查詢結果分析
來源資料
頁籤選單縮合
題名 | 出版偏誤對以相關為基礎之整合分析的影響=The Influence of Publication Bias on the Meta-Analytic Model Based on Correlations |
---|---|
作者姓名(中文) | 蔡坤宏; | 書刊名 | 管理學報 |
卷期 | 18:1 2001.03[民90.03] |
頁次 | 頁153-172 |
分類號 | 487.7 |
關鍵詞 | 出版偏誤; 整合分析; 相關; Publication bias; Meta analysis; Correlation; |
語文 | 中文(Chinese) |
中文摘要 | 本研究主要目的在試圖建構出版偏誤下以相關係數為基礎之整合分析(後設分析),繼而,進一步觀察忽略出版偏誤對模式估計的影響效果。根據模擬分析結果發現: (1)構念間平均相關程度愈低、文獻間樣本大小變異愈小,則不論文獻數目的多寡或干擾效果是否明顯,構念間平均相關程度(截距參數)很可能高估; 而當干擾效果存在時,同樣條件下,則很可能低估干擾效果(變數參數); (2)構念間平均相關程度愈高、文獻間樣本大小變異愈小、文獻數目少,無論干擾效果是否明顯,很可能高估構念間平均相關程度; 而當干擾效果存在時,同樣條件下,則很可能高估干擾效果; (3)構念相關程度趨中,無論文獻間樣本大小變異、文獻數目的多寡或干擾效果是否存在,參數估計值皆相當精確; (4)型一錯誤發生頻率,除在構念相關程度低且文獻間樣本大小變異小的情形下可能低於5%外,其餘條件下則皆明顯高於臨界值(5%)。明顯地,這些發現有助於更深入瞭解整合分析(以相關為基礎者)應用上忽略出版偏誤的影響。 |
英文摘要 | This article aims to develop a meta-analytic model based on correlations in the presence of publication bias. Subsequently, the effect caused by neglecting the publication bias is further explored in the paper. Several results from Monte Carlo simulation have been revealed. (1) While the mean of correlations is low and the dispersion extent of sample sizes among studies is small, the average of correlations (intercept) is most likely to be overestimated; under the same conditions, the effect of mediation (non-intercept) is underestimated if the mediator is present. (2) Under low mean correlation, small extent of dispersion of sample sizes among studies, and small number of studies, the average correlation is most likely to be overestimated and if the mediator is present, the effect of mediation is prone to be overestimated. (3) While the mean of correlation is medium, the bias of estimation is not observed. (4) Type one errors are above the critical value on most conditions of the simulation. Obviously, these findings provide the deeper understanding to the influence due to the negligence of publication bias in the application of meta analysis of correlations. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。