CU ratings are based on actual survey reports from owners of the cars being rated. The frequency of repair records are subdivided into various categories. The data base is probably the largest "private" one available. I believe they are as correct and unbiased as you will find. Of course, not every car in a group has every problem, and some cars may be totally trouble free, but the averages for a given car make and model are a real indication of the likelyhood of problems. If you go back and look at certain car models that show problems, you'll also probably find that TSB's were issued at some point. TSB would be an indication that the data are correct.
JD Powers is also mentioned. They survey individual owners for initial ownership experience, pretty much the car as delivered and problems within the first 90 days, I think. I don't think there's any bias in their data either.
In the end, you get what you get, but the ratings can tell you the odds as to whether your car will be a problem or not.