The census is a level playing field. Each religion has its own internal measures of membership, but censuses simply ask what each person identifies as.
So for example, when the Mormons claimed 1.1 millions members Brazil in 2010, and yet only 226,000 identified themselves as Mormons in the census, it seems fair to conclude that Mormons in some real meaningful sense overestimate their membership.
At the same time, when Jehovah’s Witnesses claimed 700,000 members in Brazil in 2010, and yet nearly 1.4 million identified themselves as members in the census, it also seems fair to conclude that in a real meaningful sense Jehovah’s Witnesses underestimate their membership.
So there really is a lot of empirical data to support JW figures and to show that they underestimate their membership compared with other groups.
See this chart of census results compared with official memberships.