เอกสารการประชุมวิชาการและเสนอผลงานวิจัย มหาวิทยาลัยทักษิณ ครั้งที่ 19 2552 - page 148

„µ¦ž¦³»¤ª·µ„µ¦Â¨³Áœ°Ÿ¨Šµœª·‹´¥ ¤®µª·š¥µ¨´¥š´„¬·– ‡¦´ÊŠš¸É 19 ž¦³‹Îµže 2552
„µ¦Áž¦¸¥Áš¸¥ª·›¸„µ¦Âž¨Š o°¤¼¨š¸É¤¸„µ¦Â‹„‹ŠÅ¤nž„˜·®¨µ¥˜´ªÂž¦
Ä®o ¤¸„µ¦¨¼nÁ oµ¼n „µ¦Â‹„‹Šž„˜·®¨µ¥˜´ªÂž¦
Comparison of Data Transformation Methods of
Non-Multinormal Distribution to Multinormal Distribution
‹˜»¡¦ ˜´ÊŠ„µ‹œ£µœr1*
Jatuporn Tangkanjanapas1*
š‡´—¥n°
„µ¦ª·‹´¥Äœ‡¦´ÊŠœ¸Ê¤¸ª´˜™»ž¦³Š‡rÁ¡ºÉ°«¹„¬µÂ¨³Áž¦¸¥Áš¸¥¦¼žÂ„µ¦Âž¨Š o°¤¼ ¨š¸É„µ¦Â‹„‹ŠÅ¤nž„˜·
®¨µ¥˜´ ªÂž¦Ä®o¤¸„µ¦¨¼nÁ oµ¼n „µ¦Â‹„‹Šž„˜·®¨µ¥˜´ªÂž¦ ×¥¡·‹µ¦–µ„µ¦Âž¨Š o°¤¼ ¨š´ÊŠ®¤— 3 ª·›¸ ‡º° „µ¦Âž¨Š
o°¤¼¨Â â¦r˜ ¡µªÁª°¦r (folded power transformation) „µ¦Âž¨Š o°¤¼¨Â Guerrero-Johnson (Guerrero-
Johnson transformation) ¨³„µ¦Âž¨Š o°¤¼ ¨Â Á„¨ °Š„¼¨µ (scaled angular transformation) ×¥¡·‹µ¦–µ‹µ„
Áž°¦rÁŽœ˜r ¼Š»— °Š„µ¦¥°¤¦´ „µ¦Âž¨Š o°¤¼ ¨ªnµÅ—o ¤¸„µ¦Â‹„‹ŠÂž„˜·®¨µ¥˜´ªÂž¦‹µ„„µ¦š—°
¤¤˜· “µœ ( o°¤¼ ¨¤¸„µ¦Â‹„‹ŠÂž„˜·®¨µ¥˜´ ªÂž¦) Á¡ºÉ°Ä®oš¦µªnµ¦¼žÂ„µ¦Âž¨Š o°¤¼¨ÂÄ—Á®¤µ³¤
„´ o°¤¼¨»—œ´ÊœÇ ¤µ„š¸É »— £µ¥Ä˜o ™µœ„µ¦–rš¸É „ε®œ— ‡º° ž¦³µ„¦¤¸„µ¦Â‹„‹ŠÂÁ¨ ¸Ê„ε¨´Š „µ¦Â‹„‹Š
Åª¼¨¨r „µ¦Â‹„‹ŠÂÂ„¤¤µ ¨³„µ¦Â‹„‹ŠÂÁ˜µ Ž¹ÉŠ„ε®œ—Ä®o ¤¸¨´„¬–³Ã—nŠÁo˜nµŠÇ„´œ œµ—
˜´ª°¥nµŠ ‡º° 20 , 30 , 50 ¨³ 100 – ¦³—´œ´¥Îµ‡´ °Š„µ¦š—°Áš¸¥‡ªµ¤„¨¤„¨º œ 0.05
Ĝ„µ¦ª·‹´¥‡¦´ÊŠœ¸Ê‹Îµ¨°Š™µœ„µ¦–r„µ¦š—¨°Š—o ª¥Áš‡œ·‡¤°œ˜·‡µ¦rè Ž¹ÉŠšÎµ„µ¦š—¨°ŠŽÊεĜ˜n¨³¦°
‹œ„ªnµŸ¨˜nµŠ¦³®ªnµŠ X~ Ĝ¦°š¸É i+1 „´¦°š¸É i ¤¸‡nµœo °¥¤µ„‹œ™º°ªn µÅ¤n˜„˜n µŠ„´œ ×¥¤¸ Á„–”rªnµ
0001 .0 ~ ~
1 d
i i X X (×¥š¸É i = 1,2,…,k ¨³ i ‡º°‹Îµœªœ¦°Äœ„µ¦š—¨°Š) ‹¹Š®¥»—„µ¦š—¨°Š
„µ¦ª·‹´¥‡¦´Ê Šœ¸Ê ¦»žÅ—oªnµ 宦´š»„„µ¦Â‹„‹ŠÄœš»„¦³—´ °Š‡ªµ¤Áo nªœÄ®nª·›¸„µ¦Âž¨ŠÂ â¦r ˜
¡µªÁª°¦r ¨³ª·›¸„µ¦Âž¨ŠÂ Guerrero-Johnson Ä®oÁž°¦rÁŽœ˜r „µ¦¥°¤¦´¤¤˜·“µœ¼Šš¸É »—
‡ÎµÎµ‡´ : „µ¦Âž¨Š o °¤¼¨ ; „µ¦Â‹„‹Šž„˜·®¨µ¥˜´ªÂž¦ ; „µ¦Â‹„‹ŠÅª¼¨¨r
Abstract
The objective of this research was to investigate and compare the data transformation forms which can
transform data to multinormal distribution. The data transformation forms were folder power transformation,
Guerrero-Johnson transformation and scaled angular Transformation. The highest percentage of accepted
multinormal distribution was used in evaluating the capability of the most appropriate data transformation forms
1 œ· ·˜µ µª·µ™· ˜· ®¨´ „¼˜¦ž¦·µ™·˜·«µ˜¦¤®µ´–”·˜ ‡–³¡µ–·¥«µ˜¦r¨³„µ¦´¸ ‹» ¯µ¨Š„¦–r¤®µª· š¥µ¨´¥
* Corresponding author : Ú¦«´¡šr/Ú¦µ¦ 081-6442979Email-gui-bing@hotmail.com