News:

Version 5.0 Beta Released. Works with Windows 11!

Main Menu

Converting Cohen's d to an Odds ration

Started by mfs, 16 October, 2007, 08:28:42 AM

« previous - next »

0 Members and 1 Guest are viewing this topic.

Dev

I am logged on, therefore I am.

Dev

Hi mfs,

By the end of the day the new version of clintools will have this feature. Download the new version, uninstall the old version, install the new version (you won't have to re-register the product) - this feature is on the last tab of the effect size generator - it computes both directions (OR to d and D to OR). Please let me know if this is what you were wanting.

Best,
Dev
I am logged on, therefore I am.

mfs

Dev: I am trying to find the reference for the 1.65. When I do I will send it your way. However, given that Chinn has a published article which includes the 1.81 I believe it is best to stick with 1.81.  thanks, mfs

Quote from: Dev on  18 October, 2007, 06:28:42 AM
Hi mfs,

Can you tell me where the 1.65 came from. Reversing Chinn's equation makes it roughly

d=Exp(d*1.81)

Someone must have made a correction for something to make the multiplier 1.65.

Thanks,

Dev

Dev

Hi mfs,

Can you tell me where the 1.65 came from. Reversing Chinn's equation makes it roughly

d=Exp(d*1.81)

Someone must have made a correction for something to make the multiplier 1.65.

Thanks,

Dev
I am logged on, therefore I am.

mfs

Quote from: Dev on  18 October, 2007, 06:08:06 AM
I have made a quick programme for converting OR to d and have attached it here. This is very BETA as I am awaiting checking of the computations (particularly for the confidence intervals) and, of course, it doesn't convert d to OR yet.

Is this useful?

Dev
Yes, this is helpful. I look forward to see the program to convert d into an OR. Also, I found the following equation for converting d into an OR.
OR = exp(1.65*d).

d = ln(OR)/1.65.

Dev

I have made a quick programme for converting OR to d and have attached it here. This is very BETA as I am awaiting checking of the computations (particularly for the confidence intervals) and, of course, it doesn't convert d to OR yet.

Is this useful?

Dev
I am logged on, therefore I am.

Dev

#3
OK - if the data is from continuous data then one may approximate d by dividing ln Odds Ratio (i.e., Natural Log Odds Ratio) by 1.81 (well, pi divided by the squareroot of 3).  See Chinn, S. (2000). A simple method for converting an odds ratio to effect size for use in meta-analysis. Statistic in Medicine, 19, 3127-3131. (I've attached the paper here - might have to remove it though soon due to IP issues so please grab it quick [and then cite it, of course]).

Rearranging the formula would mean that to get an OR from a d then one would:

multiply d by 1.81 to get ln(OR)

antilog the result to get an OR

But - that's a lot of power gone missing (translating a continuous effect size into a binary type effect size).

Does this help?

Would it be useful to have this built into ClinTools do you think?

Best,

Dev
I am logged on, therefore I am.

Dev

I take it that Cohen's d has been obtained from binary data somehow (e.g., Treatment / Control & Dead / Alive)? I'm not sure that this could be done accurately. I'll check with some colleagues and get back to you.

Cheers,

Dev
I am logged on, therefore I am.

mfs

I am attempting to calculate the OR given Cohen's d. Does anyone have a reference or knows the equation for the conversion. thanks,  maring