Talk:Fractional factorial design

Context Needed

This article needs going over and context added. For example, it talks about "case III" and "resolution III, IV, and V" without any indication as to what these terms mean. -- 21:03, 27 January 2007 (UTC)

effect multiplication needs explaining

This concept is crucial to understanding what a defining relation is and how it is derived. But it is not explained (how does D=A*B imply I=ABD?). I have done some tweaks in other ways but may revisit this soon. Feel free to beat me to it. Baccyak4H (Yak!) 17:13, 23 August 2007 (UTC)

Factorial designs using the + and - convention for the two levels make effect multiplication rules intuitive. 830199wk (talk) 04:56, 27 October 2016 (UTC)From a user's perspective, any factor interracting with itself generates 1 (or I in this essay) :

              e.g. if the pattern (total of 4 experiments)  for factor 'd'   is (-,-,+,+), 
       then the effect "product"  dd  = (-,-,+,+)(-,-,+,+)=(+,+,+,+) = I ).   

e.g The defining relation for generator e, where e is confounded with the 4-level interraction,

                               e=abcd 

is I = ee = eabcd , or I = abcde . — Preceding unsigned comment added by 830199wk (talk • contribs) 11:31, 26 October 2016 (UTC)

Major overhaul needed

The problem is not that the article is "too technical for most readers", but that the topic is poorly explained. It is not hard to imagine that it was written by a non-expert. The earlier comments in the Talk section are an indication, but there is much more that one can say:

• A fraction is described as a subset of a full factorial experiment, but there is no illustration of this – that is, a display of a full factorial and of a fraction taken from it.

• Contrary to what is written, the notation l^k-p is not the general notation, but is used for regular fractions.

• The claim that "one rarely encounters l > 2 levels in fractional factorial designs" is wrong. (It is probably true, though, that 2-level designs are more common than others.)

• Resolution is only described for 2-level designs, and is badly explained. The notions of a "generator" and a "relation" are unclear. The link used for relation is inappropriate.

• Finney (1945) should be cited as a reference, not just mentioned, and there should be more citations to the research literature. One external link is dead.

• The term "confounding" should be replaced by "aliasing"; ditto for "confound", "confounded". (The article can include a note about terminology.) The link to Confounding is incorrect, as that article deals with causal inference, not fractional factorial designs. That link can now be replaced by a link to Aliasing (factorial experiments).

• The claim that non-regular designs are "sometimes known as Plackett-Burman designs" is false: There are many other non-regular designs. (And a few Plackett-Burman designs are regular.)

Lesser issues:

• The term "run" has synonyms "treatment combination" and "cell" in the literature. The notation for runs should also be revised or broadened.

There are other issues as well, but the above items may serve as a roadmap for revising this article. I hope to take a crack at it in the near future. Johsebb (talk) 17:58, 22 August 2024 (UTC)