully about bias, confounding, and measurement.
Statistical Computing
Reproducible analysis in R and Python, from cleaning untidy data to presenting transparent results.
How I approach the work
Careful in the method. Clear in the telling.
-
01
Ask a better question
A precise method cannot rescue a vague or irrelevant question. I start with who the answer is for.
-
02
Show the uncertainty
Assumptions, missingness, and confidence belong in the story—not hidden at the bottom of a notebook.
-
03
Explain it plainly
If a result matters, the people it affects should not need a statistics degree to understand it.