完整後設資料紀錄
DC 欄位 | 值 | 語言 |
---|---|---|
dc.contributor.author | 吳俊育 | en_US |
dc.contributor.author | Wu Jiun-Yu | en_US |
dc.date.accessioned | 2014-12-13T10:42:55Z | - |
dc.date.available | 2014-12-13T10:42:55Z | - |
dc.date.issued | 2011 | en_US |
dc.identifier.govdoc | NSC100-2410-H009-002 | zh_TW |
dc.identifier.uri | http://hdl.handle.net/11536/99464 | - |
dc.identifier.uri | https://www.grb.gov.tw/search/planDetail?id=2204520&docId=351565 | en_US |
dc.description.abstract | 中文摘要資料相關性的議題已經廣泛引起社會與教育領域研究者的關心與注意。「基於設計」(design-based) 與「基於取樣」(model-based)已經被證明在分析複雜性取樣資料時,會產生一致且不偏的參數估計與統計推論結果;不過,這樣的一致結果只會出現在複雜性取樣資料有一模一樣的高低階層架構的時候 (Muthé;n and Satorra, 1995)。只要不同階層的模型擁有不一樣的架構,「基於設計」的模型建構法就會產生偏差的固定效果與隨機效果的參數估計及錯誤的統計推論結果(Wu & Kwok, In press). 事實上,不同階層的模型擁有不一樣的架構是很常見的情況。我們至今仍不清楚不同分析方法在處理這些違反高低結構同質性的資料時,產生統計結果的異同處。本研究案將使用蒙地卡羅模擬與結構方程式模型來分析實際的多階層資料。我提出一個新的模型建構方法即「最大化模型」,並利用實證性資料仔細檢驗過上述三種分析方法的效果。結果顯示「基於設計」方法在分析多階層資料時,多會產生偏誤的固定效果估測以及總合的隨機效果估測。在模型建構絕對正確的情況下,「基於取樣」模型在相同與不同的高低層級結構下,對因素結構與負載量不衡等的情況,仍能產生與母體值相符的參數估計。 相對的、即便在對高階層結構一無所知的情況下,「最大化模型」在分析多階層資料的兩種高低階層不一致的情況下,學者仍可以藉此獲得不偏的參數估計與統計推論。因此、在分析多階層資料,當我們不了解高階層架構或只對低階層模型有興趣時,「最大化模型」比另兩種方法有更高的建模自由度與不偏的統計分析結果,值得推薦給社會與教育領域研究者有此需求時使用。 | zh_TW |
dc.description.abstract | AbstractThe issue of data dependency has drawn a great deal of attention to social and educational researchers. The design-based and model-based approaches have shown to perform equally well for complex survey data under the same model specification for all data levels (Muthé;n and Satorra, 1995) but the design-based approach yielded biased estimation for the fixed effect and random effect for unequal data structure across levels (Wu & Kwok, In press). A commonly found condition, yet not systematically examined, is the violation of pattern invariance for level-varying structures. The current study proposed the use of maximum models to analyze the multilevel dependent data with evidence from real data simulation and mathematical derivation. The result showed the design-based approach produced biased estimates for both the fixed and random effects. The model-based approach yielded level-varying parameter estimates consistent to the population value iff the true multilevel model is specified; nevertheless, the maximum model generated parameter estimates consistent to the population values in the within level by estimating the full rank of variance-covariance matrix in the between level. The maximum model is recommended rather than design-based approach for substantive researchers if limited information about the higher-level structure is available for the dependent data. | en_US |
dc.description.sponsorship | 行政院國家科學委員會 | zh_TW |
dc.language.iso | zh_TW | en_US |
dc.subject | 複雜性取樣資料 | zh_TW |
dc.subject | 多階層資料結構 | zh_TW |
dc.subject | 結構方程模式 | zh_TW |
dc.subject | 多階層結構方程模式 | zh_TW |
dc.subject | 多階層模型分析 | zh_TW |
dc.subject | 穩健標準誤估測器 | zh_TW |
dc.subject | 基於取樣方法 | zh_TW |
dc.subject | 基於設計方法 | zh_TW |
dc.subject | 最大化模型 | zh_TW |
dc.subject | Complex survey data | en_US |
dc.subject | Multilevel data structure | en_US |
dc.subject | SEM | en_US |
dc.subject | Multilevel SEM | en_US |
dc.subject | Multilevel modeling | en_US |
dc.subject | Robust SE estimator | en_US |
dc.subject | Design-based approach | en_US |
dc.subject | Model-based approach | en_US |
dc.subject | Maximum model | en_US |
dc.title | 在結構方程模式的範疇下以實際資料探究三種解決數據相關性方法的有效性 | zh_TW |
dc.title | Testing the Effectiveness of Three Modeling Approaches in Addressing Data Dependency with Empirical Data under Structural Equation Modeling Framework | en_US |
dc.type | Plan | en_US |
dc.contributor.department | 國立交通大學教育研究所 | zh_TW |
顯示於類別: | 研究計畫 |