I'm not familiar with the courses in question, but my impression is that basic statistics is much easier than basic crypto. Obviously, either one could be made arbitrarily difficult, but I've seen more people struggle with things like RSA than with T-tests and distributions.

Also, I think statistics knowledge is more relevant, especially to everyday life, than crypto. Not to say that crypto is unimportant, but statistics is applicable everywhere and changes how you view pretty much everything.

So I think there are some reasons for the stats course to have less attrition than crypto.

It would be interesting to see the attrition of other similar classes at both Udacity and Coursera.

Given that nearly 20,000 students completed ai-class, I think the number for this course will be much bigger than 3-4000.

The success in this courses cannot be measured only by those that achieve a pass or certification. I did one of Udacity's courses, but being short on available time I didn't do all the homework or the final exam. I still learned a lot...