Prepare and connect data
Joins and cohort assembly, reshaping repeated observations, reusable mapping functions, field cleaning, and categorical recoding.
STATISTICS / ENGINEERING / ANALYTICS
A toolkit with
a purpose.
The tools matter because of the decisions they make possible. Follow the portfolio through a role, then explore how methods and libraries connect across the analytical workflow.
FIND YOUR READING PATH
Five roles.
A connected foundation.
Each path highlights a different part of the same analytical process and takes you directly to the supporting case studies.
A READING PATH FOR BIOSTATISTICIAN ROLES
Make the research question measurable.
A biological or behavioral question becomes an endpoint, an observation window, and an analysis population. My work connects those definitions to survival models, repeated measurements, and study planning.
survival · lme4 · lmerTest · TrialSize · effectsize
See all related workstreams →010203
Start with the endpoint
Recovery analysis turns repeated cognitive measurements into event times while retaining incomplete follow-up.
Explore this part of the story →Respect the study design
Power calculations and mixed-model simulations explore pairing, participant variation, and missingness before recruitment.
Explore this part of the story →Explain uncertainty
Rest-scheme comparisons keep subgroup counts, confidence intervals, and observational limits beside the estimates.
Explore this part of the story →A READING PATH FOR STATISTICIAN ROLES
Connect assumptions to conclusions.
The important choices happen before and after model fitting: defining the estimand, choosing a distribution, checking dependence, and interpreting uncertainty. These projects make that reasoning visible.
survival · lme4 · glmmTMB · car · DHARMa · equivalence
See all related workstreams →010203
Choose the right model
Cox, linear mixed-effects, and exploratory Poisson formulations answer different versions of a recovery question.
Explore this part of the story →Evaluate the design
Analytical non-inferiority calculations and Monte Carlo experiments test the consequences of study assumptions.
Explore this part of the story →Make comparisons interpretable
Effect sizes and interval estimates communicate magnitude and precision alongside group differences.
Explore this part of the story →A READING PATH FOR DATA ANALYST ROLES
Turn records into an answer people can use.
A useful analysis starts with a clear denominator and ends with an understandable output. My work combines cohort preparation, visual exploration, and repeatable reporting so collaborators can follow the evidence.
dplyr · ggplot2 · Plotly · Shiny · DT · openxlsx
See all related workstreams →010203
Define what is being counted
Reporting distinguishes unique people, repeat trips, and usable outcomes before creating monthly summaries.
Explore this part of the story →Make exploration accessible
Shiny filters connect selected cohorts to interactive charts, tables, and participant counts.
Explore this part of the story →Explain differences visually
Timepoint plots and subgroup comparisons show when patterns change and where the data become sparse.
Explore this part of the story →A READING PATH FOR DATA ENGINEER ROLES
Build the reliable dataset behind the analysis.
Research data arrives as heterogeneous exports, repeated records, and imperfect timestamps. The engineering story is how those inputs become keyed, comparable, validated datasets and reproducible outputs.
dplyr · tidyr · purrr · lubridate · stringr · readxl · openxlsx
See all related workstreams →010203
Resolve identity and time
Wearable exports, logs, and flight events are aligned to composite trip keys and shared time windows.
Explore this part of the story →Make transformations checkable
Rule-based screening preserves an audit trail and reconciles saved outputs against source records.
Explore this part of the story →Deliver repeatable outputs
Batch reporting applies completion rules and aggregates route-by-month metrics into structured workbooks.
Explore this part of the story →A READING PATH FOR DATA SCIENTIST ROLES
Connect feature engineering, models, and decisions.
My strongest data-science work spans the full analytical chain: transforming longitudinal data, engineering meaningful features, evaluating statistical models, and building interfaces for interpretation.
R · survival · lme4 · DHARMa · purrr · ggplot2 · Shiny
See all related workstreams →010203
Engineer a meaningful signal
Sleep-window features encode temporal context while making allocation assumptions explicit.
Explore this part of the story →Model and diagnose
Recovery models connect covariates to an event-time outcome, with proportional-hazards and residual checks.
