## Three Centuries of Categorical Data Analysis: Log-linear Models and Maximum Likelihood Estimation

### Cached

### Download Links

Citations: | 6 - 3 self |

### BibTeX

@MISC{Fienberg_threecenturies,

author = {Stephen E. Fienberg and Alessandro Rinaldo},

title = {Three Centuries of Categorical Data Analysis: Log-linear Models and Maximum Likelihood Estimation},

year = {}

}

### OpenURL

### Abstract

The common view of the history of contingency tables is that it begins in 1900 with the work of Pearson and Yule, but it extends back at least into the 19th century. Moreover it remains an active area of research today. In this paper we give an overview of this history focussing on the development of log-linear models and their estimation via the method of maximum likelihood. S. N. Roy played a crucial role in this development with two papers co-authored with his students S. K. Mitra and Marvin Kastenbaum, at roughly the mid-point temporally in this development. Then we describe a problem that eluded Roy and his students, that of the implications of sampling zeros for the existence of maximum likelihood estimates for loglinear models. Understanding the problem of non-existence is crucial to the analysis of large sparse contingency tables. We introduce some relevant results from the application of algebraic geometry to the study of this statistical problem. 1

### Citations

1581 | Generalized linear models - Nelder, Wedderburn - 1972 |

1205 | Causality: models, reasoning, and inference - Pearl - 2000 |

1141 | Graphical models - Lauritzen - 1996 |

454 | Graphical models in Applied Multivariate Statistics - Whittaker - 1990 |

338 | Statistical aspects of the analysis of data from retrospective studies - Mantel, Haenszel - 1959 |

200 |
Discrete Multivariate Analysis
- Bishop, Fienberg, et al.
- 1975
(Show Context)
Citation Context ...ssessment of fit, model selection and interpretation. The existence of the MLE is essential for the usual derivation of large-sample chi-square approximations to numerous measures of goodness of fit (=-=Bishop et al., 1975-=-; Agresti, 2002; Cressie and Read, 1988) which are utilized to perform hypothesis tests and, most importantly, are an integral part of model selection. If the distribution of the statistic measuring t... |

185 | Algebraic algorithms for sampling from conditional distributions - Diaconis, Sturmfels - 1998 |

170 | Introduction to graphical modelling - Edwards - 2000 |

156 | Fundamentals of Statistical Exponential Families - Brown - 1987 |

136 | On the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling - Pearson - 1900 |

132 | The Analysis of Cross-Classified Categorical Data - Fienberg - 1989 |

114 | Categorical data analysis. 2nd edition - Agresti - 2002 |

108 | Goodness-of-fit statistics for discrete multivariate data - Read, Cressie - 1988 |

100 |
Categorical Data Analysis. 2nd ed
- Agresti
(Show Context)
Citation Context ...el selection and interpretation. The existence of the MLE is essential for the usual derivation of large-sample chi-square approximations to numerous measures of goodness of fit (Bishop et al., 1975; =-=Agresti, 2002-=-; Cressie and Read, 1988) which are utilized to perform hypothesis tests and, most importantly, are an integral part of model selection. If the distribution of the statistic measuring the goodness of ... |

97 | Polymake: A framework for analyzing convex polytopes - Gawrilow, Joswig - 2000 |

90 | Markov fields and log-linear interaction models for contingency tables,” Ann. of [13] U. Grenander, “A unified approach to pattern analysis - Darroch, Lauritzen, et al. |

72 | Analysis of categorical data by linear models - Grizzle, Starmer, et al. - 1969 |

51 | On the interpretation of χ2from contingency tables, and the calculation of P - Fisher - 1922 |

49 | Simple Models for the Analysis of Association in CrossClassifications Having Ordered Categories - Goodman - 1979 |

47 | Some Aspect of multivariate analysis - Roy - 1957 |

46 | Algebraic Statistics: Computational Commutative Algebra in Statistics - Pistone, Wynn - 2001 |

39 | Mathematical contributions to the theory of evolution. XI. On the influence of natural selection on the variability and correlations of organs - Pearson - 1903 |

38 | The x2 test of goodness of fit - Cochran - 1952 |

38 | Causation, Prediction, and Search (2nd Edition - Spirtes, Glymour, et al. - 2001 |

