# Paired ratio significance test -- is Chi-squared test correct? :|

#### PB_1989

##### New Member
Dear Community,

Thanks very much for picking up this thread.

I’m having some trouble finding the right statistical test for paired ratios (n = 10,000 patients), having not been able to do my usual thing of googling my way to certainty.

I have data in the following layout:

The data are paired e.g. before and after an intervention, and we are looking at the differences in the ratio of canceled appointments : total appointments (in the hope that our intervention reduces cancellations).

Is it correct to use a chi squared test for something like this? I’m trying to do this in R.

Thank you very much for any and all feedback.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Hmm, a direct solution isn't popping into my head, but I will paraphrase your scenario to make sure I understand it.

You have the individual RATEs of no-shows for 10,000 patients, then you instituted an intervention to decrease no-shows (e.g., reminder calls or something). Now you have the individual rates of no-shows for this exact same 10,000 patients post-intervention. All patient got the intervention.

So as I alluded to, these are the exact same patients in both groups, no new patients or lost patients? Do you have secondary characteristics you would like to control for?

I suppose a Poisson regression model could be fit predicting the post rate given baseline rates and controlling for characteristics (e.g., age, etc.)

#### Karabiner

##### TS Contributor
I’m having some trouble finding the right statistical test for paired ratios (n = 10,000 patients),
Why should you bother about a statististical significance test with a huge data set like this?
The test deals with the question whether the difference might be exactely 0.0000000 in the
population your 10.000 cases were sampled from. You can judge from your dataset whether
there was an increase or decrease.

What could be interesting (as mentioned by hlsmith) is the possible association of
the difference with patient variables / subgroup characteristics.

With kind regards

Karabiner