SAND longitudinal data infrastructure

A six-wave residence-hall study of alcohol use, perceived norms, and social-network change among 255 first-year students. This page covers the project scope and what I built. The full technical reference lives at sand.shangshanggu.com.

Build scope: study design, participant operations, REDCap/MySQL setup, de-identified linkage, QA checks, chapter datasets, proxy data, and public-release notes.

This page excludes: raw records, real identifiers, contact fields, free text, and real network ties.

Full technical reference Source code Network visualisations

375 invited residents 255 enrolled students 87% retained to Wave 6 Six waves, Sep 2022 to Oct 2023 REDCap, MySQL, R, Make Proxy/schema preview

Build Scope

The study required the machinery around the analysis: recruitment, repeated survey delivery, identity separation, network-array preparation, and a reproducible pipeline that can run without exposing restricted data.

Participant operations

Recruited and onboarded 255 of 375 invited residents, then managed reminders, retention, incentives, and participant-facing support across six waves.

Survey and database system

Maintained REDCap/MySQL infrastructure with study-specific survey links, email validation, anonymous enquiries, referral-code logic, and automated incentive workflows.

Confidential linkage

Separated contact details from behavioural responses and converted named important-peer nominations into pseudonymous IDs for analysis.

Reproducible analysis

Built R/Make pipelines from raw exports to QA reports, longitudinal datasets, network arrays, model inputs, tables, figures, and provenance logs.

Network modelling prep

Prepared aligned behaviour arrays and directed adjacency matrices for network autocorrelation models and stochastic actor-oriented models.

Public-release tooling

Added proxy-data mode, generated dictionaries, privacy checks, checksums, validation notes, and this public study-book preview.

Timeline of the SAND study across six waves from September 2022 to October 2023 — Six survey waves from September 2022 to October 2023, showing accommodation periods and the academic-calendar spacing used to capture early and longer-term network change.

Design Problem

Sociocentric network data create an awkward requirement: students need to see names when they nominate peers, but the analysis files should not expose names, emails, or contact fields. SAND handled that with separate REDCap projects, an identity-manager key, and de-identified exports for the research workflow.

Roster, blinded

Participants selected names in the survey. Research exports carried coded identifiers, so the analysis could rebuild nominations without publishing names.

Six waves, one linkage

Baseline and follow-up data used different REDCap records. The linkage file connected those records while keeping names and emails outside the analysis repo.

Restricted topology

Real edge lists, adjacency matrices, and anonymized network plots stay out of the public page. The site shows structure, schema, and proxy-mode documentation instead.

REDCap workflow showing participant, validation officer, identity manager, and researcher data lanes — REDCap workflow. Participants saw names when they nominated peers; researchers received only coded exports for linkage and analysis.

Fieldwork Record

Operational numbers that give context for the infrastructure: recruitment, retention, and response coverage.

68% Baseline response rate: 255 of 375 invited residents enrolled.

87% Six-wave retention, with 223 respondents at the final wave.

215-247 Complete drinking-behaviour responses across the six waves.

220-223 Network nomination responses per follow-up wave from Wave 2 onward.

49% to 92% Increase in participants responding within seven days, from Wave 1 to Wave 6.

R/Make Chapter pipelines generate QA reports, model inputs, figures, tables, and logs.

Explore Further

Study design, wave maps, data dictionaries, key findings, the full REDCap infrastructure, the R/Make pipeline, and release verification are documented in the technical reference. The analysis code is open-source.

Data access: public material includes design notes, schema summaries, and synthetic proxy data. Participant records, contact fields, linkage keys, and real network files stay restricted.