Variáveis politômicas são comuns em experimentos agronômicos, apresentando natureza nominal ou ordinal. O modelo dos logitos generalizados é uma classe de modelos que pode ser empregada para a análise desses dados. Uma das características deste modelo é a pressuposição de que a variância é uma função conhecida da média e, espera-se, que a variância observada esteja próxima da variância pressuposta pelo modelo assumido. Contudo, quando ela é maior do que a especificada pelo modelo, tem-se o fenômeno da superdispersão. Nesse contexto, o presente trabalho objetivou caracterizar o problema da superdispersão associado a dados nominais em estudos \"cross-sectional\". Como motivação apresentam-se dois estudos adaptados da área de ciências agrárias relativos à fruticultura e zootecnia, ambos planejados no delineamento inteiramente casualizado. Verifica-se indicativo de superdispersão nos dados dos dois exemplos e como uma alternativa metodológica utilizou-se o modelo Dirichlet-multinomial. Por meio do gráfico de diagnóstico half-normal plot avaliou-se o ajuste do modelo dos logitos generalizados e do Dirichlet-multinomial. Adicionalmente, foi proposta uma extensão do índice de dispersão para os dados politômicos, com performance avaliada sob simulação. O modelo Dirichlet-multinomial mostrou-se adequado para o ajuste aos dados com superdispersão comparativamente ao modelo dos logitos generalizados. Apesar dos resultados satisfatórios obtidos, ressalta-se que este trabalho é uma introdução ao problema. / Polytomic variables are common in agronomic experiments, presenting nominal or ordinal nature. The generalized logits model is a class of models that can be used to analyze these facts. One of the characteristics of this model is the assumption that variance is a known function of the mean and. It is expected, that the analyzed variance is close to that assumed by the model. However, when it is larger than the one specified by the model, it has the phenomenon of overdipersion. In this context, the present work aims to characterize the problem of overdispersion associated with nominal data in cross-sectional studies. As motivation, it is showed two adapted studies of the agricultural sciences area, related to fruit growing and zootechnics, both planned in the completely randomized design. The Dirichlet-multinomial model was used as a methodological alternative and was indicated as an overdispersion in the facts of the two examples. The model of the generalized logits and the Dirichlet-multinomial model were evaluated using the half-normal plot. In addition, it was proposed an extension of the dispersion index for the polytomic data, with performance evaluated under simulation. The Dirichlet-multinomial model proved to be adequate for the adjustment to the overdispersed fact compared to the generalized logit model. Despite the satisfactory results obtained, it is emphasized that this work is an introduction to the problem.

進行調查時，最後的抽樣單位常是從不同的群集取得的，而同一群集內的樣本對象，因背景類似而對於某些問題常會傾向相同或類似的反應，研究者若忽略這種群內相關性，仍以獨立性樣本進行分析時，因其共變異數矩陣通常會與多項模式的共變異數矩陣相差懸殊，而造成所謂的額外變異或不足變異的現象。本文在不同的情況下，提出了Dirichlet-Multinomial模式(簡稱DM模式)、擴展的DM模式、以及兩種平均數-共變異數矩陣模式，以適當的彙整所有的群集資料。並討論DM與EDM模式中相關之參數及格機率之最大概似估計法，且分別對此兩種平均數-共變異數矩陣模式，提出求導廣義最小平方估計的程序。此外，也針對幾種特殊的二維表及三維表結構，探討對應的參數及格機率之估計方法。並提出計算簡易的Score統計檢定量以判斷群內相關(intra-cluster correlation)之存在性，及判斷資料集具有額外或不足變異，而對於不同母體的群內相關同質性檢定亦提出討論。 / This paper presents a modelling method of analyzing categorical data with overdispersion or underdispersion.　In many studies, data are collected from differ clusters, and members within the same cluster behave similary.　Thus, the responses of members within the same cluster are not independent and the multinomial distribution is not the correct distribution for the observed counts.　Therefore, the covariance matrix of the sample proportion vector tends to be much different from that of the multinomial model.　We discuss four different models to fit counts data with overdispersion or underdispersion feature, witch include Dirichlet-Multinomial model (DM model), extended DM model (EDM model), and two mean-covariance models.　Method of maximum-likelihood estimation is discussed for DM and EDM models.　Procedures to derive generalized least squares estimates are proposed for the two mean-covariance models respectively. As to the cell probabilities, we also discuss how to estimate them under several special structures of　two-way and three-way tables.　More easily evaluated Score test statistics are derived for the DM and EDM models to test the existence of　the intra-cluster correlation.　And the test of homogeneity of intra-cluster correlation among several populations is also derived.