Re: assess design profiles for experiment

Old Mac User wrote:
On Oct 22, 10:15 am, "Sun" <a...@xxxxxx> wrote:
I can get design files from pre-constructed tables or from software like
SPSS, still I 'd like to check the design file to make sure that the effect
I am interested in are there.

I 'd like to know how do I evaluate a design file, in terms of which effect
it allows to be measured?
I assume the main effects can be assessed by covariance/correlation between
columns, but I do not know to see if interactions also measurable from the
design.

Thanks in advance,
Sun



Sun...

It would be helpful if you would post the essence of what you are
trying to do. In all honesty,
I'd feel uncomfortable making a suggestion concerning your question.
It seems to me that if you have no training whatsoever in DOE, then
getting "design files" from a piece of software is a very risky
venture. The possible correlations between main effects an
interactions can corrupt whatever you are doing. If the data are
expensive... or if you are working on a thesis needed for
graduation... or if there's anything of value at stake here... then
you need to gain understanding of the fundamental characteristics of
"design files". OMU


Thanks Old Mac user for your kind answer and sorry for not being able to reply in time.

I spent some time reading some basic texts about experimental design (I already read something about it before, though not fully grasp the essence), let me elaborate the problem a bit.

I have 8 variables, among them 4 are ordinal(can be thought as factors too) and 4 factor variables, 2x3x3x3x4x4x4x4. I am interested to measure main effects and 2 interactions between variable 5,6 and variable 7,8.

I have a design table at hand which I can look up for pre-constructed profiles but I did not find the particular combination of my variables. So I turned to use softwares that can generate design files such as spss(which gives only main effect design, of what I know), statgraphics and AlgDesign package in R. Based on some search in forums, I think that there maybe no fractional factorial design for this variable combination since it mixs different levels. Than I 'd like to get a optimal design which is efficient and small enough to be feasible in practice.

my question is after I get this optimal design, say, a 64 profiles design, with a D-efficiency of 0.67, how do I check whether this design is able to measure the effects that I am aftering. And if the 0.67 efficiency sufficient to allow the design be used in practice. I have check the correlation between variables and they are slightly correlated.

I hope this somehow gives more information about my question, if further elabration needed, I will provide more details.

Thanks.

Sun
.

Relevant Pages

  • Re: QuartusII 4.2 problem
    ... > design we are getting an error saying that it cannot find the design ... Hi Monica, ... design files either have to reside in the project directory or the directory ...
    (comp.arch.fpga)
  • Hard coded paths in OrCAD
    ... OrCAD has all these hard coded paths embedded in the design files. ... hired an engineer to do the initial design of a board and he worked ...
    (sci.electronics.cad)
  • Re: using Principal Component Analysis to create a quality index
    ... I have a mechanical design characterized by 6 different metrics, ... all have different units and they differ by orders of magnitude. ... then you should use covariances, not correlations, because differences ... A high value in Energy means a good design, and a low value in Mass, ...
    (sci.stat.math)
  • Re: using Principal Component Analysis to create a quality index
    ... I have a mechanical design characterized by 6 different metrics, ... then you should use covariances, not correlations, because differences ... the measures are all logs. ...
    (sci.stat.math)
  • Re: using Principal Component Analysis to create a quality index
    ... I have a mechanical design characterized by 6 different metrics, ... then you should use covariances, not correlations, because differences ... the measures are all logs. ...
    (sci.stat.math)