Empirical Assessment of Effect of Publication Bias on a Meta-Analysis of Validity Studies on University Matriculation

Abstract

<p>This study examined the impact of publication bias on a meta-analysis of empirical studies on validity of University</p><p>Matriculation Examinations in Nigeria with a view to determine the level of difference between published and unpublished</p><p>articles. Specifically, the design was an ex-post facto, a causal comparative design. The sample size consisted of 30</p><p>empirical studies selected on the basis of empirical status and relevance. The characteristics of these studies were</p><p>recorded, coded and analyzed. The results revealed that the impact of publication bias was minimal on the meta-analysis</p><p>study since the two versions i.e. published and unpublished articles yielded essentially similar estimates of effect</p><p>sizes. (Mean Fisher =0.393, Weighted Fisher =0.398 with associated r =0.375 for both). Even though the use of non-</p><p>representative proportion of significant studies in the two positive directions led to a non-representative set of</p><p>studies, published articles had a higher effect size than the unpublished articles. ( = 0.460 for published, while =0.279</p><p>for unpublished). The effect size of published articles was significantly different from unpublished articles ( χ2=5.42 p</p><p>< 0.05 sig). This was an indication that studies with more significant results were the published. On the overall, there</p><p>was no significant difference in the level of significance for both published and unpublished articles ( χ2 = 0.0031 p ></p><p>0.05). Hence, publication bias did not affect the results of this meta-analysis.</p