Cluster analysis was used to develo

Cluster analysis was used to develop a taxonomy based on the constructs for lean
and agile supply chain characteristics. We followed several rules similar to those
used in Kathuria (2000) and Frohlich and Dixon (2001) to select the number of
clusters. As in Frohlich and Dixon (2001), we used a hierarchical cluster analysis
to generate a hierarchical dendogram and an agglomeration schedule table. Our
goal was to balance parsimony or few clusters with accuracy (more clusters retains
more data). The choice of the final number of clusters is a subjective one that is
generally guided by a few rules of thumb. Lehmann (1979) indicates the number
of clusters should lie between n/30 and n/60. Because our sample size is 604,
this would suggest a solution with between 10 and 20 clusters. Unfortunately,
this high number of clusters makes managerial interpretability difficult due to
the lack of parsimony. We employed K-mean cluster analysis to generate three-,
four-, and five-cluster solutions. After a careful examination of the three-, four-
, and five-cluster solutions, we decided that the four-cluster solution shown in
Table 4 provided the best interpretability. The five-cluster group had two clusters
that were very similar in that they were heavily focused on lean with very little
emphasis on agile. The three-cluster choice did not have the group that focused
primarily on agile to the exclusion of lean. We feel that the four-cluster choice
provides more explanatory power than the choice of three clusters without the
added complexity of a five-cluster solution, thus we chose four clusters, which is
a common number in the literature (Roth & Miller, 1992; Boyer, Ward, & Leong,
1996). We prefer the four-cluster solution because it provides more explanatory
power—in fact we believe the difference between Lean/agile and Traditional to be very important.

結果 (繁體中文) 2:

聚類分析被用於開發基於構建精益分類法
和敏捷供應鏈的特點。我們遵循類似於多個規則
中Kathuria（2000）和弗羅利希和狄克遜（2001）用於選擇的數量
的群集。如在弗羅利希和狄克遜（2001），我們使用了聚類分析
，以產生一個分層dendogram和附聚調度表。我們的
目標是要平衡簡約性或幾簇精度（多個群集保留
更多的數據）。簇的音響最終數目的選擇是一個主觀一個是
通常遵循拇指的一些規則。萊曼（1979）表示數字
集群應該N / 30和n / 60之間。因為我們的樣本大小是604，
這將表明與10至20簇的溶液。不幸的是，
這種高數集群管理，使DIF解釋性邪教科幻由於
缺乏簡約性。我們採用K均值聚類分析，生成三，
四和網絡VE-集群解決方案。經過仔細檢查三中，四
和網絡VE-集群解決方案，我們決定在所示的四個集群解決方案
表4中提供最好的可解釋性。該網絡已經群集組兩個組
的人在，他們過於注重精益很少很相似
強調敏捷。這三個集群的選擇沒有那麼專注於集團
主要在敏捷精益的排斥。我們認為，四集群的選擇
提供了更多的解釋力比三個集群，而不選擇
增加了網絡的複雜性，VE-集群解決方案，因此我們選擇了四組，這是
在文獻中常見的一些（羅斯和米勒，1992;博耶，沃德，與梁，
1996年）。我們看好四集群解決方案，因為它提供了更多的解釋
權力，實際上我們認為精益之間/敏捷和傳統是非常重要的區別。