Browsing by Author "Stewart, Connie"

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    A folded model for compositional data analysis
    (Wiley, 2020) Tsagris, Michail; Stewart, Connie
    A folded type model is developed for analysing compositional data. The proposed model involves an extension of the α-transformation for compositional data and provides a new and flexible class of distributions for modelling data defined on the simplex sample space. Despite its rather seemingly complex structure, employment of the EM algorithm guarantees efficient parameter estimation. The model is validated through simulation studies and examples which illustrate that the proposed model performs better in terms of capturing the data structure, when compared to the popular logistic normal distribution, and can be advantageous over a similar model without folding.
  • Thumbnail Image
    Item
    An approach to measure distance between compositional diet estimates containing essential zeros
    (Taylor & Francis, 2016) Stewart, Connie
    For many applications involving compositional data, it is necessary to establish a valid measure of distance, yet when essential zeros are present traditional distance measures are problematic. In quantitative fatty acid signature analysis (QFASA), compositional diet estimates are produced that often contain many zeros. In order to test for a difference in diet between two populations of predators using the QFASA diet estimates, a legitimate measure of distance for use in the test statistic is necessary. Since ecologists using QFASA must first select the potential species of prey in the predator's diet, the chosen measure of distance should be such that the distance between samples does not decrease as the number of species considered increases, a property known in general as subcompositional coherence. In this paper we compare three measures of distance for compositional data capable of handling zeros, but not satisfying some of the well-accepted principles of compositional data analysis. For compositional diet estimates, the most relevant of these is the property of subcompositionally coherence and we show that this property may be approximately satisfied. Based on the results of a simulation study and an application to real-life QFASA diet estimates of grey seals, we recommend the chi-square measure of distance.
  • Thumbnail Image
    Item
    Prey selection for fatty acid signature analysis using the Akaike information criterion
    (University of New Brunswick, 2025-03) Rideout, Tyler; Stephenson, Matthew; Stewart, Connie
    The estimation of marine predator diet compositions through fatty acid signature analysis allows valuable insights into the trophic structures of marine ecosystems by comparing the fatty acid signatures of predators with a library of their potential prey. These prey libraries consist of individual fatty acid signatures grouped by species, and in order to obtain unique diet estimates the number of groups in the library must be less than the number of dietary fatty acids in the analysis. We propose a novel application of the Akaike information criterion to identify the correct subset of species from a wider set of potential prey. The estimation of true zeroes through the removal of species from the prey library may result in reduced variability and greater accuracy in the estimation of non-zero proportions. Outcomes from simulation studies as well as the analysis of real-life grey seal data are used to explore the performance of this method.
  • Thumbnail Image
    Item
    QFASA R Package
    (University of New Brunswick, 2019) Kamerman, Justin; Stewart, Connie
    The primary contribution of this project is to package R source code created to support the fitting and evaluation of Quantitative Fatty Acid Signature Analysis (QFASA) models into a FOSS module available on CRAN. The existing code is widely used by the QFASA community but is inconsistently documented and maintained. This makes it difficult for new users to get up to speed with new and current QFASA methodologies, and to distribute code fixes and improvements. Creating an R package is a well-defined process and encourages the use of software engineering best practices and the production of well-documented modules that are easy to install and maintain. This report describes the process of diet estimation via the QFASA methodology and reviews some of the underlying statistical methodologies. We detail the R packaging process and our interaction with CRAN to publish the package, and our implementation of parallel computing methods to improve the speed and efficiency of model inference by making use of multi-core processors. Finally, for comparison, we review a similar QFASA module, qfasar, which was released subsequently.
  • Thumbnail Image
    Item
    Regression for compositional predictor data
    (University of New Brunswick, 2022-10) Huang, Zhenduo; Stewart, Connie; Tsagris, Michael
    Compositional data refer to proportions of a whole. This type of data arise in many different disciplines, such as geology, biology, economics, and sociology, and require nontraditional methods for their analysis. In this report we consider regression methods for compositional data and introduce two new regression methods for compositional predictor data. The first method is unconstrained log-contrast (ULC) regression which is a less restrictive form log-contrast (LC) regression. The second proposed method is Greenacre’s transformation regression which is an extension of ULC that allows for zeros in the compositional data. In this report we are interested in evaluating an F-test, in terms of its type I error rate and power, to compare LC regression with ULC regression, for various sample sizes and composition dimensions. Lastly, we used cross validation to assess the predictive performance of two new methods on three real-life datasets.
  • Thumbnail Image
    Item
    Simultaneous maximum unified fatty acid signature analysis
    (University of New Brunswick, 2022-01) McNichol, Jennifer; Stewart, Connie
    Quantitative Fatty Acid Signature Analysis (QFASA) has been the cornerstone of dietary estimation for marine predators since its introduction in 2004. However, QFASA relies upon calibration coefficients (CCs) to account for the differences in fatty acids (FAs) between a predator and its prey. CCs are determined by way of captive feeding studies and must be uniquely determined for each species of predator, creating a major limitation for QFASA. One recent approach proposed expanding QFASA to simultaneously estimate diet and CCs, though it has not been thoroughly tested. Another takes a maximum likelihood approach to QFASA which has shown promising results but still relies on predetermined CCs. In this thesis we take inspiration from both of these approaches to develop a maximum likelihood model to estimate both diet and CCs. In addition to two real life applications, a simulation study is conducted to evaluate our model in comparison to existing models.
  • Thumbnail Image
    Item
    Statistical analysis of bioassay data with dependent replicates
    (University of New Brunswick, 2025-03) Cann, Liam; Stewart, Connie; Stephenson, Matthew
    Bioassays play a key role in ensuring that every batch of drug produced is safe and effective for release. They play a critical role in testing the potency of biologic drugs, such as vaccines or monoclonal antibodies. Due to the importance of bioassays in ensuring a safe and effective product, it is critical that the statistical methods used are appropriate. However, it is the current common practice to treat the replicate responses at each dose group as if they are independent despite the fact they are often correlated. In this research, we look at quantitatively assessing the risks of the conventional analysis methods using a simulation study to investigate the impact of correlation on the statistical analysis of bioassays. Specifically, we consider parallelism assessment, model goodness-of-fit, and relative potency estimation. We also make recommendations within the constraints of the current available commercial bioassay analysis software to provide valid statistical inference.