Yazar "Van Lissa, Caspar J." seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • [ X ]
    Öğe
    An AIC-type information criterion evaluating theory-based hypotheses for contingency tables
    (Springer, 2025) Altinisik, Yasin; Hessels, Roy S.; Van Lissa, Caspar J.; Kuiper, Rebecca M.
    Researchers face inevitable difficulties when evaluating theory-based hypotheses in the context of contingency tables. Log-linear models are often insufficient to evaluate such hypotheses, as they do not provide enough information on complex relationships between cell probabilities in many real-life applications. These models are usually used to evaluate the relationships between variables using only equality restrictions between model parameters, while specifying theory-based hypotheses often also requires inequality restrictions. Moreover, high-dimensional contingency tables generally contain low cell counts and/or empty cells, complicating parameter estimation in log-linear models. The presence of many parameters in these models also causes difficulties in interpretation when evaluating the hypotheses of interest. This study proposes a method that simplifies evaluating theory-based hypotheses for high-dimensional contingency tables by simultaneously addressing each of the above problems. With this method, theory-based hypotheses, which are specified using equality and/or inequality constraints with respect to (functions of) cell probabilities, are evaluated using an AIC-type information criterion, GORICA. We conduct a simulation study to evaluate the performance of GORICA in the context of contingency tables. Two empirical examples illustrate the use of the method.
  • [ X ]
    Öğe
    Evaluation of Inequality Constrained Hypotheses Using a Generalization of the AIC
    (Amer Psychological Assoc, 2021) Altinisik, Yasin; Van Lissa, Caspar J.; Hoijtink, Herbert; Oldehinkel, Albertine J.; Kuiper, Rebecca M.
    In the social and behavioral sciences, it is often not interesting to evaluate the null hypothesis by means of a p-value. Researchers are often more interested in quantifying the evidence in the data (as opposed to using p-values) with respect to their own expectations represented by equality and/or inequality constrained hypotheses (as opposed to the null hypothesis). This article proposes an Akaike-type information criterion (AIC; Akaike, 1973, 1974) called the generalized order-restricted information criterion approximation (GORICA) that evaluates (in)equality constrained hypotheses under a very broad range of statistical models. The results of five simulation studies provide empirical evidence showing that the performance of the GORICA on selecting the best hypothesis out of a set of (in)equality constrained hypotheses is convincing. To illustrate the use of the GORICA, the expectations of researchers are investigated in a logistic regression, multilevel regression, and structural equation model. Translational Abstract Evaluation of Inequality Constrained Hypotheses Using a Generalization of the AIC: Researchers are interested in evaluating equality and/or inequality constrained hypotheses in the context not only of normal linear models, but also of the families outside of normal linear models using a suitable information criterion. However, the available information criteria in the literature are not capable of evaluating (in) equality constrained hypotheses under such a broad range of statistical models. The main aim of this paper is to close this research gap by proposing a new information criterion named the GORICA which can be utilized to evaluate these hypotheses for generalized linear (mixed) models and structural equation models. The GORICA enables researchers to quantify the evidence in the data for two or more (in) equality constrained hypotheses. Like all the other information criteria, the GORICA has the log likelihood and penalty parts. The superiority of the GORICA over the other information criteria lies behind the use of a simple formula when calculating its log likelihood. We investigated the performance of the GORICA on choosing the true hypothesis out of a set of competing hypotheses using simulation studies for logistic regression, multilevel regression, and structural equation model. The findings in these simulation studies suggest that the GORICA has a convincing performance on choosing the true hypothesis. The use of the GORICA is illustrated for (real) data sets in line with these simulation studies.