Portfolio

• National scope: Supported a large network of sites across the country — states, major cities, and territories — representing the great majority of U.S. births. Coordinated intake, reviews, and releases so each site moved smoothly from start to finish.
• Season planning and coordination: Set the calendar, intake checklist, and checkpoints. Ran standing touchpoints, unblocked issues quickly, and kept everyone aligned on what was needed and when.
• Data intake and validation: Converted plain-text “birth” and “frame” files into structured datasets using a standard template. Verified counts and identifiers, checked required fields, and documented any quirks before analysis so there were no surprises later.
• Sampling verification: Reviewed how each site drew its sample, compared expected and observed patterns, and flagged anything unusual for follow-up. Kept a clear record of findings and fixes so the next run started cleaner.
• Fair results when some people do not respond: Compared who answered with everyone who was eligible, grouped people with similar chances of responding, and adjusted their influence so results reflected the whole population, not just those who replied.
• Coverage checks: Compared important totals to official counts and looked for signs that certain groups might be missing. Built in stop-points so a run would pause if something needed attention.
• One-click packaging: Automated the creation of consistent tables, figures, and notes for reviewers. Produced tidy, share-ready documents and a short summary of what was checked and what changed.
• Lookup-driven, repeatable runs: Used a single “lookup” table to hold the site-specific settings. The system read that row and applied the right rules automatically, which reduced manual edits and made results consistent across sites and seasons.
• Partner support and problem solving: When a file did not match the plan, worked directly with the site to identify the cause, agree on a correction, and record the exact steps taken so the fix did not have to be rediscovered.
• Training and handoff: Led short, practical sessions on the process; shared simple checklists and examples; and kept a clear trail of decisions so teammates could reproduce the full run.
• Outcomes: On-time releases across many jurisdictions, fewer back-and-forth cycles, faster reviews, and clearer paperwork for later audits — with a season playbook that the team can reuse and improve.

• Regular mentorship: Met twice per month over the six-to-nine-month program, setting goals, reviewing progress, and helping mentees plan analysis steps that matched their study design.
• Regression & modeling guidance: Supported logistic and generalized linear models; explained model selection (AIC, BIC, ROC/AUC), interaction terms, confounding control, and interpretation of coefficients and effect sizes.
• Advanced methods: Taught when and how to use Lasso and Ridge regression, cross-validation, decision trees, and random forests—highlighting tradeoffs between prediction, interpretability, and sample size.
• Latent & structural models: Guided structural equation modeling and latent-construct approaches for projects involving behavioral, mental-health, and disclosure outcomes.
• Exploratory analysis: Helped researchers evaluate missingness, build correlation matrices, define variables, and create reproducible, well-documented data pipelines.
• Interpretation & communication: Worked with mentees to explain results clearly—risk ratios, odds ratios, post-hoc tests, diagnostics—and craft manuscripts and presentations suitable for scientific review and clearance.
• Impact: Mentees advanced their studies with stronger analysis plans, clearer statistical reasoning, cleaner pipelines, and ready-to-share results; several projects moved from early concept to polished products.

• Reusable tools: Templates for common tables with options for how to handle missing values and direct export to spreadsheets.
• Season enablement: Trainings on pre-weighting checks and the weighting workflow, plus step-by-step notes and checklists.
• Maintenance: Kept site-specific variable lists and format catalogs up to date so runs were smooth and consistent.
• Impact: Faster reviews, fewer errors, and easier handoffs across sites.

• Sampling change reviews: Verified proposed changes and documented decisions so the next season started cleaner.
• Data-quality triage: Investigated miscoding or missing fields and wrote clear next steps for sites.
• Accessible documentation: Modernized codebooks to be easier to read and use; standardized filenames and formats.
• Team operations: Helped plan working sessions and retreats; kept cross-team priorities moving.
• Safeguards in analysis: Protected small cells, labeled outputs clearly, and exported “ready-to-share” tables.

• Analytical review: Read and discussed the team’s nonresponse-bias analyses (led by colleagues), offering technical feedback and clarifying where conclusions were well-supported.
• Interpretation support: Helped frame findings in clear, plain-language terms so program leadership could weigh risks and understand practical implications.
• Internal guidance shift: Contributed feedback as the team moved toward focusing on transparent evaluation of potential bias rather than relying on a single response-rate cutoff.
• Outcome: The revised approach was accepted through federal review, supporting a transition away from a fixed threshold and toward routine bias assessment.

• What changed: partnerships and scheduling across time zones that work for families, schools, and mentors; cleaner operations (simple sign-ups, dependable communications, better instructor onboarding); clear roles, timelines, and checklists so the program scales without confusion.

• How I run it: audience of ~3,000+ contacts across universities and schools (kept fresh and organized); plain-spoken messages tailored for graduate departments, faculty, and educators; measured each send (opens, clicks, replies) and adjusted timing and subject lines; negotiated 100,000 high-school contacts with strong deliverability terms.
• Sample results: recent campaigns show strong engagement (high opens and meaningful clicks); faster mentor recruiting and a larger, more diverse applicant pool.

• Highlights: One clear “front door” with a single row of buttons; reusable pieces (hero, buttons, stat chips, cards, accordions) so updates are quick and consistent; mobile-friendly with strong keyboard focus and reduced-motion support.
• Impact: Fewer “where do I find…?” emails and faster onboarding for new mentors and families; content is easier to maintain—no heavy framework or plugins to fight.

• What I built: one combined data set with duplicates removed and common fields; simple charts that answer practical questions—where interest is growing, what support is needed, and which messages work.
• Why it matters: better decisions about scholarships, mentor staffing, and where to advertise; less time cleaning data and more time supporting students and mentors.

• What changed: reusable sections for common items (consent, disclosures) that we edit once and reuse; guided questions to reduce back-and-forth and cut down on typos; cleaner exports for reviewers—names line up, fields match, and sorting is straightforward.
• Impact: publish a new cycle in minutes, not hours; less manual cleanup and quicker hiring decisions.

• Data & workflow design: Shaping the process for ingesting user uploads, merging external economic indicators, and creating a clean, analysis-ready dataset.
• Indicator & metric framework: Defining the core signals (trend, risk, volatility, comparisons) and establishing consistent, reproducible logic behind each metric.
• Forecasting architecture: Evaluating and selecting simple, interpretable forecasting approaches suitable for a first-generation prototype.
• Power BI model groundwork: Outlining the schema, relationships, calendar structure, and measure logic that will support automated visuals and summaries.
• Product scope & tiers: Working with the founder to shape the initial offering, use cases, and roadmap without over-promising beyond what early tooling can reliably support.
• Documentation & clarity: Maintaining clear notes, plain-language definitions, and rationale behind design choices so future team members can extend the system safely.

• Data prep & cleaning: standardized fields, converted checkbox/“checked vs. unchecked” values into analysis categories, and aligned site/region mappings.
• Reporting in SAS: reliable routines for totals and subgroup breakouts; SAS/GRAPH visuals for the survey’s core questions and demographics; exports suitable for sharing.
• Deliverables: two reports (overall totals + demographic subgroups), plus CSV/Excel files for handoff and archival.
• Public, privacy-safe example: one modernized SAS program, a synthetic dataset, and a sample PDF so others can see the process without any PHI/PII.

Links: 🔗 View the repository · 📄 See the sample PDF · 📊 Download the synthetic CSV · 🧩 Open the SAS program

Licensing: MIT for code; CC BY 4.0 for reports/docs. Privacy: synthetic data only (no PHI/PII).