Some results on \(2^{n-k}\) fractional factorial designs and search for minimum aberration designs. (English) Zbl 0770.62063

As is well-known, a \(2^{n-k}\) fractional factorial design is determined by \(k\) generating words which in turn determine the defining contrasts subgroup. Associated with a \(2^{n-k}\) design is the wordlength pattern of the words in the underlying subgroup. The wordlength pattern of a design \(d\) is given by \(w(d)=(A_ 1(d),A_ 2(d),\dots)\), where \(A_ i(d)\) is the number of words of length \(i\). The smallest \(r\) such that \(A_ r(d)\neq 0\) is called the resolution of \(d\). If \(d_ 1\) and \(d_ 2\) are two \(2^{n-k}\) fractional factorial designs and \(r\) is the smallest \(i\) such that \(A_ i(d_ 1)\neq A_ i(d_ 2)\) then \(d_ 1\) has less aberration than \(d_ 2\) if \(A_ r(d_ 1)<A_ r(d_ 2)\). A design has minimum aberration if no other design has less aberration.
Among various results established in the paper the following are important:
(i) An upper bound is given for the length of the longest word in the defining contrasts subgroup; (ii) Minimum abberation \(2^{n-k}\) designs are presented for \(k=5\) and any \(n\); (iii) A method is given to test the equivalence of \(2^{n-k}\) designs; (iv) It is shown that minimum aberration \(2^{n-k}\) designs are unique for \(k\leq 4\).


62K15 Factorial statistical designs
62K05 Optimal statistical designs
90C90 Applications of mathematical programming
Full Text: DOI