This dissertation is a three-piece dissertation, including two empirical research (Chapter 1 and Chapter 2) and a methodological improvement of prior work (Chapter 3), to address issues of the effects of approaches to learning on academic achievement in childhood and power analysis for a three-level model in meta-analysis.Approaches to learning as a key domain of school readiness has shown significant effects on student academic achievement. The study in Chapter 1 was designed to examine the potential moderation effects of problem behaviors on the association between approaches to learning and academic achievement (mathematics, reading, and science) in early grades using a recent nationwide longitudinal dataset (ECLS-K:2011). The correlated random effects estimation was applied to deal with the omitted variable issue. At the same time, the estimation method was allowed to compute the impacts of important time-constant variables (e.g., socioeconomic status) on academic achievement. The results indicated non-significant moderation effects of problem behaviors on the relations between approaches to learning and academic outcomes. However, the main effects of approaches to learning were significant associated with academic achievement. Complete data analysis and bootstrap with multiple imputation were conducted in the research to address the missing data issue in large-scale assessments. Similar results were shown in the two approaches, which demonstrated robust findings in the study. To better understand the general relations of approaches to learning on academic performance in childhood in recent years, the study in Chapter 2 conducted a systematic review employing meta-analytic methodology to combine and summarize the results from empirical quasi-experimental studies. The study filled the literature gap and extended the theory to understand the relations between approaches to learning and achievement. The results indicated medium effect sizes of the relations on approaches to learning and concurrent/future achievement (reading and mathematics). The effects on reading achievement were slightly larger than the effects on mathematics achievement. The single timepoint evidence showed stronger effects compared with longitudinal designs. In conclusion, the meta-analysis emphasized the positive and significant effects of approaches to learning on academic achievement in childhood. The methodological improvement in Chapter 3 aimed to address the potential biased power statistics when introducing group dependence in meta-analysis. The study extended the prior work about power analysis for two-level random-effects models to three-level models in the univariate case. The three-level model assumed research teams/labs at the third level. A three-level random-effects model provides more accurate estimates of power under the assumption that variability between research teams is not negligible. Each model in the study was followed by an illustrated example to show how to calculate the power. A simulation study provided evidence about how group-level heterogeneity affected statistical power in meta-analysis in the three-level model. The present study introduced more complicated data structures in meta-analysis and provided the power measures in advanced meta-analytic models.
Read
