0

Suppose I have a data from Center A and Center B. I record the number of arrivals per hour(0,1,2,3,4,5) for center A and center B. I need to test whether the means from two centers are equal. My data has three columns. First column is Number of arrivals per hour, second column is center A frequency and last column is center B frequency. I also assume that the distribution of arrivals at both centers are Poisson.

Can I use a z test in this case?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Hello
  • 121
  • 1
  • 8
  • How large are your sample sizes? If your samples are not large you could always use a Wilcoxon rank sum test (https://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test) as a nonparametric alternative to the $t$-test. Just be aware that the hypothesis you are testing with the rank sum test is slightly different than the hypothesis considered by the $t$-test. – dsaxton Aug 06 '18 at 19:39
  • 350 in center A and 400 in center B. – Hello Aug 06 '18 at 19:42
  • 1
    In that case you are probably fine using a $t$-test. Also remember that the Poisson distribution itself is well-approximated by a Gaussian for large $\lambda$. – dsaxton Aug 06 '18 at 19:44
  • 1
    Okay. But in my case the sample mean for center A is about 2.31 and for center B is 2.60. Would it still be okay to use t -test? – Hello Aug 06 '18 at 19:55
  • 1
    That information is not consistent with a Poisson assumption, for center B ought then to be observing about 5% (or 20) hours with more than five arrivals per hour. – whuber Aug 06 '18 at 20:04
  • 2
    Yes, the underlying distributions need not be normal if your sample sizes are large. – dsaxton Aug 06 '18 at 20:27

0 Answers0