發(fā)布時間：2019-04-03 11:47

【摘要】：隨著當(dāng)今心理與教育領(lǐng)域測驗的發(fā)展,測驗已經(jīng)出現(xiàn)了由采用獨立多項選擇題向采用題組的過渡趨勢。為了滿足實際測驗的需要,對題組進(jìn)行研究的必要性日顯突出。本文首先對題組的相關(guān)概念給予綜述,在闡明了處理題組數(shù)據(jù)時若忽略題組效應(yīng)會導(dǎo)致參數(shù)估計偏差等諸多問題后介紹了6個已有題組反應(yīng)模型對題組效應(yīng)的處理方法。并探討了“廣義局部獨立性假設(shè)”、“題組反應(yīng)模型和多維項目反應(yīng)模型的關(guān)系”以及“題組反應(yīng)模型、二階模型和雙因素模型間的關(guān)系”三個問題。 基于上述回顧,筆者認(rèn)識到已有的題組反應(yīng)模型研究均假設(shè)題組項目反應(yīng)僅受到1個共同刺激的影響,這導(dǎo)致了“題組”概念的內(nèi)涵與外延不互補(bǔ)。在對題組本質(zhì)的詮釋(即一個存在共同刺激的項目集合)的基礎(chǔ)上,又提出了項目內(nèi)多維題組效應(yīng)的概念,而這是已有題組反應(yīng)模型所無法處理的。對此,本研究先將單維題組效應(yīng)參數(shù)γ擴(kuò)展為多維題組效應(yīng)矩陣r并設(shè)定判定矩陣U以期有效合理地抽離項目內(nèi)多維題組效應(yīng),基于此提出了(項目內(nèi))多維題組效應(yīng)模型,且采用馬爾可夫鏈蒙特卡洛算法實現(xiàn)了模型的參數(shù)估計。在研究一中與單維題組效應(yīng)模型和標(biāo)準(zhǔn)項目反應(yīng)模型的對比研究表明：(1)當(dāng)測驗存在項目內(nèi)多維題組效應(yīng)時,采用單維題組效應(yīng)模型或標(biāo)準(zhǔn)項目反應(yīng)模型均會導(dǎo)致參數(shù)的偏差估計；(2)多維題組效應(yīng)模型相對于單維題組效應(yīng)模型和標(biāo)準(zhǔn)項目反應(yīng)模型更具普適性。 之后,本文進(jìn)一步拓廣了多維題組效應(yīng)模型,提出：將目標(biāo)潛質(zhì)與多維題組效應(yīng)設(shè)定為具有不同區(qū)分度參數(shù)的廣義多維題組效應(yīng)模型,同樣采用馬爾可夫鏈蒙特卡洛算法實現(xiàn)了模型的參數(shù)估計。在研究二中與多維題組效應(yīng)模型的對比研究表明：(1)當(dāng)不考慮模型復(fù)雜性時,廣義多維題組效應(yīng)模型優(yōu)于多維題組效應(yīng)模型；(2)當(dāng)考慮模型復(fù)雜性時,多維題組效應(yīng)模型比廣義多維題組效應(yīng)模型更適用于處理較簡單的測驗情景�？梢哉f,廣義多維題組效應(yīng)模型與多維題組效應(yīng)模型各有優(yōu)勢,適用于不同的測驗情景。 再后,本文探討了多維題組效應(yīng)模型、分層模型和高階模型之間的關(guān)系,認(rèn)為多維題組效應(yīng)模型與分層模型和高階模型三者近似等價。 最后,本文對國際閱讀素養(yǎng)進(jìn)步研究2006進(jìn)行了實證分析。分析結(jié)果表明多維題組效應(yīng)模型在實際測驗分析中具有可行性。
[Abstract]:With the development of psychological and educational field tests, there has been a transition from independent multi-choice questions to adoption groups. In order to meet the needs of the actual test, the necessity of studying the group is becoming more and more prominent. In this paper, the related concepts of cluster are summarized firstly. After clarifying many problems such as parameter estimation deviation caused by neglecting cluster effect when dealing with cluster data, this paper introduces the treatment methods of cluster effect by six existing cluster response models. Three problems are also discussed: "the generalized local independence hypothesis", "the relationship between the cluster response model and the multi-dimensional project response model" and "the relationship between the cluster response model, the second-order model and the two-factor model". Based on the above-mentioned review, the author recognizes that the existing research on cluster response model assumes that the cluster project response is only influenced by one common stimulus, which leads to the incomplementarity of the connotation and extension of the concept of "cluster". On the basis of the interpretation of the essence of the cluster (that is, a set of items with common stimuli), the concept of multi-dimensional cluster effect within the project is proposed, which can not be dealt with by the existing cluster response models. In this paper, the one-dimensional cluster effect parameter 緯 is extended to multi-dimensional cluster effect matrix r and the decision matrix U is set up in order to extract the multi-dimensional cluster effect effectively and reasonably. Based on this, a multidimensional cluster effect model (within the project) is proposed. The Markov chain Monte Carlo algorithm is used to estimate the parameters of the model. Compared with the one-dimensional cluster effect model and the standard item response model in study one, the results show that: (1) when there is multi-dimensional cluster effect in the item, the results show that: (1) when there is multi-dimensional cluster effect, The use of one-dimensional group effect model or standard project response model will lead to parameter deviation estimation; (2) the multi-dimensional group effect model is more universal than the one-dimensional group effect model and the standard project response model. Then, this paper extends the multi-dimensional group effect model, and proposes that the target potential and the multi-dimensional group effect are set as the generalized multi-dimensional group effect model with different discriminative parameters, and that the target potential and the multi-dimensional group effect are set as the generalized multi-dimensional group effect model. Markov chain Monte Carlo algorithm is also used to estimate the parameters of the model. The comparative study of the second middle model and the multi-dimensional group effect model shows that: (1) when the complexity of the model is not considered, the generalized multi-dimensional group effect model is better than the multi-dimensional group effect model; (2) when the complexity of the model is considered, the multi-dimensional group effect model is more suitable than the generalized multi-dimensional group effect model for dealing with simple test situations. It can be said that the generalized multi-dimensional group effect model and the multi-dimensional group effect model have their own advantages and are suitable for different test scenarios. Then, the relationship among multi-dimensional group effect model, hierarchical model and higher-order model is discussed, and it is considered that the multi-dimensional group effect model is approximately equivalent to the hierarchical model and higher-order model. Finally, this paper carries on the empirical analysis to the international reading literacy progress research 2006. The results show that the multi-dimensional group effect model is feasible in practical test analysis.
【學(xué)位授予單位】：浙江師范大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：B84-0

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 葉萌;;對IRT局部獨立性假設(shè)問題的認(rèn)識[J];考試研究;2010年02期

2 辛濤;樂美玲;張佳慧;;教育測量理論新進(jìn)展及發(fā)展趨勢[J];中國考試;2012年05期

3 鄭蟬金;郭聰穎;邊玉芳;;變通的題組項目功能差異檢驗方法在篇章閱讀測驗中的應(yīng)用[J];心理學(xué)報;2011年07期

4 劉s，

本文編號：2453184