Mathematical Sciences: Analysis of Qualitative Data with Log-Linear Models

定性数据的对数线性模型通常采用大样本最大似然估计和加权最小二乘估计的近似方法进行分析。计算问题使得相应的精确测试和可信区间很难使用。使用牛顿-拉夫森算法进行计算，并采用稍微不那么优化的程序来减少问题。将模型推广到具有随机效应的定性数据，以及仅与模型近似一致的数据。

Log-linear models for qualitative data are usually analyzed by means of the large-sample approximations for maximum- likelihood and weighted-least-squares estimators. Computational problems have made the corresponding exact tests and confidence intervals difficult to use. Employing the Newton-Raphson algorithm for the computation and falling back on slightly less optimal procedures reduces the problems. The models are extended to qualitative data with random effects as well as data which is only approximately consistent with the model.