Devpsych Study Design

Study Design (devpsych-study-design)

A developmental design must support a claim about change, not just measure something at different ages. Developmental Psychology reviewers probe the four threats that are specific to this field: age vs. cohort confounds, attrition, measurement invariance across ages, and age-appropriate ethics and task validity. This skill hardens the design before data collection.

When to trigger

• Planning a cross-sectional, longitudinal, accelerated, micro-genetic, or experimental developmental study
• Writing a preregistration / pre-analysis plan for developmental work
• A reviewer questioned age confounds, attrition, invariance, power, or ethics with minors
• Justifying sample size for a growth model or an age interaction

Developmental design standards

1. Match the design to the change claim.
   • Cross-sectional age comparison: cheap, but age is confounded with cohort and with selection/era; you can describe age differences, not within-person change.
   • Longitudinal: supports within-person change but introduces attrition and retest effects.
   • Accelerated longitudinal (cohort-sequential): overlapping age cohorts to cover a wide span fast — state the convergence assumption.
   • Micro-genetic: dense repeated observation to catch the process of change; justify sampling rate.
2. Address age vs. cohort. For any age effect, say what could be cohort/era instead, and how the design (or covariates, or a sequential design) addresses it. Do not call a cross-sectional age difference "development" without this.
3. Plan for measurement invariance. The same instrument can mean different things at different ages. Pre-specify a configural → metric → scalar invariance test across ages/waves; interpret change only at the level of invariance you establish (handoff to devpsych-data-analysis).
4. Plan for attrition. Estimate expected dropout, design retention, pre-specify the missing-data model (FIML/MI) and an attrition (MCAR/MAR) analysis comparing completers vs. dropouts.
5. Justify sample size. Power for the change parameter you care about — a growth slope, an age × condition interaction, or a cross-lagged path — not just a group mean difference. State the assumed effect size and its source.
6. Age-appropriate ethics and validity. Child assent plus parental/guardian consent; age-appropriate measures (a task valid at 4 and at 8); special protections for vulnerable populations.

Pre-data lockdown checklist (developmental)

Sample-size justification — worked example (illustrative)

For a three-wave latent-growth study (ages 4, 6, 8), justify N for the slope and the scaffolding × time interaction, not a t-test.

Target parameter: latent slope variance + scaffolding × time interaction.
Method: Monte Carlo power simulation (lavaan/Mplus) under a plausible
        growth model with 20% per-wave attrition and FIML.
Result: N = 300 at wave 1 gives ~85% power for the interaction at the
        smallest developmentally meaningful slope difference; precision
        goal is a slope-CI half-width small enough to sign the trajectory.
Invariance: configural→metric→scalar tested across waves before growth is
        interpreted; partial scalar invariance plan if a few intercepts differ.
Attrition: MAR assumed; completers-vs-dropouts compared on baseline covariates.

Design-stage reviewer pushback and the venue fix

• "This is a cohort effect, not development" → add a sequential element or model cohort; soften to age-difference language if you cannot separate them.
• "Is the construct the same at every age?" → pre-specify and report measurement invariance; interpret change only at the invariance level achieved.
• "Attrition could bias the trajectory" → report the attrition analysis and a principled missing-data model, not listwise deletion.
• "Task isn't valid across this age range" → justify age-appropriateness or use age-anchored measures.

Anti-patterns

• Calling a cross-sectional age difference "developmental change"
• Interpreting trajectories without testing measurement invariance
• Listwise deletion / ignoring differential attrition
• Powering for a mean difference when the claim is a growth slope or an age interaction
• Consent without child assent; one task used across ages it is not valid for

Output format

【Design】cross-sectional / longitudinal / accelerated / micro-genetic / experiment
【Change claim supportable】age vs. cohort addressed? [Y/N]
【Invariance plan】configural→metric→scalar across ages/waves? [Y/N]
【Attrition plan】expected dropout + missing-data model + attrition analysis? [Y/N]
【Sample size】N + power for the change parameter (slope/interaction)
【Ethics】consent + child assent + age-appropriate measures? [Y/N]
【Next】devpsych-data-analysis

Supplementary resources

• ../../resources/external_tools.md — Mplus/lavaan, simr, power simulation, longitudinal design references
• ../../resources/official-source-map.md — JARS design/reporting requirements and ethics policy

Source: brycewang-stanford/Awesome-Journal-Skills → Developmental-Psychology-Skills/skills/devpsych-study-design/SKILL.md