You can do science on any made up word and reach genuine conclusions with flawed premises. Look at phrenology and scientific racism. If you ignore the question “is this thing real?” and skip straight to “what are the associations with this thing”, you’ll find something. It’ll be nonsense, but it’ll be there.
For example, suppose I look at the habits of clowns and roofers. I don’t question why clowns and roofers are associated, I just assume they are and check the data. The data I find will be the overlay of two different trends. I’ll reach all sorts of conclusions about clowns that are only true of roofers, and vice versa. The data will say clowns love a good beer after being outdoors all day, and roofers visit party stores a lot. That’s nonsense, but if I don’t question the association, the data will show it.
Associating small religions with abusive religions is the same mistake. The data will tell you all sorts of things about small and abusive religions, but it won’t tell you which trend belongs to which group, and people will make all sorts of wrong assumptions based on the wacky data.