0
$\begingroup$

Coming back to statistics after a long time, so please forgive any ignorance.

I'm curious how I might be able to argue for statistical significance of a test. Specifically, I'm after a p-value for my test.

I ran a test along with a control with approximately the same, reasonably large group size. For example:

Group A (control) = list of incidence rates in units of (seizures / (rat * yr))
                  = [0.01, 0.014, ... 0.012]

Group B (test)    = list of incidence rates in units of (seizures / (rat * yr))
                  = [0.012, 0.01, ... 0.011]

I want to know whether my test shows a statistically significantly difference - whether the frequency of the event was meaningfully reduced.

It appears that this magic calculator does exactly this. How can I do this in python?

enter image description here

http://www.cdc.gov/nhsn/PS-Analysis-resources/PDF/StatsCalc.pdf

$\endgroup$
1
$\begingroup$

Yes, you can obtain a p-value, but it will take some effort to assure the model you pick models your data well.

Your data are most likely a count of seizures over some measure of time. If your time of observation is the same for each experimental unit (rat), consider using a Poisson regression. The dependent variable is the number of seizures for each rat and the independent variable is Group (A/B).

You may also consider negative binomial regression if the Poisson model does not fit well.

$\endgroup$
  • $\begingroup$ yes the duration and start of observation is exactly the same between the groups $\endgroup$ – tarabyte Feb 18 '16 at 5:45
  • $\begingroup$ can you elaborate more on the Poisson regression and how that can lead me to a p-value using scipy? statsmodels.sourceforge.net/devel/generated/…, docs.scipy.org/doc/scipy/reference/tutorial/stats.html $\endgroup$ – tarabyte Feb 18 '16 at 7:03
  • $\begingroup$ is there any additional help you can provide? $\endgroup$ – tarabyte Feb 20 '16 at 0:57
  • $\begingroup$ This seems like the most standard case of Poisson regression, if you have the counts (rather than rates). The rates alone are (almost) meaningless (0.01 could be 1 event per 100 time units or 1000 events per 100,000 time units and the amount of information contained in the two is enormous). $\endgroup$ – Björn Oct 17 '18 at 6:06

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.