Test Whether Two Regression Lines IntersectOver a Finite Interval

• Main Authors: Hai-shia Ren, 任海峽
• Other Authors: Jungpin Wu
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/72195591793080110730

碩士 === 逢甲大學 === 統計與精算所 === 97 === While discussing the difference of two regression lines with one covariate, two statistical tools, ANCOVA, and Johnson-Neyman Technique, are often to be used. Generally, the most important reason to decide which one would be chosen, is whether two regression lines are parallel. If these lines are not parallel, that is, intersect at some point; then, we would choose Johnson-Neyman technique to analyze data. Existence of interection point somehow means two regression lines are not parallel; however, if this point is far away from our research range for covariate, JN tech''s non-significant region would be also away from the range. JN tech''s non-significant region has strongly affected by interection point so it may cause some problem hard to explain. In this situation, the choice of JN tech should be discussed again. If the interection point lies in research range, we can definitely use JN tech to analyze data; nevertheless, when interection point''s x-coordinate estimated value exceed our range and regression lines are not parallel, use JN tech to solve question may be not appropriate. For discussing if JN tech is appropriate over a finite interval, this article suggests one hypothesis testing procedure to determine whether two regression lines intersect over a finite interval. We use bootstrap hypothesis testing procedure to solve this question, and use Monte-Carlo simulation method to discuss type I error and power in different sample, sigma, and slope of two regresiion lines.