I asked him to explain the general similarity of the results (lots and lots of Christians, tiny amount of A/A's) between to compared groups if self-reporting was so inaccurate and unreliable.
The comparison is not between two groups. It is comparing a subgroup to its group. You are in a way simply taking the large group (people) and comparing it to a sample of that group (people in prison) and getting similar results because going to prison isn't something that is particularly related to religious identification (or perhaps a coincidence).
In essence, the reason you are getting similar results is possibly due to {people in prison} being a representative sample of {people}. Those {people} statistically identify with various religions, and representative subgroups will do likewise. If everyone rolled a ten sided die when they were asked (purely random selection), 1-8 they identify Christian, 9-10 Other, then you'll find you'll get basically the same results as your Christian identify and Prison Christian identify statistics.
In that case, self reporting would be irrelevant.
Now if they didn't roll a die, but instead weighed their upbringing, social group, perception of being judged, actual beliefs etc in a complex series of pulses in the brain and 80-odd percent decided they would say they were Christian, then we'd also get the same results. It doesn't mean that the majority reporting what they truly believe, it could just be what they want someone to think they believe.
What is the percentage of honest answers? An unknown number. You assume most people accurately report, rr doesn't accept the assumption. I agree in a way. Many people say they are Christian, but when further questioned ("You believe in God, Christ and Heaven/Hell and all that?", "Well...there has to be something I guess") will agree that it is only culturally. Religiously they are usually little more than agnostic or perhaps Deists etc.