meta分析中固定效应模型和随机效应模型的选择

在metan命令中，有四个选项来做随机效应模型和固定效应模型，分别为fixed, fixedi, random, randomi, 请问选择加i和不加i的选择原则是什么？另外，我想同时显示固定效应和随机效应，为什么把两个选项fixedi, randomi同时放入进去不行，显示


. metan depigr nodepigr depngr nodepngr, label(namevar=study, yearvar=year) or fixedi randomi
Invalid specifications for combining trials

As regards to "请问选择加i和不加i的选择原则是什么？", please read this article:
http://www.stata-journal.com/article.html?article=sbe24_2

As regards to "为什么把两个选项fixedi, randomi同时放入进去不行", you can do like this:

metan depigr nodepigr depngr nodepngr, label(namevar=study, yearvar=year) or randomi  second(fixedi) counts texts(120) astext(60)

hplcdadong 发表于 2013-8-5 15:19
As regards to "请问选择加i和不加i的选择原则是什么？", please read this article:
http://www.stata-j ...

同时显示的问题解决了，但是那个网站的pdf我下载不了，哥能不能简单给说说，加i和不加i选择的依据是什么，一个是M-H, 一个是I-V，具体操作中一般怎么选择

另外是不是两个连用的时候，必须是一样的才能联用？
比如如果是fixedi，就必须用randomi？
求哥帮忙

There is no clear-cut simple rule for meta-analysis.

When you play with (explore) your data, you can try both random and fixed effect model at the same time. For final results, you either use random effect model or fixed effect model but not both at the same time. Which model you select depends on your data and the aim of your study.

Generally speaking, when studies are gathered from the public literature, the random effect model is generally a more plausible first choice. The inverse variance (I-V) method and the Mantel-Haenszel (M-H) method uses different weighting scheme to compute summary (final) effect. Although in many cases, I-V and M-H give similar results, the inverse variance method may perform poorly for studies with very low or very high event rates or small sample size.

Take your igr-ngr study in your recent post as an example, if I were you, I would use the following command as the backbone of the final analysis:

metan depigr nodepigr depngr nodepngr, label(namevar=study, yearvar=year) or random counts texts(120) astext(60)

Good luck

hplcdadong 发表于 2013-8-7 03:28
There is no clear-cut simple rule for meta-analysis.

When you play with (explore) your data, you ...

非常感谢哥，明白多了。

thx for sharing

看看，