Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: RE: meta analysis modelling and regression in stata
From
Timothy Mak <[email protected]>
To
"[email protected]" <[email protected]>
Subject
st: RE: meta analysis modelling and regression in stata
Date
Mon, 11 Nov 2013 13:49:34 +0800
Dear Azreen,
First, I think I said in my earlier reply that if you want to involve other variables in your meta-analysis than Y alone, you're looking at a meta-regression, not a meta-analysis. And I said that a meta-regression is often not advisable, especially if you do not have a huge number of studies.
A meta-regression is not unlike a normal regression model, and there's a huge body of literature on variable selection in regression model. For an introduction, you might want to look the FAQ on stepwise regression:
http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/
or Frank Harrell's Statistics in Medicine paper in 1996:
http://onlinelibrary.wiley.com/doi/10.1002/(SICI)1097-0258(19960229)15:4%3C361::AID-SIM168%3E3.0.CO;2-4/abstract
In summary, there is no right/wrong answer on what variables should be included in a regression or not, although most advise not using any kind of automatic algorithms, but rather do what is "sensible". A meta-regression is more problematic than a normal regression, because studies differ wildly in many ways, and I think most would advise not to do it at all, unless it really is sensible.
Also, I think you're confused about what we mean by "weighting". The fact that all your numbers are percentages doesn't imply that they therefore have the same weight. As Red Owl pointed out, the weights can be related to sample size.
-metan- is relevant only if you feel that different studies should be contributing different weights to the meta-analysis/regression. Otherwise, simple -mean- and/or -regress- would do.
I wonder if it is best for you to consult a textbook on meta-analysis. I recommend Borenstein et al (2009): http://onlinelibrary.wiley.com/book/10.1002/9780470743386
HTH,
Tim
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Azreen Karim
Sent: 11 November 2013 07:13
To: [email protected]
Subject: st: meta analysis modelling and regression in stata
Hi all,
Its been really great to get some really good comments on this issue. I would like to start from the point where people seems to agree - that meta analysis is basically done on a weighted average of effect sizes. In my current project, all of my data regarding the dependent variable are in percentage changes i.e. I have taken percentage changes in income, consumption and other variables. I think the weighted problem has been taken care through this as all of these numbers are out of 100 (same weight). The basic problem is, I am still confused which model should I go for in my case? and as I am new to stata, how am I going to write down the regression commands? my meta dataset has got the following format:
Y paper income consumption poverty wealth health labor education hhcommunity time region demographic socioeconomic geognature method disaster
-9.7/ 3 /3 /0 /0 /0 /0 /0 /0 /0 /0 /0 /0 /1 /0 /1 /1
-7.56 /4 /3 /0 /0 /0 /0 /0 /0 /1 /0 /1 /0 /1 /0 /1 /2
-3.08 /1/4 /3 /0 /0 /0 /0 /0 /0 /1 /1 /1 /0 /1 /0 /1 /1
Here, to describe the first line of the dataset, -9.7 is the value from paper 3, 3 is the type of income (e.g.farm income)and this is an income value the rest are 0s, method 1 means here regression(the type of method the paper has used) and disaster 1 means the specific type of disaster (e.g. flood). Therefore, all values are grouped under Y (they are all taken in percentage terms) and the subsequent variables represent the types of variables that I have defined. I am going to look at the impacts of disasters on all these variables and their types separately. The dependent variables are from Y till education, from hhcommunity to geognature are the controls (written in binary format whether diffrent papers put these controls or not, method and disaster are the independent variable. This study is to look at the impacts of disasters on different poverty variables taken from different studies.
I am still confused whether I need the metan command or not? or I could simply do the regression straight away? which model should be better? and how to write the commands in stata?
Cheers,
Azreen.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/