Empirical Study on Applying Meta-analysis Methods in the Social Sciences and Psychological Research

点击翻译按钮获取中文摘要

In the social sciences and psychological research, it is not unusual for plural research with the same research problems to have inconsistent results, or even have diametrically apposite results. Integrating separate research results into a coherent and comprehensive whole involves inferences as central to the validity of knowledge as the inferences involved in drawing conclusions from primary data analysis. As a method of research integrating, meta-analysis is recognized as useful and important, that it has rapidly gained acceptance and becomes popular in many foreign countries.However in Japan meta-analysis methods are still not being widely used in the social sciences and psychological research The process of meta-analysis can be conceptualized as containing five stages: (1) problem formulation, (2) data collection or the literature search, (3) data evaluation: literature coding, (4) analysis and interpretation, and (5) presentation of results. In this study, we especially focused o … More n stages (2) and (4) which are considered as big obstacles to practice the meta-analysis methods for Japanese researchers.For (2) data collection part, we used PsychInfo, Scopus and CJP (Citation Database for Japanese Papers) for selecting the relevant literatures based on bibliometric strategies, such as by using citation indexing, author indexing, number of citations and keywords extracting from the research papers. The effective strategies for searching the literatures comprehensively were investigated and discussed in our study. Concerning the stage (4), we paid attention to effect sizes, the base of the statistical methods in meta-analysis. In the social sciences and psychological research it seems that hypothesis significance testing has been given too much emphasis. In our study, we have shown the importance of reporting effect sizes which indude those information people can never obtain just from significance testing. The results of the studies were presented in a number of annual meetings ofJapanese academic societies.Various factors may hinder using meta-analysis methods, and one of the reasons could be that there is currently no software available for calculating effect sizes and confidence intervals easily ; as we know it is practically impossible to calculate the values without computers. In this study, we developed an Excel macro computer program to estimate confidence intervals around the standardized mean difference 6 based on the bootstrap BC_a procedure. The results of our simulation show that our program can precisely estimate confidence intervals except when all the following situations happen simultaneously : the normality assumption is extremely violated, effect sizp is large, and sample size is very small. Less