Longitudinal impact of early childhood science instruction on 5th grade science achievement

ABSTRACT

This study investigated if student placement in a primary grade 1–3 classroom with a teacher who had been trained in a U.S. science Framework-aligned [National Research Council. (2012). A framework for K-12 science education: Practices, crosscutting concepts and core ideas. The National Academies Press] professional development science approach impacted student science achievement as measured in 5th grade. Students in the treatment group also received take-home science materials and treatment families were invited to participate in community-based science events. A two-level, random-slope mixed regression model was used to assess the effect of the treatment on later student achievement as measured by performance on the Science Subtest of the Ohio Achievement Assessment. This study found that students from the treatment group scored significantly higher on 5th grade science tests as compared to their peers. Overall, this study suggests that providing Framework-aligned science instruction, coupled with parent support, during early years improves science skills in later elementary grades.

KEYWORDS:

• Early years/early childhood
• longitudinal study

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 It should be noted that other possible STAR content area scores (i.e. Reading and Mathematics scale scores) could not be used as baseline measures because many students were not tested in first grade in these content areas.

2 In order to obtain replacements for missing values we used a pan algorithm Markov Chain Monte Carlo (MCMC) technique. The missing values procedure used two phases, a so-called burn-in, and an imputation stage. The burn-in phase uses iterations to stabilize estimation parameters, and the imputation phase draws replacement for missing values into the desired number of imputed datasets (Grund et al., 2016). As recommended for many missing values, 50,000 burn-in iterations and 100 datasets were created, each with 5,000 iterations apart. A separate, third procedure was used to combine all values (missing and non-missing) into separate datasets in the R mitml package. The following auxiliary variables were used in the imputation model: science scale score, retention status, minority status, gender, and treatment condition. A random-intercept model was used with school as a grouping factor. A random-intercept model was deemed to be more stable than a random-slope model that included the treatment condition variable as randomly varying as a function of schools. The highest R value was below 1.002 for all parameters and autocorrelation of consecutive elements of the MCMC chain were substantially diminished after 1000 iterations for the obtained baseline intercept variance. This value indicated that the convergence diagnostics in the data imputation procedures were stable. The random-intercept model presented in the equations modelled in this analysis used school-mean centred values for the STAR Early Literacy baseline measure. All other predictor variables were uncentred. An lmer function in the lme4 R package (Bates et al., 2015) was used to estimate the parameters of interest in pooled analysis for the 100 imputed datasets with the default restricted maximum likelihood estimation (REML) method.

3 In other parts of the world there exist similar documents such as the National Curriculum in the UK, the National Innovation and Science Agenda in Australia, the Nurturing Early Learners’ Framework in Singapore, and the European Commission’s report Science Education for Responsible Citizenship.

Additional information

Funding

This work was supported by National Science Foundation [grant number 1102808].