This three-article format dissertation aims to provide evidence of the effect size of the relationship between writing self-efficacy and writing achievement in an empirical study and a meta-analysis, and compare three methods, the averaging method, three-level meta-analysis (3LM), and robust variance estimation (RVE), in handling multiple dependent effect sizes within studies in doing the meta-analysis. In the first article, data were collected from college students in an English as a foreign language (EFL) context. The second article extracted 575 effect sizes from 76 primary studies, which were also used in the third article. Results revealed a medium to large observed effect size (r = .47) with EFL students and a medium true effect size (r = .29) of the relationship between writing self-efficacy and writing achievement. Writing in English as a first language (L1) or a second language (L2) was found to moderate the relationship between writing self-efficacy and writing achievement, with the effect size estimated with L2 learners (r = .44) being statistically significantly larger than that yielded with L1 learners (r = .23). Results also revealed that statistical procedure moderated effect size estimates. The third article found that while the averaging method, 3LM, and RVE yielded similar estimates of the average effect size and standard error, the averaging method resulted in a much smaller variance of true effect sizes. 3LM produced the smallest variance estimate when the number of primary studies is small. This dissertation provides evidence of the observed and true effect sizes of the relationship between writing self-efficacy and writing achievement. It also informs meta-analysts of the performance of the three methods in handling dependent effect sizes in doing meta-analyses.