How to check gender and age matching

#1
Hello girls and guys,

I have worked on a study for which I was blinded in the first part. I always assumed that the patients and controls were gender and age matched because I was told so.
Now I have the list of birth dates and gender affiliation and I would like to state in my paper/thesis that the subjects are matched. Of course, I can't just say that without having proven or checked it. Now my question: How can I test this with the above mentioned knowledge? Through Google I came across instructions on how to match in advance, but unfortunately not on how to test for matching afterwards.

FYI: I have:
23 patients,17 men, 6 women, mean age 63,9 years (on 31. December 2016)
20 controls, 17 men, 3 women, mean age 58,9 (on 31. December 2016).

Any help would be very appreciated. Thank you in Advance,

Zweistein
 

hlsmith

Not a robit
#2
So you truly don't know if they were matched. Also, what matching process used would be beneficial to know, so you can test whether it was successful.
 
#3
I was hoping there'd be a test to see how two groups match. Just like you can with t-test or ANOVA also match the difference of a measurement.

Unfortunately, I don't know much about the acquisition of the test persons which is some years ago. The Acquisition happened in at least 2 steps as far as I know. It's not that I don't trust the people, but I want to Control/measure the matching for myself. So you think the only reasonable way is to get more Information about the acquisition? I was hoping to "just" prove/evaluate this by myself. I can't change the group anyway, unless I take subjects from the study.

Thank you very much for the quick answer.
 
Last edited:

hlsmith

Not a robit
#4
In order for you to publish or disseminate results based on these data, you should ideally know and report the matching mechanism. It would allow the reader to understand the internal representativeness of the sample. You usually match on a confounder to remove imbalances between the groups, and you do it based on criteria. If you think gender is a confounder, well controls have more females, why - what caused this to happen. Also, age is not exactly the same. As a reviewer I would say, well the pool of controls must not have been that large if this is the best they could do. Also, was age considered a bigger confounder then sex and criteria tried harder to match age? Were there more women eligible controls, but their ages were too high, so they didn't match. Regardless if you can move forward you should try to know exactly the source of your sample, in that the process could introduce selection bias to your statistical question and you wouldn't know, or perhaps matching was not needed and performing it may have underpowered the analyses, when those variables could have just been controlled for in a model incorporating covariates.

In randomized control trials, characteristics are supposed to be balanced, but if a researcher sees that they aren't, they can still control for them in the model after randomization to address residual differences. Your sample is pretty small, so this may not be an option for you. You can't see if they were match if you don't know the criteria and probably source population, but you can still check their balance. Tests can be over lauded, and underpowered in many cases. Typical measures for balance can be based on:


absolute standardized mean difference (ASMD; also referred to as the absolute standardized bias or the effect size (ES)) and the Kolmogorov-Smirnov (KS) statistic.
 
Last edited: