我将首先做单向方差分析。我已经决定使用频率喝李克特量表:0饮料在过去3个月= 1;1 - 10饮料在最后3个月= 2;10 - 20饮料在最后3个月= 3;20 - 30饮料在最后3个月= 4;30 - 40饮料在过去3个月= 5。喝酒被认为是8盎司的酒。消费可能发生在1(3 8盎司饮料一天或10 8盎司饮料超过3个月)。很明显,我们需要清楚地定义这个,看看青少年讲述这件事。我将使用标准使用过然后问青少年这喝酒和醉酒对他们意味着什么。我的假设是,会意味着不同的那些参与了干预和那些没有。我怀疑意味着将明显不同的干预和不干预和那些在学校干预可能高于学校但不显著。这里我想看看f统计量(观察之间的差异/内差异,预期偶然),看看它是重要的。我将想要运行一个方差分析对比的区别所在。我也想把我的一些控制变量,性别、SES,学校,预先测试,看看这些影响我的结果。如果是这样,我将离开或如果他们不占方差结果我将离开了。我这里可能会发现,某些类别的干预工作更好的学生比其他人更如此。我也可以把在不同时间点的测试,然而,方差分析不能考虑这些时间点的生长。我可能要看分数为每个时间点随时间的变化。这将是有趣的发展这些时间点。我可能想看看这些其他的时间点在最终结果产生影响;可能有一些发展喝像如果你15岁就开始更有可能继续下去。我可能会想要使用一个潜在的增长曲线模型来测试这些影响。现在我不会把这些措施我认为他们将与最终结果高度相关,会导致multicolinearity在回归问题。因此,我也使用分层回归测试干预效果和测试的结果。我想使用一个阻塞方法。我的拳头放在人口变量的块1;进行预测块2,然后是否参与干预块3结局变量的消费。在运行此回归之前我想要确保所有的假设OLS得到满足。可能我有一些测量问题最终的结果。我必须满足的假设是1)x和y的关系必须是线性的。2)没有测量误差(测量的研究可能是违反饮酒但我会寻找强大的心理测验学。3)所有变量的理论模型中包含我们很可能留下很多。4)残差正态分布。5)残差6)Homoscedascity -独立变量的传播residueals是一致的。
I will first do a one-way anova. I have settled on using frequency of drinking on a likert scale: 0 drinks in last 3 months = 1; 1-10 drinks in last 3 months = 2; 10-20 drinks in last 3 months = 3; 20-30 drinks in last 3 months = 4; 30-40 drinks in last 3 months = 5. A drink is considered 8 oz of alcohol. Consumption could happen in 1 sitting (three 8 oz drinks on one day or 10 8 oz drinks over last 3 months). Obviously we need to clearly define this and see if teens relate to this. I would use standards used before and then ask teens about this and what a drink and getting drunk means to them. So my hypothesis is that there will be mean differences in those that participated in the intervention and those that did not. I suspect the means will be significantly different for those in intervention and not in intervention and those in the school with intervention might be higher than the school with out but it will not be statistically significant. Here I will want to look at the F-statistic (looking at between difference/within differences, expected by chance) and see if it is significant. I will then want to run an anova contrast to see where the difference lies. I may also want to bring in some of my control variables, gender, SES, school, and pre-test to see if these influence my outcome. If so, I will leave in or if they do not account for variance in outcome I will leave out. I may find here that the intervention worked better for some categories of students more so than others. I could also put in the different time points of the test, however, ANOVA will not be able consider the growth of these time points. I may want to look at the change scores over time for each time point. It would be interesting to see the development of these time points. I may want to see if these other time points make a difference in the final outcome; it may be that there is something developmental about drinking like if you start at 15 you are more likely to continue. I would likely want to use a latent growth curve model to test for these effects. Right now I won't be putting these measures in as I assume they will be highly correlated with the final outcome and would cause multicolinearity problems within a regression. Thus, I would also use hierarchical regression to test the intervention effects and the results of my test. I would want to use a blocking method. I would fist put in the demographic variables as Block 1; pretest as Block 2; then whether or not participated in intervention as Block 3 on my final outcome variable of consumption. Before running this regression I would want to make sure all the assumptions of OLS were met. It is likely I have some measurement issues with the final outcome. The assumptions I must meet are 1) The relationship between x and y must be linear. 2) No measurement error (the study is probably violating with the alcohol consumption measurement but I would look for strong psychometrics. 3) All variables from the theory included in model We are likely leaving a lot out. 4) Residuals normally distributed. 5) Residuals independent 6) Homoscedascity - the spread of variables residueals are consistent.