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:

  • Novel Contributions in Dependence Analysis

    Authors: Pier Alda Ferrari (Università degli Studi di Milano), Emanuela Raffinetti (Università degli Studi di Milano)
    Primary area of focus / application: Business
    Keywords: concordance curve, Ranks-based Concordance Index (RCI), mixed variables, ordinal independent variables
    Submitted at 16-Apr-2012 11:56 by Emanuela Raffinetti
    Accepted
    In applied research, many multivariate data sets consist of observations from ordinal rather than continuous variables. In such situations, the study of dependence relationships among the variables represents an interesting issue, since the ordinal nature of the variables could not successfully allow the application of the main standard dependence measures based on a metric scale.
    Recently, some new contributions in dependence analysis have been provided in literature to solve this problem. In this direction, we propose a more extended and meaningful development based on a novel dependence measure, named “Ranks-based Concordance Index” (RCI). This approach is focused on the Y dependent variable and a concordance curve. In more detail, supposing to evaluate the dependence of Y from a set of covariates of ordinal nature by applying a linear regression model, the Lorenz curve of the Y quantitative response variable is built. Then, the concordance curve and the related proper concordance index between the Y original values and the Y original values ordered according to the ranks of the Y linear estimated values are constructed. A statistical interpretation of the aforementioned index in terms of linear dependence relationships between the dependent variable and the set of ordinal independent ones is thus provided. The Lorenz curve and the corresponding RCI in special cases of mixed variables assuming both positive and negative values are also considered.
    The adequacy and robustness of the proposed index are validated by a simulation study and its applicability is evaluated through the analysis of a real dataset.
  • Kernel PLS GLM Regressions

    Authors: Frédéric Bertrand (Université de Strasbourg), Myriam Maumy-Bertrand (Université de Strasbourg), Nicolas Meyer (Université de Strasbourg)
    Primary area of focus / application: Modelling
    Keywords: PLS, GLM, Kernel PLS, bootstrap techniques
    Submitted at 18-Apr-2012 00:34 by Frédéric Bertrand
    Accepted
    10-Sep-2012 12:30 Kernel PLS GLM Regressions
    There are mainly two aims for the plsRglm library written by the authors for the R software (R Development Core Team 2008). The extension of PLS regression to generalized linear models, and for instance to logistic regression models (Bastien and al. 2005), and the need to provide tools to PLS users to deal with incomplete datasets using cross-validation. These models were successfully applied to datasets of various kind: by Bastien and al. (2005) to multiple regression problems linked to the famous mixture dataset of Cornell (Kettaneh-Wold 1992), to generalized linear regression and especially to a study of Bordeaux wine quality thanks to an ordinal logistic regression model. More recently, the authors carried out the analysis of allelotyping data thanks to PLS binary logistic regression models (Meyer and al. 2009) in order to enhance the understanding of mecanisms involved in the evolution of cancers. We now put forward Kernel extensions of PLS GLM regression which are available in the plsRglm library for R.

    References:
    Bastien, Ph., Esposito Vinzi, V., and Tenenhaus, M. (2005). PLS generalised linear regression. Computational Statistics & Data Analysis, 48(1), 17-46.
    Meyer, N., Maumy-Bertrand, M. & Bertrand, F. (2010). Comparaison de variantes de régressions logistiques PLS et de régression PLS sur variables qualitatives : application aux données d’allélotypage. Journal de la Société Française de Statistique, 151, 1-18.
    R Development Core Team (2008). A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. http://www.R-project.org.
    Wold, S., Sjöström, M., & Eriksson, L. (2001). PLS-regression: a basic tool of Chemometrics. Chemometrics and Intelligent Laboratory Systems, 58, 109-130.
  • Improving an Electron Beam Soldering Process by Design of Experiments

