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]

From |
Christina Wei <[email protected]> |

To |
[email protected] |

Subject |
Re: st: RE: meta analysis in stata |

Date |
Tue, 5 Nov 2013 17:49:16 -0500 |

```
I've performed a couple of meta-analyses using metan command in Stata
so here are my thoughts:
I concur with most of the replies to your post that meta-analysis is a
way to summarize the effect size of a group of studies. In other
words, you are summarizing outcomes from all studies that supposedly
are similar (as you do not want to summarize apples and oranges; you
will get a high I squared or Cochrane Q). You can elect to use random
or fixed effect model when you are summarizing the effect sizes. To
err on the conservative side, I would just use random effect model.
If you need more explanations for this I'd be happy to explain the
reasons to you in another separate email. Also, you would need to
interpret I-squared value to make sure that most of your studies are
similar enough to justify the summation. Also I concur that it is
better that you weight your study based on study size, rather than
inverse variance -> unless most of your studies have similar study
sizes.
It appears that you have many outcomes so I am not sure as to what you
want to study.
For binary outcomes, the results are usually expressed as odds ratios
or hazards ratios in papers. Hence you would be summarizing ORs and
HRs (and the associated confidence intervals). You would also need
standard error when you use then metan command. And don't forget to
log transform the ORs/HRs and the associated CIs when you input your
data for your meta-analysis (it is required if you are using metan
command).
I hope this helps
~Christina
On Tue, Nov 5, 2013 at 12:27 AM, Timothy Mak <[email protected]> wrote:
> Dear Azreen,
>
> A meta-analysis is basically a (weighted) average of the effects as reported by the different studies/papers. In Stata, the usual command that is used is -metan-, from SSC. As I can tell from your description below, your situation is a bit unusual, because you don't seem to have a variable for the standard error/confidence interval for your study-specific estimate.
>
> In a meta-analysis, the standard error/CI controls the weight that is given to a particular study. Given that you don't have this information, it appears that the next reasonable step would be to assume equal weights for all your studies, in which case, you can just use the -mean- command, or the -summ- command, e.g.:
>
> mean Y
> sum Y
>
> and you will be able to obtain the average of Y across studies. As your weights are equal, the estimate would be the same for both a fixed effects and a random effects model. The random-effects variance, however, is unidentifiable without information on standard error/CI.
>
> A normal meta-analysis does not involve additional variables, such as those you have: wealth, health, etc. It is straightforward to extend the meta-analysis model to include covariates, and this is known as meta-regression in the literature. However, the practice is not recommended by many people, especially when you don't have a lot of studies. Since you don't have standard errors/CIs, if you assume equal weights, your meta-regression is again the same as a normal linear regression. Thus you can do it by, e.g.:
>
> reg Y i.Income I.consumption
>
> HTH,
> Tim
>
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Azreen Karim
> Sent: 05 November 2013 08:36
> To: [email protected]
> Subject: st: meta analysis in stata
>
>
> Dear Stata users,
>
> I am a beginner in using Stata and would like to use it in my current project which is a meta analysis in economics. My current meta dataset have got the following variables:
> Paperid (unique paper code)
> Y (all percentage changes of specific variables)
> Income (1, 2, 3 to define different types of income)
> Consumption (1 ,2, 3 to define different types of consumption)
> Poverty (0, 1)
> Wealth (0,1)
> Health (1,2 to define different categories)
> Education (1,2 to define different categories)
> Labor (1,2 to define different categories)
> Disaster (1,2,3,4 to define different types of disasters)
> Method (1,2 to define different techniques used in various studies)
> Control variables (household, time, region, demographic, socioeconomic, geography) in binary (0,1) format
> Dependent variables in this model would be Y, income, consumption, poverty, wealth, health,education,labor i.e looking at the impacts of disasters on these variables.
> Method and disaster would supposedly be independent variables (although i am not sure how to utilize unique paper code inside the model.
> All the data are in percentage changes and had been grouped together in Y column. The other dependent variables(e.g. income, consumption and so on) basically represents the different categories of which the respective data had been grouped under Y.
> To look at the implications on different categories in particular, I have generated a couple of variables as of following:
> gen consume_1 if consumption=1
> replace consume_1=0 if consumption!=1
> gen consume_2 if consumption=2
> replace consume_2=0 if consumption!=2
> drop consume_2
> gen consume_2
> gen consume_2 if consumption=1
> replace consume_2=0 if consumption!=2
> gen consume_3 if consumption=1
> replace consume_3=0 if consumption!=3
> gen health_1 if health=1
> replace health_1=0 if health!=1
> gen educ_1 if education=1
> replace educ_1=0 if education!=1
> gen labor_1 if labor=1
> replace labor_1=0 if labor!=1
> That means inaddtion to the above, I have income_1,income_2 and income_3 and similar for the rest as well in binary (0,1) format.To conduct a meta analysis in Stata, could anyone help me in suggesting what kind of regression I should be possibly utilizing and giving me an example of the commands I should be using in Stata? Also please explain a bit on how to use the control variables inside this model.
>
> Cheers,
>
> Azreen Karim
> PhD Candidate
>
> *
> * 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/
*
* 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/
```

**Follow-Ups**:**Re: st: RE: meta analysis in stata***From:*Azreen Karim <[email protected]>

**References**:**st: meta analysis in stata***From:*Azreen Karim <[email protected]>

**st: RE: meta analysis in stata***From:*Timothy Mak <[email protected]>

- Prev by Date:
**st: RE: Need help with Programming/ running a forvalues loop inside foreach loop** - Next by Date:
**Re: st: Using bootstrapping to validate model with svy data** - Previous by thread:
**st: RE: meta analysis in stata** - Next by thread:
**Re: st: RE: meta analysis in stata** - Index(es):