One sided testing of two Poisson distributions?

[Gerenuk]

I want to test if one Poisson distributed result a is large than another one b.
I don't know much about statistics, but I understood the Wiki article about testing normal distribution however they need the number of samples there.

Basically I measure two Poisson distributed variables, I get two values and want to know the probability that one is larger than the other.

Can someone give my a quick reference (online or good book), where I can find my problem as close as possible?

 

[EnumaElish]

You can compute the difference between the two Poisson means and see whether the difference is significantly different from zero under the Skellam distribution.

Or (with a large enough sample) you can assume that the normal distribution will be a reasonable approximation.

 

[Gerenuk]

You can compute the difference between the two Poisson means and see whether the difference is significantly different from zero under the Skellam distribution.

Or (with a large enough sample) you can assume that the normal distribution will be a reasonable approximation.

I think I just about know what to do with the Skellam. But it has a funny Bessel function.
I have both means approx. 500.
What would be the method with the normal distribution?

 

[EnumaElish]

If you know the true variances, the z test.

If you are using computed variances, then technically you should use t test with equal or unequal variances, as the case may be. As the sample size increases, the z-test becomes a good approximation to the t-test (e.g. for n > 40).

 

[Gerenuk]

If you know the true variances, the z test.

If you are using computed variances, then technically you should use t test with equal or unequal variances, as the case may be. As the sample size increases, the z-test becomes a good approximation to the t-test (e.g. for n > 40).

I tried to look through these tests, but I'm not sure what to take. I only know that I measured a single value
x and single value y. Both are supposed to be Poisson (so I expect x+- sqrt(x) and y+-sqrt(y)).

In this case I'm not sure how interpret "computed variance" or "sample size".
I know about mathematics, but not of the formalities of statistics :(

 

[Gerenuk]

I found the following (attachment) in
"An improved approximate two-sample poisson test" (M.D.Huffman)

Just to make sure I got it right and plug in the right values:
I use \alpha=0.05, p=0.90 as sensible values?
I look up z in a table? (i.e. z_{0.95}=1.65, z_{0.90}=1.28)
I estimate \varrho from initial measurements.
Should I use equal counting time d=1 for best results?
By equation (4) I will find how long I have to measure...

 

[EnumaElish]

I tried to look through these tests, but I'm not sure what to take. I only know that I measured a single value
x and single value y. Both are supposed to be Poisson (so I expect x+- sqrt(x) and y+-sqrt(y)).

In this case I'm not sure how interpret "computed variance" or "sample size".
I know about mathematics, but not of the formalities of statistics :(

In a z-test you are assumed to know both means and variances. Poisson is a one-parameter distribution (say k) where mean is a function of k, and variance is also a function of k; so you can derive both means and both variances if you know the k parameter for each of the distributions. Put differently, if you know the mean then you know the variance.