Re: [UAI] About Data Analysis

From: Russell Almond (ralmond@ets.org)
Date: Tue Mar 21 2000 - 09:32:11 PST

  • Next message: David L Dowe: "Re: [UAI] Number of Gaussians in a histogram"

              I'm help a friend who's working in analysis
      of business data. The data was acquired from a questionnaire.
      Each question has a set of answers and the the user marks
      one of that. Futhermore, the answers are made in some way
      that the second involves the first, third involves the
      second, and so on.
              I wold like to know if is correct treat this data
      how mutually exclusive. If it's incorrect could any one
      gimme some suggestion.

    It is impossible to answer this question in the abstract. The answer
    really depends on what your friend's objectives are. For many
    questions, he may be interested in conditional or marginal rates for
    which the dependency just doesn't matter. In other cases, you can
    treat it like missing data. Often, ignoring the dependency is not a
    problem; usually, you just over/under estamate the variance; only if
    the question you are asking the data is very complex is it likely to
    make a big difference. Another approach is to test to see if the
    dependency is ignorable (often is) and then proceed with the
    appropriate marginal analysis.

    There is an extensive literature dealing with such items in the
    Statistical Sampling/Survey world. Often a local university with have
    a stat clinic run by graduate students which can help people with such
    problems for free or at a reduced rate. This is probably a good
    source of help. (Hopefully your friend has asked questions which
    contain the answer he is looking for.)

            --Russell Almond
    Educational Testing Service
    Research Statistics Group, 15-T
    Princeton, NJ 08541
    Phone: 609-734-1557 FAX: 609-734-5420
    Email: --almond@acm.org, --ralmond@ets.org
    http://www.stat.washington.edu/bayes/almond/almond.html
    [Remove -- from email addresses]



    This archive was generated by hypermail 2b29 : Tue Mar 21 2000 - 09:34:58 PST