Near-exact distributions for the likelihood ratio test statistic for testing multisample independence - The real and complex cases

Research output: Contribution to journalArticle

Abstract

We consider a generalization of the well-known independence of several groups of variables test, which we designate by multisample independence test of several groups of variables. This new generalization is of great interest whenever we want to test if in different populations, which may follow a multivariate complex or real normal distribution, the Hermitian covariance matrices have the same structure and if there is independence between different groups of variables.We show that the test statistic has the distribution of the product of independent beta random variables; however, the explicit expressions for the probability density and cumulative distribution functions turn out to be very complicated and almost impossible to use in practice. Our objective is to use a breakthrough technique to develop near-exact distributions for the test statistic. These approximations are known to be highly accurate and easy to use, which facilitates and encourages their use in practice. Using a decomposition of the null hypothesis of the test into two null hypotheses we obtain, in a simple way, the likelihood ratio test statistic, the expression of its hth null moment, and the characteristic function of its logarithm. The decomposition of the null hypothesis also induces a factorization on the characteristic function of the logarithm of the test statistic, which enables the development of near-exact distributions. The numerical studies presented highlight the good properties of these approximations and show their great precision. Simulation studies conducted show the good power of the test proposed even for alternatives quite close to the null hypothesis. An example of application of the test is also provided.
Original languageUnknown
Pages (from-to)37-58
JournalJournal of Statistical Theory and Practice
Volume9
Issue number1
DOIs
Publication statusPublished - 1 Jan 2015

Keywords

    Cite this

    @article{a52f8bf5e9094053a8b4d376234c062a,
    title = "Near-exact distributions for the likelihood ratio test statistic for testing multisample independence - The real and complex cases",
    abstract = "We consider a generalization of the well-known independence of several groups of variables test, which we designate by multisample independence test of several groups of variables. This new generalization is of great interest whenever we want to test if in different populations, which may follow a multivariate complex or real normal distribution, the Hermitian covariance matrices have the same structure and if there is independence between different groups of variables.We show that the test statistic has the distribution of the product of independent beta random variables; however, the explicit expressions for the probability density and cumulative distribution functions turn out to be very complicated and almost impossible to use in practice. Our objective is to use a breakthrough technique to develop near-exact distributions for the test statistic. These approximations are known to be highly accurate and easy to use, which facilitates and encourages their use in practice. Using a decomposition of the null hypothesis of the test into two null hypotheses we obtain, in a simple way, the likelihood ratio test statistic, the expression of its hth null moment, and the characteristic function of its logarithm. The decomposition of the null hypothesis also induces a factorization on the characteristic function of the logarithm of the test statistic, which enables the development of near-exact distributions. The numerical studies presented highlight the good properties of these approximations and show their great precision. Simulation studies conducted show the good power of the test proposed even for alternatives quite close to the null hypothesis. An example of application of the test is also provided.",
    keywords = "multivariate complex normal distribution, equality of covariance matrices test, mixtures, generalized near-integer gamma distribution, independence of several groups of variables test",
    author = "Coelho, {Carlos Manuel Agra} and Marques, {Filipe Jos{\'e} Gon{\cc}alves Pereira}",
    note = "Embora publicado (impresso) com data de 2015 foi aceite e publicado on-line em 2014",
    year = "2015",
    month = "1",
    day = "1",
    doi = "10.1080/15598608.2014.915157",
    language = "Unknown",
    volume = "9",
    pages = "37--58",
    journal = "Journal of Statistical Theory and Practice",
    issn = "1559-8608",
    publisher = "Taylor & Francis",
    number = "1",

    }

    TY - JOUR

    T1 - Near-exact distributions for the likelihood ratio test statistic for testing multisample independence - The real and complex cases

    AU - Coelho, Carlos Manuel Agra

    AU - Marques, Filipe José Gonçalves Pereira

    N1 - Embora publicado (impresso) com data de 2015 foi aceite e publicado on-line em 2014

    PY - 2015/1/1

    Y1 - 2015/1/1

    N2 - We consider a generalization of the well-known independence of several groups of variables test, which we designate by multisample independence test of several groups of variables. This new generalization is of great interest whenever we want to test if in different populations, which may follow a multivariate complex or real normal distribution, the Hermitian covariance matrices have the same structure and if there is independence between different groups of variables.We show that the test statistic has the distribution of the product of independent beta random variables; however, the explicit expressions for the probability density and cumulative distribution functions turn out to be very complicated and almost impossible to use in practice. Our objective is to use a breakthrough technique to develop near-exact distributions for the test statistic. These approximations are known to be highly accurate and easy to use, which facilitates and encourages their use in practice. Using a decomposition of the null hypothesis of the test into two null hypotheses we obtain, in a simple way, the likelihood ratio test statistic, the expression of its hth null moment, and the characteristic function of its logarithm. The decomposition of the null hypothesis also induces a factorization on the characteristic function of the logarithm of the test statistic, which enables the development of near-exact distributions. The numerical studies presented highlight the good properties of these approximations and show their great precision. Simulation studies conducted show the good power of the test proposed even for alternatives quite close to the null hypothesis. An example of application of the test is also provided.

    AB - We consider a generalization of the well-known independence of several groups of variables test, which we designate by multisample independence test of several groups of variables. This new generalization is of great interest whenever we want to test if in different populations, which may follow a multivariate complex or real normal distribution, the Hermitian covariance matrices have the same structure and if there is independence between different groups of variables.We show that the test statistic has the distribution of the product of independent beta random variables; however, the explicit expressions for the probability density and cumulative distribution functions turn out to be very complicated and almost impossible to use in practice. Our objective is to use a breakthrough technique to develop near-exact distributions for the test statistic. These approximations are known to be highly accurate and easy to use, which facilitates and encourages their use in practice. Using a decomposition of the null hypothesis of the test into two null hypotheses we obtain, in a simple way, the likelihood ratio test statistic, the expression of its hth null moment, and the characteristic function of its logarithm. The decomposition of the null hypothesis also induces a factorization on the characteristic function of the logarithm of the test statistic, which enables the development of near-exact distributions. The numerical studies presented highlight the good properties of these approximations and show their great precision. Simulation studies conducted show the good power of the test proposed even for alternatives quite close to the null hypothesis. An example of application of the test is also provided.

    KW - multivariate complex normal distribution

    KW - equality of covariance matrices test

    KW - mixtures

    KW - generalized near-integer gamma distribution

    KW - independence of several groups of variables test

    U2 - 10.1080/15598608.2014.915157

    DO - 10.1080/15598608.2014.915157

    M3 - Article

    VL - 9

    SP - 37

    EP - 58

    JO - Journal of Statistical Theory and Practice

    JF - Journal of Statistical Theory and Practice

    SN - 1559-8608

    IS - 1

    ER -