Propensity score matching in JMP

[#1]

Hi all.

I have a combined dataset (cases and controls). The total number of cases is fixed and controls has double the number compared to cases . Main outcome of the study to see the recurrence of disease after treatment. I would like to run propenisty score matching on this data set. Is there anyway i can run propensity socre in JMP?

Appreicate for your advice and suggestions.

Pablo

[#2]

Hi Pablo. Did you ever find a solution to your question? I am also trying to do a propensity score matching with cases and controls, but cannot figure out how to do this JMP. Thanks!

lindsey

[#3]

Hi,
This capability is not currently out of the box functionality within JMP. There was a team in Japan who made an add-in to do this as few versions ago (link below but in Japanese).
https://www.jmp.com/content/dam/jmp/documents/jp/support/propensityscore201412.pdf
I would add this request to the Wish List section of the community page.
https://community.jmp.com/t5/JMP-Wish-List/idb-p/jmp-wish-list
Best,
Chris Kirchberg

[#4 · Accepted solution]

Hello,

While JMP doesn't have  Propensity Score Analysis (PSA) platform, you can definitely accomplish PSA in JMP by regressing the Treatment/Control factor on the suspected covariates by using the Fit Model (Logistic) platform.  The Pscores are the XB portion of the model. You can use the PScores to identify matches or as weights. Also, given that JMP connects with R, you can take advantage of the PSA algorithms that R has, including the Optimal Matching routine.  If you know how to do PSA in a different software, you could try using the platforms that JMP has to reproduce the results.

Cheers,

MG

[#5]

FYI, this recent paper has some cautions about the use of propensity scoring in situations where matching is possible: Why Propensity Scores Should Not Be Used for Matching.

[#6]

Hi Xan,

Thanks for the heads up.  Gary King is definitely an authority in this space.  Observational studies are exciting to say the least.

Sincerely,

Matt

[#7]

Gary King (one of the authors of the paper @XanGregg linked to) also gave an "International Methods Colloquium" on this topic and has the video available on YouTube. I've included it below, and here is the direct link.