Statistical Methodology for the Analysis of Repeated Duration Data in Behavioral Studies Purpose Repeated duration data are frequently used in behavioral studies. Classical linear or log-linear mixed models are often inadequate to analyze such data, because they usually consist of nonnegative and skew-distributed variables. Therefore, we recommend use of a statistical methodology specific to duration data. Method We propose a ... Research Note
Research Note  |   March 15, 2018
Statistical Methodology for the Analysis of Repeated Duration Data in Behavioral Studies
 
Author Affiliations & Notes
  • Frédérique Letué
    Université Grenoble Alpes, CNRS, Grenoble INP, Laboratoire Jean Kuntzmann, France
  • Marie-José Martinez
    Université Grenoble Alpes, CNRS, Grenoble INP, Laboratoire Jean Kuntzmann, France
  • Adeline Samson
    Université Grenoble Alpes, CNRS, Grenoble INP, Laboratoire Jean Kuntzmann, France
  • Anne Vilain
    Université Grenoble Alpes, CNRS, Grenoble INP, GIPSA-Lab, France
  • Coriandre Vilain
    Université Grenoble Alpes, CNRS, Grenoble INP, GIPSA-Lab, France
  • Disclosure: The authors have declared that no competing interests existed at the time of publication.
    Disclosure: The authors have declared that no competing interests existed at the time of publication. ×
  • Correspondence to Marie-José Martinez: marie-jose.martinez@univ-grenoble-alpes.fr
  • Editor-in-Chief: Julie Liss
    Editor-in-Chief: Julie Liss×
  • Editor: Bharath Chandrasekaran
    Editor: Bharath Chandrasekaran×
Article Information
Special Populations / Genetic & Congenital Disorders / Research Issues, Methods & Evidence-Based Practice / Speech, Voice & Prosody / Speech / Research Notes
Research Note   |   March 15, 2018
Statistical Methodology for the Analysis of Repeated Duration Data in Behavioral Studies
Journal of Speech, Language, and Hearing Research, March 2018, Vol. 61, 561-582. doi:10.1044/2017_JSLHR-S-17-0135
History: Received April 13, 2017 , Revised July 19, 2017 , Accepted October 13, 2017
 
Journal of Speech, Language, and Hearing Research, March 2018, Vol. 61, 561-582. doi:10.1044/2017_JSLHR-S-17-0135
History: Received April 13, 2017; Revised July 19, 2017; Accepted October 13, 2017

Purpose Repeated duration data are frequently used in behavioral studies. Classical linear or log-linear mixed models are often inadequate to analyze such data, because they usually consist of nonnegative and skew-distributed variables. Therefore, we recommend use of a statistical methodology specific to duration data.

Method We propose a methodology based on Cox mixed models and written under the R language. This semiparametric model is indeed flexible enough to fit duration data. To compare log-linear and Cox mixed models in terms of goodness-of-fit on real data sets, we also provide a procedure based on simulations and quantile–quantile plots.

Results We present two examples from a data set of speech and gesture interactions, which illustrate the limitations of linear and log-linear mixed models, as compared to Cox models. The linear models are not validated on our data, whereas Cox models are. Moreover, in the second example, the Cox model exhibits a significant effect that the linear model does not.

Conclusions We provide methods to select the best-fitting models for repeated duration data and to compare statistical methodologies. In this study, we show that Cox models are best suited to the analysis of our data set.

Acknowledgment
All of the authors have been supported by the LabEx PERSYVAL-Lab (ANR-11-LABX-0025-01) for this work. This article was developed in the framework of the Grenoble Alpes Data Institute, supported by the French National Research Agency under the investissements d'avenir program (ANR-15-IDEX-02). The authors thank Jeanne Clarke for the data acquisition and Wen Wang for the first statistical analysis with survival models.
Order a Subscription
Pay Per View
Entire Journal of Speech, Language, and Hearing Research content & archive
24-hour access
This Article
24-hour access