36 | Some methods for strengthening the common χ2 tests - Cochran - 1954 |

34 |
Maximum likelihood in three-way contingency tables
- Birch
- 1963
(Show Context)
Citation Context ...tence of the MLE” to signify lack of solutions for the maximum likelihood optimization problem, in accordance with a terminology long established in the log-linear model literature (see, for example, =-=Birch, 1963-=-; Fienberg and Gilbert, 1970; Haberman, 1974). Alternatively, we can say that the MLE of the cell mean vector does not exist whenever there is no strictly positive solution to the MLE defining equatio... |

31 | The analysis of cross-classified data having ordered and/or unordered catecories:association models, correlation models, and asymmetry models for contingency tables with or without missing entries. The Annals of Statistics - Goodman - 1985 |

29 | The multivariate analysis of qualitative data: Interactions among multiple classifications - Goodman - 1970 |

23 | Markov bases of three-way tables are arbitrarily complicated - Loera, Onn |

23 | An iterative procedure for estimation in contingency tables - Fienberg - 1970 |

21 | Collapsibility and response variables in contingency tables. Biometrika - Asmussen, Edwards - 1983 |

21 | Generalized linear models - Hastie, Pregibon - 1992 |

18 | A note on the equivalence of two test criteria for hypotheses in categorical data - Bhapkar - 1966 |

18 | Log-linear models and frequency tables with small expected cell counts,” The Annals of Statistics - Haberman - 1977 |

17 | Central limit theorems for multinomial sums,” The - Morris - 1975 |

15 | On the existence of maximum likelihood estimators for the binomial response models - Silvapulle - 1981 |

14 | On the hypothesis of no ‘interaction’ in a multi-way contingency table - Roy, Kastenbaum - 1956 |

13 | Goodness-of-fit tests for log-linear models in sparse contingency tables - Koehler - 1986 |

11 | Polyhedral conditions for the nonexistence of the mle for hierarchical log-linear models - Eriksson, Fienberg, et al. - 2006 |

11 | An introduction to some non-parametric generalizations of analysis of variance and multivariate analysis - Roy, Mitra - 1956 |

10 | The geometry of a two by two contingency table - Fienberg, Gilbert - 1970 |

10 | Association models and canonical correlation in the analysis of cross-classifications having ordered categories - Goodman - 1981 |

10 | Contribution to the theory of the χ2 test - Neyman - 1949 |

9 | Full contingency tables, logits, and split contingency tables - Bishop - 1969 |

9 | Interactions in multi-factor contingency tables - Darroch - 1962 |

9 | Making the release of confidential data from muti-way tables count. Retrieved September 1, 2004, from http://www.niss.org/dgii/techreports.html Fienberg - Fienberg, Slavkovic - 1998 |

9 | Theory Anticipated - Heyde, Seneta - 1977 |

7 |
Contingency table interactions, Supplement to the
- Bartlett
- 1935
(Show Context)
Citation Context ...e added and subtracted to the table cells in an appropriate order. Norton (1945) extended Bartlett’s results to the case of 2 × 2 × t tables. Figure 1: Bartlett’s representation of a 2 × 2 × 2 table (=-=Bartlett, 1935-=-, page 248). Deming and Stephan (1940) proposed the method of iterative proportional fitting (IPF) for estimating the cell values in a contingency table subject to constraints coming from “known” marg... |

7 |
On the hypotheses of ‘no interaction’ in contingency tables
- Bhapkar, Koch
- 1968
(Show Context)
Citation Context ...Kastenbaum. One of his other Ph.D. students Vasant P. Bhapkar was to follow up on these ideas in a series of papers (e.g., see Bhapkar, 1961, 1966) and also in collaboration with Gary Koch (e.g., see =-=Bhapkar and Koch, 1968-=-). This work led to the paper by Grizzle et al. (1969) and a number of subsequent contributions by Koch and his students and colleagues. 2.3 The Emergence of Log-Linear Models and Methods The 1960s sa... |

6 |
Some tests for categorical data
- Bhapkar
- 1961
(Show Context)
Citation Context ...ions. Roy’s influence ran deeper than the two papers with Mitra and with Kastenbaum. One of his other Ph.D. students Vasant P. Bhapkar was to follow up on these ideas in a series of papers (e.g., see =-=Bhapkar, 1961-=-, 1966) and also in collaboration with Gary Koch (e.g., see Bhapkar and Koch, 1968). This work led to the paper by Grizzle et al. (1969) and a number of subsequent contributions by Koch and his studen... |