เอกสารการประชุมวิชาการและเสนอผลงานวิจัย มหาวิทยาลัยทักษิณ ครั้งที่ 19 2552 - page 169

2
„µ¦ž¦³» ¤ª·µ„µ¦Â¨³Áœ°Ÿ¨Šµœª·‹´¥ ¤®µª·š¥µ¨´ ¥š´„¬·– ‡¦´Ê Šš¸É 19 ž¦³‹Îµže 2552
Abstract
The objective of this study is correction for heterogeneity of variance in the completely randomized
design. The fixed-effect completely randomized design model is ij i ij Y H W P when k i ,..., 1 and
n j ,..., 1 where ij Y is the th j response variable for the th i treatment, P is the grand mean, i W is the th i
treatment effect and ij H is a random error of the th j response variable for the th i treatment. The ij H is
independently and normally distribution with mean 0 and variance k i
i ,..., 1 , 2 V when k is the number of
treatment. In this study, The generation of response data is done by S-PLUS 2000 package. The data are generated
with 3,5 and 7 treatments. The data are generated with 3,4,5 and 6 number of replication for each treatment.
The variance ratio of data for different 3 levels; small, medium and high level. The criterion of determination in
this study are the proportion of success in correction for variances heterogeneity problem, the proportion of data is
still normally assumption after transformation. The result of this study can be summarized as follow:
The fit transformation methods for correcting the heterogeneity of variance are transformed data and the
data after transformation is still normally distribution. When the variance ratio is small level case; Treatment is
equal 3 are 0.2 O and 5.1 O ,Treatment is equal 5 are 5.0 O , Treatment is equal 7 are 5.1 O and
0.1 O . When the variance ratio is medium level case; Treatment is equal 3 are 5.1 O and
0.1 O ,Treatment is equal 5 are 5.0 O and 0.0 O , Treatment is equal 7 is 5.0 O .When the
variance ratio is high level case; Treatment is equal 3 are 5.1 O and 0.1 O ,Treatment is equal 5 is
0.0 O , Treatment is equal 7 are 5.0 O and 0.0 O
Keywords: Heterogeneity of variance Completely Randomized design Box-Cox of power transformation
‡Î µœÎµ
ŸœÂ„µ¦š—¨°Š»n¤˜¨°—(Completely Randomized Design) Ÿœ„µ¦š—¨°Šœ¸ÊÁ®¤µ³Î µ®¦´„¦–¸ š¸É
®œnª¥š—¨°Šš¸ÉœÎ µ¤µÄo¤¸¨´„¬–³¤ÉÎ µÁ¤°®¦º°‡¨oµ¥‡¨¹Š„´œ¤µ„š¸É »— „µ¦‹´—ª·›¸ š—¨°ŠÄ®o˜´ª°¥nµŠÂ˜n ¨³„¨»n¤°¥n µŠ
»n¤ Ÿ¼o ª·‹´¥¤´„čo„µ¦ª·Á‡¦µ³®r ‡ªµ¤Âž¦ž¦ªœ(Analysis of Variance:ANOVA) ÁœºÉ°Š‹µ„µ¤µ¦™°›·µ¥š¸É¤µ °ŠŸ¨
„µ¦š—°Å—o°¥nµŠ´ —Á‹œ ¨³Äo„´œ°¥nµŠÂ¡¦n®¨µ¥ œ°„‹µ„Ÿ¼o ª·‹´¥˜o°Š¦¼o‹´„Á¨º°„čo ™·˜·š—°Ä®o Á®¤µ³¤„´
¨´„¬–³ °Š o°¤¼¨Â¨oª Ÿ¼oª·‹´¥¥´Š˜o°Š‡Îµœ¹ Š™¹Š o°„ε®œ—ÁºÊ°Š˜oœ(Assumption) °ŠÂ˜n¨³™·˜·š—° Ž¹ÉŠ™o µ iµ gœ
o°„ε®œ—ÁºÊ°Š˜oœ°µ‹‹³šÎµÄ®oŗoŸ¨¦» žÅ¤n˜¦Š‡ªµ¤‹¦·Š ‹µ„„µ¦«¹ „¬µ‡oœ‡ªo µ¡ªnµ„µ¦ iµ gœ o°„ε®œ—ÁºÊ°Š˜oœ