Yes, it should be tested, not just assumed as I had said, but let's remember, Robert is (if I understand the methodology correctly) adding approximately 300 variates to form each of the 29 final numbers. Also, as each of original 300 numbers are separated by 29 days, the autocorrelation should be negligible. I would be surprised if the Normal distribution was not a good approximation. I did however do a QQ-plot on the 29 numbers and found rough agreement with the Normal. Not perfect, but not bad for a sample of 29. However, the roughness of the agreement does, as people have pointed out, affect the levels of significance that he has claimed, Overall it is not strong evidence of a cyclical pattern.

I understand also he used the Binomial distribution to obtain the significance levels. I'm not sure why - it would have been simpler to use the Normal. Maybe I've missed something. Of course, it would be better to model the daily rainfall, and test for cyclicity another way. Robert's statistical methodology lacks power and only has approximate levels of significance.

Peter