Loading ...
Sorry, an error occurred while loading the content.

494Re: Multivariate testing using the Taguchi Method

Expand Messages
  • matthewjncroche
    Sep 30, 2004
      Stephen,

      I would be curious what you would set as a threshold for "low traffic"

      Multivariate techniques are definitely not a buzzword or a flavor of
      a month, but are in fact crucial.

      Even if a company has very high traffic, there are factors that
      demand fractional factorial support:
      1. While a "home page", application form page, or cart page on
      certain sites may get a lot of traffic; category pages, calculators,
      and product pages will get substantially fewer.
      2. Less traffic also means shorter run. The duration of a test in a
      marketing environment introduces "noise" that is not as relevant a
      factor in manufacturing. We have run tests that ran into "March
      Madness" for shoes, Valentines Day for flowers, and the "Friends"
      final episode for a board game. More days of testing can mean less
      useful results.
      3. Less traffic means segment-based testing is possible. We have run
      effective tests to just Google visitors or just for new visitors.
      4. Finally, multivariate testing is best when it is done in cycles -
      find an element that contributes, then create another experiment to
      test variations on that element. Less traffic, less time, more
      cycles.

      Matthew
      http://www.offermatica.com

      --- In webanalytics@yahoogroups.com, "Stephen Turner" <yahoo@a...>
      wrote:
      > --- In webanalytics@yahoogroups.com, "Blair Gorman" <blair@i...>
      > wrote:
      > > Hi all,
      > >
      > > Wow, I just found this group, and as a somewhat fanatical tester,
      > I'm very
      > > pleased to be here.
      > >
      > > Just letting you know I have released a free multivariate
      analyzer,
      > based
      > > upon the workings of the Taguchi Method.
      > >
      > > In essence, this allows you to test various elements
      simultaneously
      > - much
      > > more effective than simple A/B split testing.
      > >
      >
      > Blair,
      >
      > Did you read the earlier messages in this group about the Taguchi
      > method? We've had a couple of interesting threads about it. Try
      > searching the archives for "Taguchi".
      >
      > I have argued that the Taguchi method is not useful except for low
      > traffic sites, or at least low traffic pages, and that this is the
      > rare case. For manufacturing engineering, which is where the Taguchi
      > method originated, it can be expensive to produce new experiments,
      but
      > for most websites, visitors are plentiful. So I suspect that for
      most
      > website experiments, it's easier and more powerful to do a full
      > factorial experiment.
      >
      > Having said that, it's nice to see some software to conduct a
      Taguchi
      > experiment where it is the right design. I just suspect that sites
      > will start to conduct Taguchi experiments because they've heard it's
      > the way to tackle this problem, even though a design which is
      simpler
      > but doesn't have as good a buzzword is preferable for them.
      >
      > --
      > Stephen Turner
      > CTO, ClickTracks http://www.clicktracks.com/
      > ClickTracks wins ClickZ Marketing Excellence Award again!
      > WINNER: Best Web Analytics Tool 2003 & 2004
    • Show all 7 messages in this topic