Читать книгу Generalized Linear Models for Bounded and Limited Quantitative Variables - Michael Smithson - Страница 7

Truth be told, many of us have not given much consideration to the kinds of models presented in this volume, but we should. Many of the variables we analyze are bounded. For example, years of education is bounded at 0, percentages at 0 and 100, and time use between 0 and 24 hours in a day. These are “absolute” bounds. Bounds may also be due to truncation and censorship. A measure of social and economic conservatism that asks respondents to “please indicate the extent to which you feel positive or negative towards” a set of issues, with anchors denoting each end of a 0–100 scale as “completely negative” or “completely positive” is an example from this monograph. There is nothing absolute about 0 and 100 in this scale. When it comes to attitudes towards gun ownership, there is a pileup of responses at each end of the scale.

Generalized Linear Models for Bounded and Limited Quantitative Variables acquaints non-specialists with approaches for modeling bounded variables. The focus is variables having continuous ranges with one or two bounds due to absolute limits, truncation, or censoring. Readers have the opportunity to learn from experts in this area. The authors, Michael Smithson and Yiyun Zhou, have been central to the development of these methods, especially for modeling doubly-bounded variables.

Generalized Linear Models for Bounded and Limited Quantitative Variables is a “second course” on generalized linear models (GLMs) with clear explanations and lots of helpful advice. The volume grounds readers in the analysis of bounded variables, equips them with the understanding and tools needed to model these variables, and introduces readers to recent advancements including the authors’ own work on CDF-quantile GLMs. Although the authors provide a brief review of GLMs in the first chapter, readers would do well to be already familiar with this material. Those needing a tutorial can find one in the just published second edition of Generalized Linear Models, by Jeff Gill and Michelle Torres.

A strength of the volume’s pedagogy is an extensive use of examples based on real data from diverse disciplines. Approaches to singly bounded variables are illustrated with analyses of average days to start a business in various countries of the world, individual response times on a Stroop test, and ambulance arrival times in regions of Wales.

Examples of doubly bounded variables come from judgments of the grammaticality of sentences generated by machine translation, probability judgments in a moral dilemma, family factors and depressive symptoms among university students in China, party affiliation and political and social conservatism, and ratings of features essential to democracy. The data come from administrative records, experiments, surveys, and other sources such as mTurk. There is something for everyone. The data for the examples, along with detailed explanations and software code in R, SAS, and Stata, are in a supplementary website so that readers can reproduce the results reported and discussed in the volume.

With this book as a foundation, analysts now have every reason to incorporate the boundedness of variables in their analyses. Its timing is propitious. Software resources for analyzing bounded variables, especially doubly-bounded variables, are increasingly available in a variety of packages, including R, SAS, and Stata. This volume provides the guidance needed to design these analyses and interpret the results.

—Barbara Entwisle

Series Editor