根據社會學習理論，接觸偏差同儕對偏差行為具有個人與脈絡層次的潛在影響，然而，過去研究受分析方式的限制，遲至近年來才有適當的統計分析方法來定量脈絡的影響。本研究考量過去相關研究在分析時所忽略的兩個層面：「測量誤差」與「分析層次」（個人與脈絡層次），進一步評估較適於探究接觸偏差同儕對偏差行為影響關係之理論模式。本研究採用單層與兩層SEM進行分析，研究發現：接觸偏差同儕與偏差行為的測量皆存有誤差，班級內青少年接觸偏差同儕與偏差行為皆存在著相似性，突顯出班級脈絡的影響。對個人而言，接觸偏差同儕程度越高的青少年從事偏差行為也越嚴重；同樣的，班級整體接觸偏差行為嚴重程度越高，班級內青少年從事偏差行為也越嚴重。更重要的是，班級脈絡效果比個人層次的影響效果大。據研究結果，社會學習理論內涵與實徵研究，應將接觸偏差同儕對偏差行為的脈絡層次影響納入考量，並採多層次SEM，以正確的探究理論變項間的關係。在教學實務面，可採情境轉換、輔導融入課程設計、增加與學生對話與觀察課間互動，來減少接觸偏差同儕對偏差行為的個人與脈絡影響效果。

According to social learning theory, deviant peer serves as learning model of deviances for adolescents. Adolescent learned deviant behavior, they learned deviant motivation and belief behind that behavior as well. It may cause diffusion of deviances. It suggested that part of young adolescents studied in the same classroom/context may behave similar behaviors. A well-known slang: ＂birds of a feather flock together＂ depicted this phenomenon, but it has been neglected in both theoretical literature and empirical researches. By inferring from social learning theory, deviant peers may exert both personal and contextual latent effects on individual deviances. However, contextual effect of deviant peers on deviances has not been addressed. One of possible reasons may also involve limitation on the usage of appropriate analytic methods in the past. There was appropriate statistic method could be used to quantify the effect of context recently. The main purpose of this study was to probe potential contextual effect of deviant peers by introducing alternative method. By using traditional and this alternative, this study considered two aspects which have been ignored: ＂measurement errors＂ and ＂analysis level＂ (single and two-level). Two theoretical models were proposed respectively to represent different consideration of one aspect or both aspects. Single level structural equation model was introduced to consider measurement errors but limited on its ability to take contextual effect into account. While two level structural equation model was used to consider measurement errors and to probe potential contextual effect. Several indices were used to evaluate the most suitable model on investigating the effect of peer deviance on individual deviance. Five thousand and seventy-nine junior high school students nested in one hundred and eighty three classes were sampled by using random cluster sampling method. All of which were consented to participate in this study. Results showed that there were considerable measurement errors resided in the measurements for deviant peer and individual deviance. There were also dependencies resided in observations. It implied the effect of the context in favor of deviances was salient. Combining these evidences suggested that was reasonable to use two level structural equation modeling for considering both effects of measurement errors and dependencies. This method may be not only more suitable and beneficial for probing contextual effect. Further evidences indicated that for individual, associated more with deviant peer resulted in more deviances. In the same token, on average, it showed the same pattern on class level. Specifically, it indicated that as the serious level of association with deviant peers on class level raised, the deeper the deviant problem resided in that context. More importantly, the contextual effect of deviant peers was greater than the individual effect of which. According to results, the contextual effect of deviant peers on individual deviances should be considered in social learning theory and empirical studies in addition to investigate the effect of deviant peers on individual deviances. Comparing to SEM, MSEM was a more useful tool which could be used to investigate explicit and implicit theoretical assumptions or relations. Because the contextual effect was more prominent than the effect on individual level. It suggested the contextual effect of deviant peers should be considered first and treated it seriously. In practical, the contextual effect may be weaken by adapting several methods. The first is context switching which is similar to time-out method. Theoretical, it may be useful for suspending the relationship between deviant peer and individual. Second, counseling methods could be incorporated into curriculum, and teacher should increasing frequency chats with students to understanding what they are thinking, doing and engaging in. The last but not least, it is important for teachers to observe interactions between students during breaking time in school.