Longitudinal Cohort Study for a
Child Language Development

RESEARCH QUESTION

What the Study Set Out to Ask

This longitudinal cohort study example follows a developmental psychology research team that recruits about 600 three-year-olds and assesses them every year until age seven. The team wants to describe how receptive vocabulary grows over those years, and whether children whose families use more of the language of schooling at home start higher, grow faster, or both.

The cohort is observational and home language use is not assigned, so differences in growth are described as associations that family income, parental education and preschool experience may partly explain. The plan lists those covariates, the growth forms to be compared and what counts as faster growth before any outcome is examined. A question about change also depends on the measure meaning the same thing at every age, which the analysis tests instead of assuming.

STUDY DESIGN AND DATA

Five Annual Waves for About 600 Children

A prospective cohort assessed at about ages three, four, five, six and seven, with a picture-vocabulary test at each wave. The test has a younger and an older form that share a set of anchor items. Each record holds the child, the exact age at assessment, the vocabulary score, home language use reported by a parent at baseline, weekly hours in group childcare at each wave and family background. Some families skip a wave or leave the study, and children are not assessed at the same age within a wave.

METHOD AND ASSUMPTIONS

Why a Growth Model, and What Must Hold

A growth model describes each child's trajectory by an intercept and a slope that vary across children, and estimates their means, their variances and their association with home language use. Time is entered as each child's age at assessment, so a wave that spreads over several months does not blur the shape. Linear, piecewise and quadratic forms are compared on fit, parsimony and interpretability, because a developmental curve is not assumed to be straight. The older form of the test differs from the younger one, so the vocabulary construct is first modeled from the shared items and tested for invariance of loadings and thresholds across ages. Without a sufficient degree of invariance, a change in score cannot be read as a change in vocabulary.

Missing waves are handled by full information maximum likelihood, which assumes that whether a score is missing depends only on observed information, not on the missing score itself. Variables that predict dropout but are not in the model, such as residential moves recorded at each contact, are added as auxiliary variables to make that assumption more credible, and a sensitivity analysis imputes the missing scores, shifts them in a plausible direction and shows how far the conclusions move. The invariance models, fitted wave by wave, are judged on several fit indices together; the growth models use each child's exact age, so their forms are compared on information criteria and likelihood ratio tests. A growth factor variance near zero or negative is treated as a sign of misspecification or too little between-child variation. The time-varying covariate, childcare hours, enters with a lag so that it comes before the score it is related to.

OUTPUT AND INTERPRETATION

How the Results Are Read

The output has the estimated mean trajectory with its 95% confidence band, the variances of the intercept and slope with intervals, and the association of home language use with each of them, reported with 95% confidence intervals and in score units per year of age. A table of the invariance steps with their fit indices, and of the information criteria behind the choice of growth form, sits beside them, with an attrition table that compares children who stayed with children who left on their baseline measures.

Sensitivity analyses compare the main model with a completers-only analysis, with the other growth forms and with other ways of handling the time-varying covariate. The write-up describes associations within this cohort and not effects of home language, and it limits conclusions to children like those recruited. What the pattern means for classroom practice is a judgment for the research team.

REPORTING

What the Researcher Received

A written report to APA 7, with the measurement model, the growth model and the missing-data handling described in methods-section wording, and tables formatted for the target journal. The research question, the developmental interpretation and the conclusions remain with the research team.