Speelman / Heylen / Geeraerts Mixed-Effects Regression Models in Linguistics

1. Auflage 2018
ISBN: 978-3-319-69830-4
Verlag: Springer Nature Switzerland
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 149 Seiten

Reihe: Quantitative Methods in the Humanities and Social Sciences

When data consist of grouped observations or clusters, and there is a risk that measurements within the same group are not independent, group-specific random effects can be added to a regression model in order to account for such within-group associations. Regression models that contain such group-specific random effects are called mixed-effects regression models, or simply mixed models. Mixed models are a versatile tool that can handle both balanced and unbalanced datasets and that can also be applied when several layers of grouping are present in the data; these layers can either be nested or crossed.  In linguistics, as in many other fields, the use of mixed models has gained ground rapidly over the last decade. This methodological evolution enables us to build more sophisticated and arguably more realistic models, but, due to its technical complexity, also introduces new challenges. This volume brings together a number of promising new evolutions in the use of mixed models in linguistics, but also addresses a number of common complications, misunderstandings, and pitfalls. Topics that are covered include the use of huge datasets, dealing with non-linear relations, issues of cross-validation, and issues of model selection and complex random structures. The volume features examples from various subfields in linguistics. The book also provides R code for a wide range of analyses.


Dirk Speelman is associate professor at the department of linguistics at the KU Leuven. Dirk's main research interest lies in the fields of corpus linguistics, computational lexicology and variational linguistics in general. Much of his work focuses on methodology and on the application of statistical and other quantitative methods to the study of language.  Kris Heylen is a research fellow at the research group Quantitative Lexicology and Variational Linguistics at the University of Leuven (KU Leuven, Belgium) and research fellow at the Institute for the Dutch Language (INT, Leiden, The Netherlands). He specialises in the corpus-based, statistical modelling of lexical semantics and lexical variation.  Dirk Geeraerts is professor of linguistics at the University of Leuven, where founded the research unit Quantitative Lexicology and Variational Linguistics. His main research interests involve the overlapping fields of lexical semantics and lexicology, with a specific descriptive interest in social variation, a strong methodological commitment to corpus analysis, and a theoretical background in Cognitive Linguistics.

Autoren/Hrsg.

