The correlation between the true standard deviation (ߪ) and the estimated
viation (ߪො) using the Equation (6.16).
pression is included into the initial tight cluster (ܢା) if it satisfies
wing condition, where ݔ stands for either a control expression or
xpression and the tight cluster box is sized by four standard
s (between െ2ߪො and 2ߪො) from the left to the right of the mean
ht cluster ߤ,
ܢାൌ∪ሼݔ|∀ݔ∈ሾߤെ2ߪො, ߤ2ߪොሿሽ
(6.17)
s way, the initial tight cluster is composed of part or all control
ns and maybe a few case expressions. The rest of expressions,
n be some control expressions and case expressions, are treated
ndidate outliers. The candidate list is denoted by ܢି. They are
ed in an ascending order. As in [Yang and Yang, 2013], each
e outlier is tested one by one in terms of its relationship with the
ter. The one in the candidate list, which is most close to the tight
ox ܢା defined above, is selected for testing at the first. If the
hip is significant, this candidate outlier is merged into the tight
ା and the parameters of the tight cluster are updated. The merged
e outlier is removed from the candidate outlier set ܢି and the test
s. If a candidate outlier cannot be merged into the tight cluster, a
ocess is terminated. At this point, all the rest candidates in ܢି are
s the outliers or the outstanding case expressions for a DEG. To
because the initial candidate list has been ordered. If the first of
ning candidates in ܢି cannot be merged into the tight cluster ܢା,
st of candidates cannot be merged into the tight cluster as well.