full2010.pdf - page 415

377
As the sample size, the number of independent variables, and the proportion of failure are kept constant, with
the proportion of failure equal to 0.5, the optimal cut-off point will decrease when the degree of multicollinearity
increases, and with the other proportions of failure, the optimal cut-off point will decrease to the degree of
multicollinearity equal to 0.67 and slightly increase when the degree of multicollinearity equal to 0.99.
As the number of independent variables, the proportion of failure, and the degree of multicollinearity are kept
constant, when the sample size increases, with the proportion of failure equal to 0.1, the optimal cut-off point will
decrease, at the proportion of failure equal to 0.5, the optimal cut-off point will be constant, this cut-off point is in the
neighborhood of 0.4623, and at the proportion of failure equal to 0.9, the optimal cut-off point will increases.
As the sample size, the degree of multicollinearity, and the proportion of failure are kept constant, the
optimal cut-off point will decrease when the number of independent variables increases.
From the multiple regression model with all interaction terms. It is found that the coefficient of multiple
determination (R2=0.907), is considerably high. From the estimated regression equation, the optimal cut-off point for
any situation can be predicted.
Keywords: Binary logistic regression model, Classification error rate, Cut-off points
‡Î µœÎµ
˜´ªÂ„µ¦™—™°¥Ã¨‹· ˜·„ 2 ž¦³Á£š (Binary Logistic Regression Model) ÁžÈ œ˜´ªÂš¸É«¹„¬µ
‡ªµ¤´¤¡´œ›r¦³®ªµŠ˜´ªÂž¦°· ¦³®¨µ¥˜´ª„˜´ªÂž¦˜µ¤Ž¹É ŠÁžÈ œ˜´ªÂž¦Á· Š‡»–£µ¡š¸É¤¸‡µÅ—oÁ¡¸¥Š n ´ n 2 ‡µn (Dichotomous
®¦º° Binary Variable) ªœ˜´ªÂž¦°· ¦³ÁžÈ œ˜´ªÂž¦Á· Šž¦· n ¤µ–®¦º°˜´ªÂž¦Á·Š‡» –£µ¡®¦º °°µ‹‹³¤¸š´Š˜´ªÂž¦Á·Š Ê
ž¦· ¤µ–®¦º °˜´ªÂž¦Á·Š‡»–£µ¡„Å—o œÎµÅžÄoĜ„µ¦ž¦³¤µ–‡µ˜´ªÂž¦˜µ¤®¦º°„µ¦¡¥µ„¦–rð„µš¸É˜¨³®œª¥‹³°¥¼ È n n n n
Ĝ„¨»¤Ä—„¨»¤®œ¹É Š Ž¹É ŠÄœž´‹‹»´œ™¼„œÎµ¤µÄoÁžÈ œ°¥µŠ¤µ„Äœ„µ¦¡¥µ„¦–r˜´ªÂž¦˜µ¤Á·Š‡»–£µ¡ªµ‹³°¥¼Äœ„¨»¤Ä—„¨» n n n n n n n ¤
®œ¹ÉŠÄœ 2 „¨» ¤ Ž¹É Š„‡º ° ‡ªµ¤ÎµÁ¦È‹ n È (Success) ¨³‡ªµ¤¨o¤Á®¨ª (Failure) °Š¨´„¬–³š¸É œÄ‹«¹„¬µÃ—¥Äo˜´ªÂž¦°· ¦³
ÁžÈ œ˜´ª¡¥µ„¦–r ˜´ªÂœ¸ ™¼„œÎµ¤µÄož¦³Ã¥œr°¥µŠÂ¡¦®¨µ¥ÄœšµŠÁ«¦¬“«µ˜¦r Ê n n , ›» ¦„‹Â¨³ª·š¥µ«µ˜¦r „µ¦Â¡š¥r ·
ÁžÈ œ˜oœ Šµœª·‹´¥ªœÄ®‹´—Ä®o˜¨³®œª¥ ° n n n n ¥¼Äœ„¨»¤Ä—„¨»¤®œ¹É ŠÄœ n n n 2 „¨»¤ ץčo‹»—Š n n (Cut-Off Point) š¸É 0.5 ×¥
Ä®oÁ®˜» Ÿ¨ªµ„¨»¤ °Š‡ªµ¤ÎµÁ¦È‹¤¸Ã°„µÁ„— ¹œÁšµ„„¨»¤ °Š‡ªµ¤¨o¤Á®¨ª °Š¨´„¬–³š¸É œÄ‹«¹„¬µ n n · n ´ n Ê
Ĝ„µ¦¡¥µ„¦–rð„µš¸É ˜¨³®œª¥‹³°¥¼Äœ„¨»¤Ä—„¨»¤®œ¹É ŠÄœ n n n n n 2 „¨»¤ °Š˜´ªÂ„µ¦™—™°¥Ã¨‹· ˜· n „
2 ž¦³Á£š ž¦³Á—Ȝε‡´ž¦³Á—Èœ®œ¹É Š ‡º° ‹»—Šš¸ÉÁ®¤µ³¤š¸É »—‡º °‹»—Ä—š¸ÉšÎµÄ®o ´—ªœ‡ªµ¤™¼„˜o°ŠÄœ„µ¦¡¥µ„¦–r n n
„µ¦‹ÎµÂœ„„¨»¤¤¸‡µ¼Š»— n n Ž¹É ŠÅ¤¤¸Šµœª·‹´¥Ä—š¸É Á‡¥šÎµ„µ¦«¹„¬µ„µ¦‡´—Á¨º °„‹»—Š °Š˜´ªÂ n n „µ¦™—™°¥Ã¨‹· ˜·„š¸É‹³
šÎµÄ®o„µ¦‹Î µÂœ„„¨»¤¤¸‡ªµ¤™¼„˜o°Š¼ Š» n —×¥¡·‹µ¦–µ™¹Š„µ¦Â‹„‹Š °Š˜´ªÂž¦°· ¦³, ‹Îµœªœ °Š˜´ªÂž¦°· ¦³,