ENBIS-7 in Dortmund

24 – 26 September 2007

My abstracts

 

The following abstracts have been accepted for this event:

  • Efficient Conjoint Choice Designs in the Presence of Respondent Heterogeneity

    Authors: Jie Yu, Peter Goos, Martina Vandebroek
    Primary area of focus / application:
    Submitted at 10-Sep-2007 13:20 by
    Accepted
    The authors propose a fast and efficient algorithm for constructing
    $D$-optimal conjoint choice designs for mixed logit models in the
    presence of respondent heterogeneity. With this new algorithm, the
    construction of semi-Bayesian $D$-optimal mixed logit designs with
    large numbers of attributes and attribute levels becomes practically
    feasible. The results from the comparison of eight designs (ranging
    from the simple locally $D$-optimal design for the multinomial logit
    model and the nearly orthogonal design generated by Sawtooth (CBC)
    to the complex semi-Bayesian mixed logit design) across wide ranges
    of parameter values show that the semi-Bayesian mixed logit approach
    outperforms the competing designs not only in terms of estimation
    efficiency but also in terms of prediction accuracy. In particular,
    it was found that semi-Bayesian mixed logit designs constructed with
    large heterogeneity parameters are most robust against the
    misspecification of the values for the mean of the individual-level
    coefficients for making precise estimations
    and predictions.
  • Fitting a Model for Predicting Leaks in Pipes for a Water Supplying Company

    Authors: Lluís Marco-Almagro (Technical University of Catalonia),Lourdes Rodero (Technical University of Catalonia),Xavier Tort-Martorell (Technical University of Catalonia),Lluís Jordan (Aigües de Barcelona),Jordi Molina (Aigües de Barcelona),Ramon Ariño (Ai
    Primary area of focus / application:
    Submitted at 10-Sep-2007 17:09 by
    Accepted
    Aigües de Barcelona is a company that supplies water to Barcelona and other
    surrounding districts. The company contacted UPC to get help in a project to identify
    and characterize variables affecting the distribution of leaks in water supply pipes.
    Although the company already had a model for predicting leaks in pipes, they were not
    sure about its validity.
    The first step was reviewing existing historical data using descriptive statistics and basic
    graphics. Information from this step was considered valuable by company technicians.
    The information gathered was used to decide the locations of a nondestructive test to
    know the current condition of the pipes. The remaining life years was deducted from
    this data. The information from this study, together with historical data from the leaks
    database, was used to develop a list of candidate variables for improving the model in
    the company.

    A loglinear model was then fitted. The new model was able to better explain the number
    of leaks than the previous one. A substantial collateral benefit of the project was
    revealing how materials and diameters affected the number of leaks in the tubes.

    Many important lessons were derived from the project. Among others, the value of
    having an updated and consistent database for actualizing the model, the falling of some
    old myths in the company that weren't based on data, and the validation of some other
    perceptions from technicians.

    The presentation will focus not only in the results, but also in the methodology followed
    and in the relationship with technicians and managers in the company.
  • MEASURING SERVICE QUALITY IN HIGHER EDUCATION SECTOR

    Authors: Ana Brochado and Rui Cunha Marques (Technical University of Lisbon, Lisbon, Portugal)
    Primary area of focus / application:
    Submitted at 12-Sep-2007 10:05 by Ana Brochado
    Accepted
    The purpose of this work is to examine the performance of five alternative measures of service quality in high education sector - SERVQUAL (Service Quality), Weighted SERVQUAL, SERVPERF (Service Preferences), Weighted SERVPERF and HedPERF (Higher Education performance). We aim at determining which instrument had the superior measurement capability. Data were collected by means of a structured questionnaire containing perceptions items enhanced from the SERVPERF and HEdPERF scales and expectation items from the SERVQUAL scale, both modified to fit into the higher education sector.
    The first draft of the questionnaire was subject to a pilot testing through a focus group and an expert evaluation. Data were gathered from a 300 student's sample of a Portuguese university in Lisbon. Scales were compared in terms of unidimensionality, reliability and validity (convergent, discriminant and predictive). Managerial conclusions were also drawn.

    Keywords: Service quality scales, higher education, reliability.
  • FAULT DETECTION IN FEEDBACK (CLOSED-LOOP) CONTROLLED SYSTEMS

    Authors: Murat Caner Testik
    Primary area of focus / application:
    Submitted at 16-Sep-2007 16:13 by
    Accepted
    Feedback control systems do not remove assignable causes of variation but attempt to compensate variations in the output by adjustments to some controllable process variables. In contrast, statistical process control methods are used with an intention to eliminate assignable causes that are the sources of variation. In the following, characteristics of feedback control systems have been discussed for their integration with statistical process control methods. Optimal controllers, arbitrary controllers and different fault types are considered. Detection of different types of faults by monitoring the deviations from target, control actions, and some other statistics are discussed.
  • Automating model selection for predictive modeling using SAS Enterprise Miner

    Authors: Stefan Ahrens, SAS Institute
    Primary area of focus / application:
    Submitted at 20-Sep-2007 10:07 by
    Accepted
    Predictive modeling, as used in the context of analytic customer relationship management or credit scoring requires building stable models that not only fit the training and validation data, but generalize well when presented with new data that were not using during the modeling stage. Statisticians usually employ techniques like cross-validation or jack-knifing to ensure that no over-fitting occurs. However, finding the best model by trial and error is still a time-consuming challenge, as many statistical methods and data mining algorithms require fine-tuning certain parameters.

    Using an example from database marketing, this presentation will demonstrate how SAS Enterprise Miner can be used to run various models with differing parameter settings and automatically select the best performing model for the final scoring task. Model selection can be based on statistical criteria such as misclassification, error or profit/loss (if a profit/loss function is supplied). SAS Enterprise Miner offers a convenient graphical user interface that supports the statistician with all necessary tasks during a data mining project, from sampling and initial data exploration via variable transformation and modeling to model assessment and scoring. Thus, using SAS Enterprise Miner can help statisticians save time during the model building process.
  • Virtual Statistics: Exploiting new web-based tools on the company intranet to improve internal discussion and learning within the user community

    Authors: Ewan Polwart (Fujifilm Imaging Colorants Ltd)
    Primary area of focus / application:
    Submitted at 21-Sep-2007 15:46 by
    Accepted
    Within FFIC we have recently been developing intranet-based tools to
    facilitate and improve discussion and information sharing. Statistics has
    been one of the pioneering areas for the use our new Community Server
    interface. Both blogs and forums have been used to promote discussion,
    learning and dissemination of information. This poster will present a
    background to this approach, detail some examples and share learning from
    the experience.

    Specifics: poster presentation