I have a data set of 22000 authors and I wish to compare female and male authors with regard to number of received citations, number of reads, etc. However, as the number of citations, etc., is positively skewed and there are many zeros, my distributions are not normal. So, I used Box-Cox transformation to try to normalize my data and be able to use t-test. As I have zeros, first, I added 1 to all citations, number of reads, etc. and then used the transformation. For some of my variables the transformation works and my distribution is normal. However, for some variables the distribution is not normal yet. So, I don't know what to do in this case.
I would recommend not transforming at all and not using a t-test.
Instead, use a count regression model; if you have a lot of 0's and skewed distributions, then a zero inflated negative binomial model might be a good choice.