Mokken scale analysis of mental health and wellbeing questionnaire item responses

a nonparametric IRT method in empirical research for applied health researchers

Jan Stochl (Lead / Corresponding author), Peter B. Jones, Tim J. Croudace

    Research output: Contribution to journalArticle

    73 Citations (Scopus)

    Abstract

    ABSTRACT: BACKGROUND: Mokken scaling techniques are a useful tool for researchers who wish to construct unidimensional tests or use questionnaires that comprise multiple binary or polytomous items. The stochastic cumulative scaling model offered by this approach is ideally suited when the intention is to score an underlying latent trait by simple addition of the item response values. In our experience, the Mokken model appears to be less well-known than for example the (related) Rasch model, but is seeing increasing use in contemporary clinical research and public health. Mokken's method is a generalisation of Guttman scaling that can assist in the determination of the dimensionality of tests or scales, and enables consideration of reliability, without reliance on Cronbach's alpha. This paper provides a practical guide to the application and interpretation of this non-parametric item response theory method in empirical research with health and well-being questionnaires. METHODS: Scalability of data from 1) a cross-sectional health survey (the Scottish Health Education Population Survey) and 2) a general population birth cohort study (the National Child Development Study) illustrate the method and modeling steps for dichotomous and polytomous items respectively. The questionnaire data analyzed comprise responses to the 12 item General Health Questionnaire, under the binary recoding recommended for screening applications, and the ordinal/polytomous responses to the Warwick-Edinburgh Mental Well-being Scale. RESULTS AND CONCLUSIONS: After an initial analysis example in which we select items by phrasing (six positive versus six negatively worded items) we show that all items from the 12-item General Health Questionnaire (GHQ-12) - when binary scored - were scalable according to the double monotonicity model, in two short scales comprising six items each (Bech's "well-being" and "distress" clinical scales). An illustration of ordinal item analysis confirmed that all 14 positively worded items of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) met criteria for the monotone homogeneity model but four items violated double monotonicity with respect to a single underlying dimension. Software availability and commands used to specify unidimensionality and reliability analysis and graphical displays for diagnosing monotone homogeneity and double monotonicity are discussed, with an emphasis on current implementations in freeware.
    Original languageEnglish
    Article number74
    Number of pages16
    JournalBMC Medical Research Methodology
    Volume12
    DOIs
    Publication statusPublished - Jun 2012

    Fingerprint

    Empirical Research
    Mental Health
    Research Personnel
    Health
    Child Development
    Health Surveys
    Health Education
    Population
    Surveys and Questionnaires
    Cohort Studies
    Software
    Public Health
    Cross-Sectional Studies
    Parturition
    Research

    Keywords

    • Epidemiology
    • Health Informatics

    Cite this

    @article{e8d5631ec75642658dd63085ea0e7f98,
    title = "Mokken scale analysis of mental health and wellbeing questionnaire item responses: a nonparametric IRT method in empirical research for applied health researchers",
    abstract = "ABSTRACT: BACKGROUND: Mokken scaling techniques are a useful tool for researchers who wish to construct unidimensional tests or use questionnaires that comprise multiple binary or polytomous items. The stochastic cumulative scaling model offered by this approach is ideally suited when the intention is to score an underlying latent trait by simple addition of the item response values. In our experience, the Mokken model appears to be less well-known than for example the (related) Rasch model, but is seeing increasing use in contemporary clinical research and public health. Mokken's method is a generalisation of Guttman scaling that can assist in the determination of the dimensionality of tests or scales, and enables consideration of reliability, without reliance on Cronbach's alpha. This paper provides a practical guide to the application and interpretation of this non-parametric item response theory method in empirical research with health and well-being questionnaires. METHODS: Scalability of data from 1) a cross-sectional health survey (the Scottish Health Education Population Survey) and 2) a general population birth cohort study (the National Child Development Study) illustrate the method and modeling steps for dichotomous and polytomous items respectively. The questionnaire data analyzed comprise responses to the 12 item General Health Questionnaire, under the binary recoding recommended for screening applications, and the ordinal/polytomous responses to the Warwick-Edinburgh Mental Well-being Scale. RESULTS AND CONCLUSIONS: After an initial analysis example in which we select items by phrasing (six positive versus six negatively worded items) we show that all items from the 12-item General Health Questionnaire (GHQ-12) - when binary scored - were scalable according to the double monotonicity model, in two short scales comprising six items each (Bech's {"}well-being{"} and {"}distress{"} clinical scales). An illustration of ordinal item analysis confirmed that all 14 positively worded items of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) met criteria for the monotone homogeneity model but four items violated double monotonicity with respect to a single underlying dimension. Software availability and commands used to specify unidimensionality and reliability analysis and graphical displays for diagnosing monotone homogeneity and double monotonicity are discussed, with an emphasis on current implementations in freeware.",
    keywords = "Epidemiology, Health Informatics",
    author = "Jan Stochl and Jones, {Peter B.} and Croudace, {Tim J.}",
    note = "M1 - Article",
    year = "2012",
    month = "6",
    doi = "10.1186/1471-2288-12-74",
    language = "English",
    volume = "12",
    journal = "BMC Medical Research Methodology",
    issn = "1471-2288",
    publisher = "Springer Verlag",

    }

    TY - JOUR

    T1 - Mokken scale analysis of mental health and wellbeing questionnaire item responses

    T2 - a nonparametric IRT method in empirical research for applied health researchers

    AU - Stochl, Jan

    AU - Jones, Peter B.

