各位大神救救我!!
我做了一个随机对照研究,其中有2组为干预组,1组为没有干预的对照组,最后结果是ITT没有意义,又因为其中一组的干预组依从率比较低,所以用了Per-Protocol,结果是阳性,但是2个审稿人均说PP不严谨,均提出需要用CACE/LATE来分析,我查了很多文献,发现我的研究是one-sided non-compliance,就是组间没有Crossover,在对照组的人不会接受任何干预,而干预1组和干预2组会出现没有接受干预的情况,但不会出现交叉,所以这个研究只会出现Complier和Never-taker,按照文献说的我应该是使用ATE(Average treamtent effect)、ATT(Average treament effect on the treated)/TOT(Treatment effect on the treated)进行分析。
随后我用R的“ATT”“ATE”包做了两两对比,得出的结果跟预期一样,但是那位Reviewer又说要用另一种方法,还给了一个PDF我参考,Reviewer是这样说的:
“-On page 11, you said that:"…..Therefore, the average treatment effect (ATE) and average treatmenteffect on the treated (ATT) were also estimated [33]". I am very confusedhere. In the intention-to-treat (ITT) analysis, you are estimating the ATE. TheATE is then the average of the slope over the entire population and the ATT isthe average of the slope over the subset of the population (those who actuallyreceived the program). Since you have only a one-side compliance, the ATT isthe coefficients of causal impact divided by the proportion of participants whoactually received in the intervention in the treated group. Hence, I willsuggest that you report the proportion of respondents who actually received theintervention in the treated group. Then estimate the ATT following my adviceabove. In Stata, you first need to estimate the causal impact of your programbased on the ITT. Then, you save the coefficients of the ITT. Lastly, youestimate the ATT: divide the coefficients of the ITT by the proportion ofrespondents who actually received the intervention in the treated group. Toreport the standard errors and 95% confidence interval, you should use thedelta method, nlcom command in Stata. The link below provides a comprehensiveunderstanding of this: https://www.hhs.gov/ash/oah/sites/default/files/estimating-program-effects-on-program-participants-brief.pdf”
而这个Brief是这样说的:
而我也没找到更多有用的信息,于是按照Reveiwer的意见在Stata操作(我是Stata小白一玫),先用probit算出干预的Coefficient:
然后用“nlcom”命令,用Coefficient除以依从率(40%),得出:
然后就懵13了!!!这个不就是普通的除法吗~有什么意义呢。。。。而且我发现无论除以什么值,z和P值都不会变,应该不是这样的吧?? 求助各位大神帮帮忙看看是什么问题啊,感激不尽啊!!!
|