ENBIS-12 in Ljubljana

9 – 13 September 2012 Abstract submission: 15 January – 10 May 2012

My abstracts


The following abstracts have been accepted for this event:

  • Introducing the COM-Poisson

    Authors: Galit Shmueli (Indian School of Business)
    Primary area of focus / application: Other: SPECIAL SESSION
    Keywords: COM-Poisson, count data, regression, exponential family
    Submitted at 3-May-2012 14:23 by Galit Shmueli
    11-Sep-2012 11:30 Introducing the COM-Poisson
    The Poisson distribution is popular for modeling count data, yet it is constrained by its equi-dispersion assumption, making it less than ideal for modeling real data that often exhibit over- or under-dispersion. The COM-Poisson distribution is a two-parameter generalization of the Poisson distribution that allows for a wide range of over- and under-dispersion. It generalizes the Poisson distribution as well as the Bernoulli and geometric distributions. Its flexibility and special properties has prompted a fast growth of methodological and applied research in various fields. This talk will introduce the main properties of the COM-Poisson distribution and regression models, and present a brief overview of methodological directions that have been taken and areas of applications.
  • Statistical Advancements Using the COM-Poisson Distribution

    Authors: Kimberly Sellers (Georgetown University)
    Primary area of focus / application: Other: SPECIAL SESSION
    Keywords: Conway-Maxwell-Poisson, regression, control chart, overdispersion, underdispersion
    Submitted at 3-May-2012 18:54 by Kimberly Sellers
    11-Sep-2012 11:50 Statistical Advancements Using the COM-Poisson Distribution
    Several significant methodological contributions stem from the COM-Poisson distribution. Two that will be featured in this talk are the development of a COM-Poisson regression, and a COM-Poisson control chart. The COM-Poisson regression generalizes the well-known Poisson and logistic regression models and further bridges the gap between these regression models, allowing for general consideration of count data regression in light of any form of data (over- or under-) dispersion. Meanwhile, COM-Poisson control charts are flexible control charts developed that encompass the classical Shewart charts based on the Poisson, Bernoulli (or binomial), and geometric (or negative binomial) distributions. Both statistical methods rely on the underlying COM-Poisson properties to circumvent the constraining assumptions associated with the corresponding classical count methods for data analysis. Examples applying these methods to business and industrial data will be discussed to demonstrate the significance of this work, and the valued flexibility these methods allow in data analysis.
  • Discussion in the Session "The COM-Poisson Model for Count Data: Methods and Applications"

    Authors: Werner G. Müller (Johannes Kepler University Linz)
    Primary area of focus / application: Other: SPECIAL ISBIS SESSION
    Keywords: copula, COM-Poisson, overdispersion, underdispersion, soccer, football
    Submitted at 4-May-2012 11:09 by Werner G. Mueller
    11-Sep-2012 12:40 Discussion in the Session "The COM-Poisson Model for Count Data: Methods and Applications"
    I will provide a discussion of the presentations by Galit Shmueli (Indian School of Business), Kimberly Sellers (Georgetown University), and Sharad Borle (Rice University). Eventually I will report abot a study, where a copula model with COM-Poisson margins was used to predict football scores.
  • Compositional T^2 Control Chart: Interpretation of Out-of-control Signals

    Authors: Marina Vives-Mestres (Universitat de Girona), Josep Daunis-i-Estadella (Universitat de Girona), Josep-Antoni Martín-Fernández (Universitat de Girona)
    Primary area of focus / application: Process
    Keywords: Multivariate control chart, Statistical process control, Log-ratio methodology, Compositional data
    Submitted at 7-May-2012 14:40 by Marina Vives-Mestres
    Accepted (view paper)
    10-Sep-2012 11:50 Compositional T^2 Control Chart: Interpretation of Out-of-control Signals
    Hotelling T^2 control chart is one of the most familiar multivariate statistical process control tools which take into account both the univariate and interrelationship effects between variables.
    Compositions are vectors of positive elements describing quantitatively the parts of some whole, which carry exclusively relative information between the parts. Their sample space is the simplex and specific methods are necessary to deal with their restriction of constant sum.
    In a control chart scenario it is not possible to calculate the T^2 statistic of a raw composition because of the singularity of the covariance matrix. One classical approach solution to solve this difficulty is to eliminate one component. Another typical solution is to apply PCA and afterwards calculate the T^2. Nevertheless, each solution generates new serious troubles in the analysis and in the interpretation of their results. In addition, both obtained limits are not coherent with the compositional nature of the data.
    In this work it is introduced a new approach based on the log-ratio methodology which provides a powerful framework to deal with this type of data. We show that it is fully consistent to applying the classical methodology the compositions expressed in coordinates with respect to an orthonormal basis.
    Classical and log-ratio approaches are compared in terms of their Run Length performance and through a practical example from the industry. Moreover, some advice is given (in terms of choosing an appropriate the basis) for easy interpretation of out of control signals based on the decomposition of the T^2 statistic.
  • Some Tracks for Extending the COM-Poisson

    Authors: Célestin Kokonendji (University of Franche-Comté)
    Primary area of focus / application: Other: SPECIAL SESSION
    Keywords: Count distribution, Duality, Multivariate discrete distribution, Weighted Poisson
    Submitted at 8-May-2012 18:02 by Célestin Kokonendji
    11-Sep-2012 12:30 Some Tracks for Extending the COM-Poisson
    The COM-Poisson model is investigated by several authors these last years for modeling count data. This family of univariate count distributions is more flexible for practical use and admits special properties such duality between under and over dispersion with respect to the Poisson distribution. As a particular case of weighted Poisson distributions this talk will present some new directions of theoretical research of the COM-Poisson, and also propose various areas of applications.
  • Imputation Methods in the Estimation of ARMA Models with Missing Observations

    Authors: Korneel Bernaert (Vrije Universiteit Brussel)
    Primary area of focus / application: Economics
    Keywords: ARMA, missing data, estimation, imputation
    Submitted at 9-May-2012 10:06 by Korneel Bernaert
    12-Sep-2012 10:25 Imputation Methods in the Estimation of ARMA Models with Missing Observations
    Missing observations in time series analysis pose difficulties in estimating models because traditional methods no longer apply. Several methods for dealing with missing data have already been developed. Velicer and Colby showed that ad hoc methods such as mean imputation often don’t result in unbiased estimators. This paper starts out by proposing an adapted Yule-Walker estimator to get unbiased results from mean imputation. We then continue discussing two existing methods for estimating ARMA models, the Kalman filter proposed by Jones (1980) and the additive outlier approach proposed by Pena (1987) and Ljung (1989). Using the Kalman filter is not based on imputation but skips the missing values in the estimation, the downside to this approach is that it relies on the assumption that the observations are missing completely at random which might not always be the case. The additive outlier approach imputes random variables and proceeds estimating the coefficients treating the imputed values as additive outliers. This paper suggests a third alternative using multiple imputation. An advantage of the imputation based methods is their ability to account for more information about the missing data generating mechanism. We also perform a simulation study to compare and assess the properties of the four methods for dealing with missing observations.