Two-wave longitudinal survey setup and SEM analysis

Problem: a scientist needed a technical and analytical partner for a complex two-wave longitudinal study across 5 countries (US, UK, South Africa, Australia, Mexico), to test hypothesis of behavioral patterns and productivity relations and confirm or refute a theoretical multi-step dysregulation model through using statistical methods.

The need for a flawless survey infrastructure (Prolific + Typeform) that could screen, recruit and re-invite the right participants across two waves 7 days apart, with zero ID mismatches or Wave 2 leakage.

Goal: deliver rigorous, publication-quality statistical analysis (CFA, SEM, bootstrap mediation) to test the scientist’ theoretical model.

Recruitment and pre-screening (Prolific)

• Designed and ran a Prolific pre-screen to identify 4 target subgroups, applying strict inclusion criteria and screening quotas to ensure balanced representation.
• Recruited across 5 countries (US, UK, South Africa, Australia, Mexico) with a Spanish-language translation prepared for the Mexico segment Structured participant payment and incentives to maximize retention.

Wave 2 allowlist system and pre-launch testing

• Built a secure Prolific allowlist so only Wave 1 completers could access Wave 2, with IDs matched automatically 7 days after Wave 1 completion.
• Use Typeform engine for survey. Typeform supported custom ID tracking and had no content-policy restrictions for sensitive-topic research.
• Validated the merge logic between the Prolific allowlist and Typeform’s data base, checking for ID mismatches before any real participant touched the link.
• Ran full end-to-end tests with multiple test IDs simulating the entire Wave 1 -> 7-day wait -> Wave 2 flow, confirming no leakage of the Wave 2 link and no drop-off caused by setup error before launch.

Data preparation and Confirmatory Factor Analysis (CFA)

• Cleaned and merged datasets across all collection waves by participant ID; check reverse code items, and created coded data base, prepared to analysis.
• Ran reliability analysis on theoretical model – six latent constructs (Predictor, Mediator 1, Mediator 2, Mediator 3, Mediator 4, Outcome 1, Outcome 2) before any structural modeling.
• Built the CFA measurement model in R (lavaan, ML estimator with FIML for missing data), iterating through 4 model versions – Model Initial → A_v1 → A_v2 → A_v3 – restructure model – removing 8 items and adding 5 residual covariances to reach excellent fit (CFI=.945, TLI=.938, RMSEA=.050, SRMR=.052).
• Validated the measurement model with convergent/discriminant validity checks (AVE, composite reliability, HTMT) and standardized residual analysis, confirming the same item-removal decisions independently replicated across all sample sizes.

Structural Equation Modeling (SEM) and robustness checks

• Built and tested the full structural model – a serial mediation chain running Predictor → Mediator 1 → Mediator 2 → Mediator 3 → Mediator 4 → Outcome 1 / Outcome 2 — with bootstrap significance testing (5,000 resamples, 95% percentile confidence intervals) on every direct and indirect path.
• Ran longitudinal measurement invariance testing (Wave 1 vs Wave 2), reaching excellent fit (CFI=.989) and confirming the autoregressive SEM design was fully justified.
• Conducted subgroup analyses across all 4 pre-screen categories, each with its own bootstrap-tested indirect effects.
• Ran a multi-group measurement invariance and gender-split analysis as an additional robustness check, confirming the core mediation finding was gender-invariant.

Interested in the report? Let us know and we will be happy to help!

Delivered a flawless technical data pipeline: Technically configured Prolific and Typeform into one seamless data-collection system. Built the waves allowlist, automated ID matching, and validated the full setup end-to-end.

Tested the theoretical model with rigorous statistical analysis: Ran the full serial mediation chain (Predictor -> Mediator 1 -> Mediator 2 -> Mediator 3 -> Mediator 4 -> Outcome 1 / Outcome 2) through CFA and SEM with bootstrap significance testing (5,000 resamples)

Confirmed and refined the structural path: Iterated the CFA measurement model through 4 versions to reach excellent, replicated fit (CFI=.945, RMSEA=.050), validated across three independent samples and confirmed under longitudinal, multi-group, and subgroup robustness checks

Prepared all materials and analysis results for the client: Publication-quality tables, fit-index reporting, and clear written interpretation of results.