Explore this part of the story →Stress-test before acting
Synthetic repeated-measures studies examine the behavior of model-based decisions under design assumptions.
Explore this part of the story →APPLIED THROUGHOUT THE RESEARCH WORK
Libraries connected
to real analytical tasks.
R is the core programming language in this research collection. These tools appear in the implemented or shared workflows, with links to the work they support.
Resolve time and deliver files
Timestamp normalization, interval calculations, workbook imports, structured exports, and audit sheets. The xlsx workflow is part of shared/legacy reporting.
Model biological and behavioral outcomes
Cox event-time models; linear and logistic mixed models; exploratory Gamma mixed models. nlme appears in the shared layover workflow.
Challenge the model
Collinearity, simulated residual behavior, dispersion, and participant-level influence diagnostics in modeling and exploratory analyses.
See these tools in context
Design studies and quantify differences
Non-inferiority sample-size calculations, equivalence comparisons, standardized effect sizes, and descriptive summaries.
Build interactive views
Statistical graphics, route panels, sleep timelines, interactive plots, reactive cohort filters, and explorable tables.
Make analysis reproducible and readable
Literate analysis, formatted tables, programmatic reports, and route-wise processing. flextable and parallel also appear in shared layover code.
WHERE THE STORY COULD GO NEXT
Transferable methods.
New questions.
These are proposed project directions, not completed projects. Each connects an existing strength to a new domain or toolchain and makes the intended work and deliverables concrete.
DESIGN CONCEPT / 01
Clinical outcomes & follow-up
Extend the recovery workflow to a public or synthetic longitudinal health dataset.
Explore the proposed project
How the story would unfold
Define an endpoint and follow-up window; build a cohort flow; compare survival estimates and adjusted associations; report missingness and model diagnostics.
Proposed deliverables
An analysis plan, reproducible cohort pipeline, survival figures, and a methods report.
Technology direction
R · survival · dplyr · ggplot2 · R Markdown
The connection to existing work
Endpoint definition, censoring, longitudinal features, and diagnostic reasoning.
Explore the foundation →DESIGN CONCEPT / 02
A testable analytics data pipeline
Translate research-file integration into a small, documented analytical data model.
Explore the proposed project
How the story would unfold
Ingest public or synthetic source files, define stable keys, separate staging from analysis tables, and test uniqueness, required values, and relationships before publishing metrics.
Proposed deliverables
A documented SQL model, data dictionary, quality checks, and a repeatable metrics output.
Technology direction
Proposed stack: Python · pandas · SQL · DuckDB · dbt
The connection to existing work
Composite keys, temporal alignment, deterministic transformations, and reconciliation.
Explore the foundation →DESIGN CONCEPT / 03
Predictive modeling with honest validation
Extend feature-engineering and diagnostics experience into a prediction-focused project.
Explore the proposed project
How the story would unfold
Use a public tabular dataset, define a realistic prediction point, keep preprocessing inside validation folds, compare a baseline with candidate models, and examine calibration and subgroup error.
Proposed deliverables
A reproducible training pipeline, held-out evaluation, error analysis, and a model card.
Technology direction
Proposed stack: Python · pandas · NumPy · scikit-learn · matplotlib
The connection to existing work
Temporal feature definitions, missingness checks, model assumptions, and uncertainty.
Explore the foundation →DESIGN CONCEPT / 04
Experiment & survey decision support
Connect study planning with an explorable analysis of an experiment or survey.
Explore the proposed project
How the story would unfold
Specify the comparison and unit of randomization or observation, run sensitivity scenarios, visualize effect sizes and intervals, and publish denominator-aware summaries.
Proposed deliverables
A design notebook, scenario explorer, comparative figures, and a plain-language decision brief.
Technology direction
R · TrialSize · effectsize · Shiny · Plotly; potential Python extension: SciPy · statsmodels
The connection to existing work
Power planning, repeated-measures simulation, survey transformations, and reactive reporting.
Explore the foundation →Technical references for the proposed implementations: scikit-learn pipelines · dbt data tests.