Thats a pretty universal problem with inferential stats. Randomization is difficult and all or most test samples have some biases. Statisticians and pollsters seem to be doing alright for themselves anyway; there results get more trust than I think is strictly warrented sometimes.
Ive seen zogby's bush approval data arranged with other polls and it seems to be telling the same story everyone else's is. Here, have a look for yourself:
![](http://www.pollkatz.homestead.com/files/approval-data_files/zzzmainGRAPHICS_14808_image001.gif)
Maybe you can tell me why he is off the mark with Hillary, but able to keep his polling in the mainstream with Bush?