Compute the treatment effect (or effect size) automatically

In every meta-analysis you start with the published summary data for each study and compute the treatment effect (or effect size). For example, if a study reports the number of events in each group, you might compute the odds ratio.  Or, if a study reports means and standard deviations, you might compute the standardized mean difference.  This process of computing effect sizes is typically tedious and time-consuming. In some cases, especially when studies present data in different formats, the process is also difficult and prone to error.

With CMA the process is fast and accurate

With CMA you enter whatever summary data was reported in the published study, and the program computes the effect size from that summary data. For example, you could enter events and sample size, and the program would compute the odds ratio.  Or, you could enter means and standard deviations, and the program would compute the standardized mean difference.  Three examples (selected from more than a hundred options) are shown here.

What if my data is in some other format?

What if your studies reported data in some other format?  Perhaps you have studies that reported only a p-value and sample size.  Or, you have studies that reported an odds ratio and confidence limits.  With any other program you would need to compute the effect size and variance for each study before proceeding to the analysis.  By contrast, CMA allows you to enter almost any kind of data – it includes 100 formats for data entry similar to the three shown above.  Simply locate your data type in a list and CMA will create the corresponding columns in the spreadsheet.

> Click here to see the entire list.

What formula is the program using to compute these effects?

To see the formula used to compute an effect size, double-click on that effect size.  The program opens a dialog box that shows the exact formula used and also all details of the computation for that specific row.

What if I want to use another index of treatment effect?

In one of the examples shown above we entered events and sample size and the program computed the odds ratio. What if you would prefer to work with the risk ratio?  Or what if you wanted to compute the standardized mean difference corresponding to the odds ratio?  In another example we entered means and standard deviations and the program computed the standardized mean difference.  What if you would prefer to work with the raw mean difference, or to compute the correlation corresponding to the standardized mean difference?

CMA allows you to work with the index of your choice, and to switch back and forth among indices. 

For example, if you have entered the events and sample size, the program will compute the odds ratio, log odds ratio, risk ratio, log risk ratio, risk difference, standardized mean difference (d), bias-corrected standardized mean difference (g), correlation, and Fisher’s z. Or, if you enter means and standard deviations the program will compute the raw mean difference, standardized mean difference (d), bias-corrected standardized mean difference (g), correlation, Fisher’s z, log odds ratio, and odds ratio. 

These examples are a subset of the supported formats and indices.

What if different studies reported different kinds of data?

Above, we showed that you can customize the data-entry screen to accept almost any kind of data.  But what different studies provide different kinds of data? For example, what if one study reported events and sample size while another reported the odds ratio and confidence interval?  How would you get both kinds of data into the program?

CMA allows you to mix and match the different data formats.  You can enter events and sample size for the first few studies, then odds ratio and confidence interval for the next few studies, log odds ratios with variances for others, and so on.  Or, you can enter means and standard deviations for some studies, p-values for other studies, t-values for others, and so on. You can customize the spreadsheet with as many kinds of data formats as you like.  The program will compute the effect size from each of them and (to the extent possible) allow you to include them all in the same analysis.  CMA is the only program to offer this feature.

What if some (or all) of my studies include pre-post or crossover designs?

CMA includes templates for more than 20 pre-post or crossover designs, which is of particular import since the standard error for these may be difficult to compute otherwise. And, you can mix and match these studies with studies that used post-tests alone. 

> For a list of all formats click here.

For more detail on the computational options for paired studies download the whitepaper.

What if I have already computed the effect size?

If you have already computed the effect size and its variance (or standard error) you may enter these directly (the same as you would enter data in any other format).

Can I mix binary, continuous, and correlational data?

As explained above, the program allows you to enter summary data in more than one format – for example, events and sample size for one study and odds ratios with confidence intervals for another.  But in this example both studies used binary data. What if some studies report binary data (events and sample size) while others report continuous data (means and standard deviations) or correlational data? 

The program is able to convert across these different classes of data.  It will convert among odds ratios, standardized mean difference, and correlations so that all may be used in the same analysis.  A whitepaper gives details on these algorithms.

What if I have studies that look at point estimates rather than effect sizes or treatment effects?

While most meta-analyses work with effect sizes (which assess the relationship between two variables) some are used to estimate a risk, rate, or mean in one group (for example, “What is the risk of Lyme disease?”).  CMA will work with these effects (or point estimates) as well.

> Click here to see the entire list.

Can I run a meta-analysis on regression weights?

Yes.  In addition to being able to work with recognized effects (such as odds ratios and mean differences) the program is able to work with generic point estimates which may be analyzed either in their original scale or on a log scale.

> Download a free software trial