Dear expert,

In my meta-analysis, some studies shared subjects. I would like to use "combine images" to combine these studies into a single map, but I don't know how to specify the assumed correlation (default is 0.3)?

In addition, if some studies shared the completely same subjects, shall I also combine them?

Best,
Yuanwei

Dear Yao,

If you don't have the raw data, you must assume one or other correlation. The assumed correlation (0.3) may be different from the true correlation in the raw data, but the error derived from this is "only" that the study receives more or less weight than it should. If you have the raw data, you can find the global correlation that best fits the data, e.g. the one that once applied minimizes the square error between t[(contrast1 + contrast2) / 2] and (t[contrast1] + t[contrast2]) / 2.

Alternatively, you may choose to just include the study or contrast with the larger sample size.

In any case, studies or contrasts from the same subjects should be included in the meta-analysis as a single study, and their coordinates cannot be simply mixed in a single text file.

Hope this helps,

Joaquim

Dear Joaquim,

Thank you for your reply, it really helps! Just for confirmation, I still have some further questions:

1. If I don't have raw data, I have two option: a). using an assumed correlation (e.g., 0.3) instead; or b). choosing only one contrast or study for later meta-analysis at the expense of losing a lot of data. Which one do you recommend?

2. If a study includes more than 2 contrasts (e.g., 5 contrasts), I will have multiple correlations (which are calculated between every 2 contrasts). How can I specify correlation in this case.

3. With regard to global correlation calculation, I have never done such analysis before. Is there any software can achieve this goal?

Thank you in advance,
Yuanwei

Dear Yuanwei,

Between assuming a 0.3 correlation or choosing only one contrast, I would generally choose to assume the correlation. However, this decision may depend on the meta-analysis, e.g. choosing one contrast may be probably preferable when tasks are imbalanced.

You can combine more than 2 contrasts.

Finally, as far as I know, currently there is no specific software to easily estimate a global correlation, I usually code my own scripts.

Hope this helps,

Joaquim