Speelman, Dirk

Heylen, Kris

Geeraerts, Dirk

Inhaltsverzeichnis

1;Preface;5
2;Contents;6
3;1 Introduction;7
3.1;1 Mixed Models;7
3.2;2 Mixed Models in Linguistics;8
3.3;3 Mixed Models in This Book;10
3.4;4 Software Used in the Book;11
3.5;5 Chapters in This Book;11
3.6;References;15
4;2 Mixed Models with Emphasis on Large Data Sets;17
4.1;1 Introduction;18
4.2;2 Mixed Models;19
4.2.1;2.1 Linear Mixed Models;19
4.2.2;2.2 Generalized Linear Mixed Models;21
4.2.3;2.3 Estimation and Inference;22
4.3;3 Mixed Models in Action: The Leuven Diabetes Project;23
4.3.1;3.1 Interpretation of the Fixed Effects;23
4.3.2;3.2 Tests for Variance Components;25
4.3.3;3.3 Empirical Bayes Estimation;26
4.4;4 Issues with Large Data Sets;29
4.4.1;4.1 The Split-Sample Idea;29
4.4.2;4.2 Examples of How Large Data Sets Can Be Split;31
4.5;5 Concluding Remarks;32
4.6;References;33
5;3 The L2 Impact on Learning L3 Dutch: The L2 Distance Effect;35
5.1;1 Introduction;35
5.2;2 Background;37
5.2.1;2.1 CCREMs with Interrelated Random Effects;37
5.2.2;2.2 Interrelated L1 and L2 Effects;38
5.3;3 Methods;39
5.4;4 Results;41
5.4.1;4.1 Model Comparison;41
5.4.2;4.2 Control Variables;44
5.4.3;4.3 The L2 Distance Effect;44
5.5;5 Discussion and Conclusion;49
5.6;References;51
6;4 Autocorrelated Errors in Experimental Data in the Language Sciences: Some Solutions Offered by Generalized Additive Mixed Models;54
6.1;1 Introduction;54
6.2;2 Generalized Additive Mixed Models;55
6.3;3 Time Series in a Word Naming Task;56
6.4;4 Pitch Contours as Time Series;61
6.5;5 Time Series in EEG Registration;68
6.6;6 Concluding Remarks;73
6.7;References;74
7;5 Border Effects Among Catalan Dialects;75
7.1;1 Introduction;76
7.1.1;1.1 Border Effects;77
7.1.2;1.2 Combining Dialectometry and Social Dialectology;77
7.1.3;1.3 Hypotheses;78
7.2;2 Material;79
7.2.1;2.1 Pronunciation Data;79
7.2.2;2.2 Sociolinguistic Data;80
7.3;3 Methods;81
7.3.1;3.1 Obtaining Pronunciation Distances;81
7.3.2;3.2 Mixed-Effects Regression Modeling;83
7.3.3;3.3 Generalized Additive Mixed-Effects Regression Modeling;85
7.3.3.1;3.3.1 Social and Lexical Variables;86
7.4;4 Results;88
7.4.1;4.1 Demographic Predictors;89
7.4.2;4.2 Predictors Specific to Lexical Identity;92
7.4.3;4.3 Comparison to Individual Linguistic Variables;92
7.5;5 Discussion and Conclusions;97
7.6;References;99
8;6 Evaluating Logistic Mixed-Effects Models of Corpus-Linguistic Data in Light of Lexical Diffusion;102
8.1;1 Mixed-Effects Models in Corpus Linguistics;102
8.2;2 The Challenges of Corpus Data Given Lexical Diffusion;104
8.3;3 Purposes of Model Evaluation;106
8.4;4 The Problem;107
8.5;5 Simulations;108
8.5.1;5.1 Documenting the Problem;108
8.5.2;5.2 The Solution: Using Mixed-Effects Models to Derive Coefficients of the Evaluated Fixed-Effects Models When the Sample is Unbalanced;112
8.6;6 Limitations;115
8.7;7 Conclusion;115
8.8;References;116
9;7 (Non)metonymic Expressions for government in Chinese: A Mixed-Effects Logistic Regression Analysis;120
9.1;1 Introduction;120
9.2;2 Methodology;122
9.2.1;2.1 Data Collection;122
9.2.1.1;2.1.1 Corpus Design;122
9.2.1.2;2.1.2 Potential Expressions for government and Data Retrieval;123
9.2.1.3;2.1.3 Meaning Identification in Contexts;124
9.2.2;2.2 The Variables;129
9.2.2.1;2.2.1 The Response Variable Meto;129
9.2.2.2;2.2.2 The Predictors;129
9.2.2.3;2.2.3 Summary of the Variables;132
9.2.3;2.3 The Mixed-Effects Logistic Regression Model;133
9.2.3.1;2.3.1 The Random Effect: Verb;133
9.2.3.2;2.3.2 Model Selection and Diagnostics;134
9.2.3.3;2.3.3 The Regression Output;134
9.3;3 The General Regression Model for government;136
9.3.1;3.1 General Impact of the Predictors;136
9.3.2;3.2 Specific Influence of Fixed Effects;137
9.3.2.1;3.2.1 The Variable Con_gp and the Variable Style;137
9.3.2.2;3.2.2 The Variable Topic;139
9.3.2.3;3.2.3 The Variable Syn;139
9.3.2.4;3.2.4 The Variable Locus;140
9.3.2.5;3.2.5 The Variable LangVar;140
9.3.3;3.3 The Random-Effect Variable of Verbs;140
9.4;4 The Separate Regression Model for Mainland Chinese government;141
9.4.1;4.1 The Separate Mixed-Effects Model;142
9.4.2;4.2 The Lectal Variation Between Mainland and Taiwan Chinese;143
9.5;5 Summary;145
9.6;References;146