After skimming through the methodology used in the latest report, I don't think there's any reason to suppose the numbers are not accurate.
Which report? Whatever the merits of the methodology used in any particular measure, for diachronic analysis I think it's always preferable to compare statistics that were gathered in the same way. That's why I think it is dubious to compare a result from one measure, one year, and a different measure from a different year, and act as though you can work out an increase or decrease between the two. It's a different measure. If the different measures were conducted in the same year the chances are you would have got a different result. So what would Joker10 do then, work out the increase within that year or what?
However what I thought was most interesting about the census figures is not simply that they differ from Watchtower figures for the 1990s, it's the fact that they show a completely different trend that is significant. According to the figures in the Watchtower the number of Witnesses in the US in 1990 was 816,417 and that increased to 945,689 in 2001. That is an increase of 16%. Contrast that with the census figures that show a 4% decrease over the same period. Combine that with census results from Canada, Australia and New Zealand that similarly contradict the trend shown in the official Watchtower figures and I think we can say there is a significant discrepancy.
I agree with you by the way that JWs are nevertheless doing a lot better than most other churches in keeping members. Most other Christian groups are declining much faster in Western countries.