技能 数据科学 发展心理学数据分析规范

发展心理学数据分析规范

v20260724
devpsych-data-analysis
本指南用于发展心理学领域的纵向数据分析和报告,确保符合高标准的学术要求。重点指导如何正确建模发展变化(如增长曲线、多层次模型、结构方程模型),如何在解释平均变化前先检验测量不变性(configural→metric→scalar),并严格使用JARS指南报告效应值和置信区间。核心目的是区分和清晰报告确认性与探索性发现。
获取技能
392 次下载
概览

Data Analysis (devpsych-data-analysis)

Developmental Psychology holds analyses to a developmental and a credibility standard at once: the model
must actually capture change (not just a cross-sectional snapshot), and reporting must meet JARS —
effect sizes with confidence intervals, full disclosure, and a clean confirmatory vs. exploratory
split. The most common fatal error is interpreting trajectories without first establishing that the
construct is measured the same way across ages.

When to trigger

  • Fitting growth-curve / multilevel / SEM models, or mediation/moderation of developmental effects
  • A reviewer asked for measurement invariance, effect sizes, intervals, or attrition handling
  • Reconciling preregistered developmental hypotheses with exploratory trajectory findings
  • Preparing analysis scripts and a data dictionary for deposit

Reporting norms Developmental Psychology expects

  1. Model change correctly. Use the method the claim requires: latent growth / multilevel models

for trajectories, SEM for latent constructs, cross-lagged / RI-CLPM for reciprocal effects,
mediation/moderation for mechanism and moderated change. State time coding and centering.

  1. Establish measurement invariance first. Test configural → metric → scalar across ages/waves

before interpreting mean change; report partial invariance honestly if full scalar fails.

  1. Effect sizes + uncertainty. Report a standardized or unstandardized effect size **and confidence

intervals** for major results — slope estimates, interactions, indirect effects — not just stars.

  1. Handle missing data and attrition principledly. Use FIML or multiple imputation; report the

attrition analysis (completers vs. dropouts) and the missingness assumption.

  1. JARS disclosure. Report how sample size was determined, all exclusions and reasons, all conditions

and measures; keep confirmatory and exploratory analyses clearly separated.

  1. Reproducibility. Provide analysis scripts and a data dictionary; the numbers should regenerate in

a fresh session (see devpsych-open-science-and-transparency).

Worked micro-example (illustrative numbers)

A preregistered three-wave latent-growth study (ages 4, 6, 8; N = 300, 18% attrition) of effortful
control, testing maternal scaffolding as a driver of the growth slope.

Invariance (reported first):
  configural fit good; metric and scalar invariance hold across waves
  (ΔCFI < .01) → mean change is interpretable.
Confirmatory (preregistered):
  Latent slope > 0: b = 0.42/year, 95% CI [0.31, 0.53] (within-person growth).
  Scaffolding × time: b = 0.18, 95% CI [0.07, 0.29] (steeper growth with
  higher wave-1 scaffolding).
  Missing data: FIML; MAR; completers and dropouts did not differ on baseline
  covariates (attrition analysis in supplement).
Exploratory (labeled):
  RI-CLPM suggests child→parent effects in later waves; reported as
  exploratory and flagged for confirmation in a future sample.

Why this passes scrutiny: invariance is reported before the growth claim; every developmental parameter
carries an effect size and a CI; missingness is modeled, not deleted; the reciprocal-effects finding is
honestly demoted to exploratory.

Analysis-stage reviewer pushback and the venue fix

Reviewer pushbackWhat it signals hereDevelopmental Psychology fix
"Is the construct the same at each age?"invariance not testedreport configural→metric→scalar before interpreting change
"You deleted dropouts"attrition biasrefit with FIML/MI; add the completers-vs-dropouts analysis
"ANOVA on age groups for a change claim"wrong model for the claimfit a growth/multilevel model on within-person data
"Stars, no effect size"pre-reform reportingreport slope/interaction effect sizes with CIs
"Is this confirmatory?"HARKing concernpoint to preregistration; relabel post hoc trajectories exploratory

Calibration anchors

  • A clean latent-growth slope with a tight CI, on an invariant measure, beats an age-group ANOVA with

stars — the venue's currency is credible change, not a snapshot contrast.

  • Prefer estimation language ("effortful control grew 0.42/year, 95% CI [...]") to "significant effect of

age." Bare p-value sentences read as thin here.

  • When attrition is non-trivial, state the missingness assumption and show the trajectory is robust to a

reasonable alternative (e.g., pattern-mixture sensitivity), rather than implying complete data.

Anti-patterns

  • Interpreting mean change without establishing measurement invariance
  • Listwise deletion or ignoring differential attrition
  • Using age-group ANOVA to support a within-person change claim
  • p-values and stars with no effect sizes or confidence intervals
  • HARKing exploratory trajectory shapes into confirmatory hypotheses

Output format

【Model】growth / multilevel / SEM / cross-lagged / mediation-moderation — matches the change claim?
【Invariance】configural→metric→scalar tested before interpreting change? [Y/N]
【Main result】effect size + confidence interval + meaning
【Missing data】FIML/MI + attrition analysis reported? [Y/N]
【Confirmatory vs exploratory】clearly separated (JARS)? [Y/N]
【Reproducible】scripts + data dictionary + fresh-session check? [Y/N]
【Next】devpsych-tables-figures

Supplementary resources

信息
分类 数据科学
名称 devpsych-data-analysis
版本 v20260724
大小 6.16KB
更新时间 2026-07-28
语言