    Authors: Pere Grima (Universitat Politècnica de Catalunya, Barcelona), Xavier Tort-Martorell (Universitat Politècnica de Catalunya, Barcelona), Lluís Marco-Almagro (Universitat Politècnica de Catalunya, Barcelona), Miguel Sarachaga (ITP)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: design of experiments, factorial designs, process improvement, aeronautic industry, welding process
    Submitted at 19-Apr-2012 19:49 by Lluis Marco-Almagro
    Accepted (view paper)
    11-Sep-2012 16:30 Improving an Electron Beam Soldering Process by Design of Experiments
    ITP (Industria de Turbopropulsores) is a company located in the Basque Country (north of Spain) dedicated to the production of engines for the aeronautic industry. One critical and expensive part, that has an irregular shape with an interior hole, is welded by an electron beam process. The number of parts that had to be reprocessed and repaired manually was too high. The team in charge conducted several studies to identify good process parameters and although some improvements were obtained the process continued to be unsatisfactory. They recognized the need to a more scientific approach. A sequential experimentation approach was used to learn more about the way the different process parameters affected the welding quality at different zones of the part. Even though there were a lot of restrictions - number of experiments and way to conduct them - and the process showed a very high variability the results were an important knowledge gain on how the process variables affected the welding quality. In the paper we will present the design and results. It is interesting to see how the welding parameters affect in different but related ways the response level and variability at different part locations.
  • Performance Measurement Systems Research in CEE

    Authors: Adriana Rejc Buhovac (University of Ljubljana, Faculty of Economics), Maja Zaman Groff (University of Ljubljana, Faculty of Economics)
    Primary area of focus / application: Business
    Keywords: contemporary performance measurement systems, PMS determinants, Central and Eastern Europe, empirical research
    Submitted at 26-Apr-2012 13:19 by Maja Zaman Groff
    Accepted (view paper)
    10-Sep-2012 17:15 Performance Measurement Systems Research in CEE
    The paper presents a synthesis of contemporary performance measurement systems (PMS) research in Central and Eastern Europe (CEE). The review is organized around four determinants of contemporary PMS: multidimensionality, strategic focus, cascading, and alignment between PMS and compensation. Discussion of findings and future research directions are suggested along these four determinants while taking into account different institutional, legislative, and cultural contexts of the CEE countries. The paper delivers tentative implications for management researchers and management accounting researchers to guide future thinking and research on PMS in CEE countries. Finally, relevance of research findings for managers is discussed.
  • Comparison of Different Forms of Sequential Designs for a Low and High Dimensional Case

    Authors: Koen Rutten (KU Leuven), Josse De Baerdemaeker (KU Leuven), Bart De Ketelaere (KU Leuven)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Design of Experiments, Sequential Design, Optimization, Evolutionary Operation, Steepest Ascent, Simplex
    Submitted at 27-Apr-2012 10:06 by Koen Rutten
    Accepted (view paper)
    10-Sep-2012 17:00 Comparison of Different Forms of Sequential Designs for a Low and High Dimensional Case
    Sequential designs in the optimization for practical industrial cases are well-described for low-dimensional problems, especially up to three factors. In higher dimensions it is often unclear which method to use. This work implements several basic sequential designs, for low-dimensional and high-dimensional simulations to highlight both problems with and opportunities for, their use. It describes the two- and three-factor simulation with basic sequential designs, namely simplex and EVOP, as presented in the previous work of Rutten et al (2011), and the steepest ascent EVOP approach, as explained in Montgomery et al (2009). For an eight-factor simulation the EVOP implementation has also been conducted with fractional factorials as base design. Results based on the simulation studies are presented and discussed. The use of more efficient designs opens a door to tackling the problem of high dimensional, practical problems.
  • Big Data - How Big is Big?

    Authors: Volker Kraft (JMP)
    Primary area of focus / application: Education & Thinking
    Keywords: Big Data, Data Discovery, Desktop Analytics, Architecture (Client, Server, Cloud), Real-World Examples, JMP, SAS, R
    Submitted at 2-May-2012 11:37 by Volker Kraft
    Accepted (view paper)
    10-Sep-2012 15:35 Big Data - How Big is Big?
    The analysis of 'Big Data' is normally associated with large-scale hardware and expensive software. But multi-processor hardware and multi-threaded software are now also ubiquitous in personal computing devices. So, in 2012, just what is possible with a laptop that is within the budget of a serious analyst?

    Most real-world analysis efforts that are given the 'big data' tag start by trying to understand issues of data quality, and by trying to identify interesting and potentially useful relationships and structure. Indeed, in such situations the data is often under scrutiny for the first time, so the investigation is necessarily somewhat open-ended. In addition, there are clear advantages to offering the analyst the capability to do significant data manipulation and data discovery without the need to call on an IT group.

    This presentation shows some real-world examples of what's currently possible using JMP, and positions this within the spectrum that covers the range from analysis of text book examples that are amenable to hand calculation to genuinely large-scale problems that can only be handled with investments in hardware and software of $1m or more.