Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

Re: [Phys-l] Vary-One-Variable Protocols (was PsychologistsRespond etc...)


Actually, the Taguchi methods typically ignores interactions. A typical experiment might be as follows. You have 6 factors (which we will call A-F) that you want to test - each of which can take on one of two values (called "1" and "2") A "one factor at a time" design would look like this - a single "control" experiment and 6 variations:

A B C D E F
1 1 1 1 1 1
1 1 1 1 1 2
1 1 1 1 2 1
1 1 1 2 1 1
1 1 2 1 1 1
1 2 1 1 1 1
2 1 1 1 1 1



A typical Taguichi experiment would look like this - 8 trials where several things are changed each time:

A B C D E F
1 1 1 1 1 1
1 1 1 2 2 2
1 2 2 1 1 2
1 2 2 2 2 1
2 1 2 1 2 1
2 1 2 2 1 2
2 2 1 1 2 2
2 2 1 2 1 1

Note in particular that each factor takes on each value 1/2 the time. If you want to see if factor A is important, compare the 1st 4 trials to the last four trials . To test Factor B, compare (1,2,5,6) with (3,4,7,8), etc. This typically is more robust than comparing just trial 1 to trial 3 to test Factor B.

There are other Taguchi designs if the factors can have more than 2 values.

If you want to check interactions (for example, Factor B might be important only when Factor A has a value "2") then you need more trials. There are other techniques that are designed to explore such interactions.


Tim Folkerts



-----Original Message-----
From: phys-l-bounces@carnot.physics.buffalo.edu on behalf of E Muehleisen
Sent: Wed 11/21/2007 1:56 PM
To: Forum for Physics Educators
Subject: Re: [Phys-l] Vary-One-Variable Protocols (was PsychologistsRespond etc...)


A powerful tool in the QC arsenal for analyzing the performance of production systems is called the Taguchi Method. It is a multivariable analysis technique that bypasses the single variable technique by reducing the number of experiments whilst also showing variable interdependence. This method is especially strong since it can show how variables interact.

E.C. Muehleisen

Too much of a good thing is wonderful!

Date: Mon, 19 Nov 2007 18:46:57 -0600
To: phys-l@carnot.physics.buffalo.edu
From: betwys1@sbcglobal.net
Subject: [Phys-l] Vary-One-Variable Protocols (was Psychologists Respond etc...)

At 02:48 PM 11/19/2007, Richard H., you wrote:

... we not only reemphasize the importance of randomized,
controlled experimental tests of competing instructional procedures,
but also indicate that altering one variable at a time is an
essential feature of a properly controlled experiment. ...

The familiar "vary one variable" approach to experiment,
is now in particular situations, substituted by
multivariate experimentation.

If all confounding factors cannot be specified in advance,
methods of varying several factors at a time can extract
useful experimental responses due to these factors, from other,
perhaps unknown or unrecognized variables.

I sometimes think of this as the spatial or parallel approach to
randomizing the protocol, as opposed to the serial randomization
which is familiar.


Brian Whatcott Altus OK Eureka!

_______________________________________________
Forum for Physics Educators
Phys-l@carnot.physics.buffalo.edu
https://carnot.physics.buffalo.edu/mailman/listinfo/phys-l

_________________________________________________________________
Share life as it happens with the new Windows Live.Download today it's FREE!
http://www.windowslive.com/share.html?ocid=TXT_TAGLM_Wave2_sharelife_112007
_______________________________________________
Forum for Physics Educators
Phys-l@carnot.physics.buffalo.edu
https://carnot.physics.buffalo.edu/mailman/listinfo/phys-l