    AU - Croudace, Tim J.

    N1 - M1 - Article

    PY - 2012/6

    Y1 - 2012/6

    N2 - ABSTRACT: BACKGROUND: Mokken scaling techniques are a useful tool for researchers who wish to construct unidimensional tests or use questionnaires that comprise multiple binary or polytomous items. The stochastic cumulative scaling model offered by this approach is ideally suited when the intention is to score an underlying latent trait by simple addition of the item response values. In our experience, the Mokken model appears to be less well-known than for example the (related) Rasch model, but is seeing increasing use in contemporary clinical research and public health. Mokken's method is a generalisation of Guttman scaling that can assist in the determination of the dimensionality of tests or scales, and enables consideration of reliability, without reliance on Cronbach's alpha. This paper provides a practical guide to the application and interpretation of this non-parametric item response theory method in empirical research with health and well-being questionnaires. METHODS: Scalability of data from 1) a cross-sectional health survey (the Scottish Health Education Population Survey) and 2) a general population birth cohort study (the National Child Development Study) illustrate the method and modeling steps for dichotomous and polytomous items respectively. The questionnaire data analyzed comprise responses to the 12 item General Health Questionnaire, under the binary recoding recommended for screening applications, and the ordinal/polytomous responses to the Warwick-Edinburgh Mental Well-being Scale. RESULTS AND CONCLUSIONS: After an initial analysis example in which we select items by phrasing (six positive versus six negatively worded items) we show that all items from the 12-item General Health Questionnaire (GHQ-12) - when binary scored - were scalable according to the double monotonicity model, in two short scales comprising six items each (Bech's "well-being" and "distress" clinical scales). An illustration of ordinal item analysis confirmed that all 14 positively worded items of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) met criteria for the monotone homogeneity model but four items violated double monotonicity with respect to a single underlying dimension. Software availability and commands used to specify unidimensionality and reliability analysis and graphical displays for diagnosing monotone homogeneity and double monotonicity are discussed, with an emphasis on current implementations in freeware.

    AB - ABSTRACT: BACKGROUND: Mokken scaling techniques are a useful tool for researchers who wish to construct unidimensional tests or use questionnaires that comprise multiple binary or polytomous items. The stochastic cumulative scaling model offered by this approach is ideally suited when the intention is to score an underlying latent trait by simple addition of the item response values. In our experience, the Mokken model appears to be less well-known than for example the (related) Rasch model, but is seeing increasing use in contemporary clinical research and public health. Mokken's method is a generalisation of Guttman scaling that can assist in the determination of the dimensionality of tests or scales, and enables consideration of reliability, without reliance on Cronbach's alpha. This paper provides a practical guide to the application and interpretation of this non-parametric item response theory method in empirical research with health and well-being questionnaires. METHODS: Scalability of data from 1) a cross-sectional health survey (the Scottish Health Education Population Survey) and 2) a general population birth cohort study (the National Child Development Study) illustrate the method and modeling steps for dichotomous and polytomous items respectively. The questionnaire data analyzed comprise responses to the 12 item General Health Questionnaire, under the binary recoding recommended for screening applications, and the ordinal/polytomous responses to the Warwick-Edinburgh Mental Well-being Scale. RESULTS AND CONCLUSIONS: After an initial analysis example in which we select items by phrasing (six positive versus six negatively worded items) we show that all items from the 12-item General Health Questionnaire (GHQ-12) - when binary scored - were scalable according to the double monotonicity model, in two short scales comprising six items each (Bech's "well-being" and "distress" clinical scales). An illustration of ordinal item analysis confirmed that all 14 positively worded items of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) met criteria for the monotone homogeneity model but four items violated double monotonicity with respect to a single underlying dimension. Software availability and commands used to specify unidimensionality and reliability analysis and graphical displays for diagnosing monotone homogeneity and double monotonicity are discussed, with an emphasis on current implementations in freeware.

    KW - Epidemiology

    KW - Health Informatics

    U2 - 10.1186/1471-2288-12-74

    DO - 10.1186/1471-2288-12-74

    M3 - Article

    VL - 12

    JO - BMC Medical Research Methodology

    JF - BMC Medical Research Methodology

    SN - 1471-2288

    M1 - 74

    ER -