0
$\begingroup$

I have a data set with one row per subject. Some variables include laboratory parameters for blood chemistry, hematology, etc. I also have some flag variables: any = 1 if the subject experienced an adverse event, 0 if not; and ser_flag = 1 if the subject experienced a serious adverse event, 0 if not.

There doesn't seem to be any difference in the distribution of laboratory parameters between subjects who experienced an adverse event (any=1) and subjects who did not (any = 0). When I do a pair plot of all the lab parameters against each other and color by the any flag, there doesn't seem to be any clustering or separation of subjects.

However, when I do the same pair plot and color by ser_flag - I notice that the 20 subjects who experienced a serious adverse events seem to be clustered together in many of the plots.

What test (if any) can I use to determine if these clusters I think I am seeing are occurring randomly, by chance...or if they statistically significant?

$\endgroup$
0
$\begingroup$

Regarding clustering in machine learning, there is the Hopkin's statistic which basically compares your data to randomly generated data and returns a score of how likely your data shows tendencies of clusters. However, you need to be able to define a distance function between two data points which I don't know is possible in your scenario.

$\endgroup$
  • $\begingroup$ Thank you for this! Gives me something interesting to research! $\endgroup$ – TheCuriouslyCodingFoxah Oct 29 '18 at 13:28
0
$\begingroup$

The easiest test is equality of means of two distributions. Student's t-test

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.