On quadratic logistic regression models when predictor variables are subject to measurement error

Jakub Stoklosa, Yih-huei Huang, Elise FURLAN, Wen-Han Hwang

    Research output: Contribution to journalArticle

    3 Citations (Scopus)

    Abstract

    Owing to its good properties and a simple model fitting procedure, logistic regression is one of the most commonly used methods applied to data consisting of binary outcomes and one or more predictor variables. However, if the predictor variables are measured with error and the functional relationship between the response and predictor variables is non-linear (e.g., quadratic) then consistent estimation of model parameters is more challenging to develop. To address the effects of measurement error in predictor variables when using quadratic logistic regression models, two novel approaches are developed: (1) an approximated refined regression calibration; and (2) a weighted corrected score method. Both proposed approaches offer several advantages over existing methods in that they are computationally efficient and are straightforward to implement. A simulation study was conducted to evaluate the estimators' finite sample performance. The proposed methods are also applied on real data from a medical study and an ecological application.
    Original languageEnglish
    Pages (from-to)109-121
    Number of pages13
    JournalComputational Statistics and Data Analysis
    Volume95
    DOIs
    Publication statusPublished - 2016

    Fingerprint

    Logistic Regression Model
    Measurement errors
    Measurement Error
    Logistics
    Predictors
    Corrected Score
    Regression Calibration
    Consistent Estimation
    Binary Outcomes
    Functional Relationship
    Model Fitting
    Logistic Regression
    Calibration
    Simulation Study
    Estimator
    Evaluate
    Model

    Cite this

    @article{d81710c819194304b72620f8ae8c6488,
    title = "On quadratic logistic regression models when predictor variables are subject to measurement error",
    abstract = "Owing to its good properties and a simple model fitting procedure, logistic regression is one of the most commonly used methods applied to data consisting of binary outcomes and one or more predictor variables. However, if the predictor variables are measured with error and the functional relationship between the response and predictor variables is non-linear (e.g., quadratic) then consistent estimation of model parameters is more challenging to develop. To address the effects of measurement error in predictor variables when using quadratic logistic regression models, two novel approaches are developed: (1) an approximated refined regression calibration; and (2) a weighted corrected score method. Both proposed approaches offer several advantages over existing methods in that they are computationally efficient and are straightforward to implement. A simulation study was conducted to evaluate the estimators' finite sample performance. The proposed methods are also applied on real data from a medical study and an ecological application.",
    keywords = "quadratic logistic regression",
    author = "Jakub Stoklosa and Yih-huei Huang and Elise FURLAN and Wen-Han Hwang",
    year = "2016",
    doi = "10.1016/j.csda.2015.09.012",
    language = "English",
    volume = "95",
    pages = "109--121",
    journal = "Computational Statistics Data Analysis",
    issn = "0167-9473",
    publisher = "Elsevier",

    }

    On quadratic logistic regression models when predictor variables are subject to measurement error. / Stoklosa, Jakub; Huang, Yih-huei; FURLAN, Elise; Hwang, Wen-Han.

    In: Computational Statistics and Data Analysis, Vol. 95, 2016, p. 109-121.

    Research output: Contribution to journalArticle

    TY - JOUR

    T1 - On quadratic logistic regression models when predictor variables are subject to measurement error

    AU - Stoklosa, Jakub

    AU - Huang, Yih-huei

    AU - FURLAN, Elise

    AU - Hwang, Wen-Han

    PY - 2016

    Y1 - 2016

    N2 - Owing to its good properties and a simple model fitting procedure, logistic regression is one of the most commonly used methods applied to data consisting of binary outcomes and one or more predictor variables. However, if the predictor variables are measured with error and the functional relationship between the response and predictor variables is non-linear (e.g., quadratic) then consistent estimation of model parameters is more challenging to develop. To address the effects of measurement error in predictor variables when using quadratic logistic regression models, two novel approaches are developed: (1) an approximated refined regression calibration; and (2) a weighted corrected score method. Both proposed approaches offer several advantages over existing methods in that they are computationally efficient and are straightforward to implement. A simulation study was conducted to evaluate the estimators' finite sample performance. The proposed methods are also applied on real data from a medical study and an ecological application.

    AB - Owing to its good properties and a simple model fitting procedure, logistic regression is one of the most commonly used methods applied to data consisting of binary outcomes and one or more predictor variables. However, if the predictor variables are measured with error and the functional relationship between the response and predictor variables is non-linear (e.g., quadratic) then consistent estimation of model parameters is more challenging to develop. To address the effects of measurement error in predictor variables when using quadratic logistic regression models, two novel approaches are developed: (1) an approximated refined regression calibration; and (2) a weighted corrected score method. Both proposed approaches offer several advantages over existing methods in that they are computationally efficient and are straightforward to implement. A simulation study was conducted to evaluate the estimators' finite sample performance. The proposed methods are also applied on real data from a medical study and an ecological application.

    KW - quadratic logistic regression

    U2 - 10.1016/j.csda.2015.09.012

    DO - 10.1016/j.csda.2015.09.012

    M3 - Article

    VL - 95

    SP - 109

    EP - 121

    JO - Computational Statistics Data Analysis

    JF - Computational Statistics Data Analysis

    SN - 0167-9473

    ER -