Bobward at Crosstabs does it much better than I ever could:

The whole idea of a statistically valid sample, i.e. one that represents the broader universe being sampled within an accepted error margin, is that a) each member of the survey universe has an equal chance at being selected in the sample, and b) the sample is selected at random.

Interactive surveys – or online surveys driven by email from a panel of pre-screened participants – such as those Zogby uses, completely ignore the first tenent of polling. Every member of the universe (in this case, voters) does NOT have an equal chance of being selected. A sample from his email panel produces at best a survey of his entire panel of participants.

To be sure, telephone polling has its limitations, and it grows less reliable every day. As more and more people move to mobile phones as their primary phone, and call screening and blocking enjoy wider practice, every voter having an equal chance being reached gets muddied. One phenomenon that has delayed this impact on campaign polling is that the age group most likely to eschew land line phones are also the least likely to vote – young adults. But the limitations of telephone polling are nowhere near those of opt-in email panels.

The truth is that on-line surveys are CHEAP, and that is why the media loves them. You don’t pay for trained interviewers, phone center managers, quality control, long distance telephone time, etc. You program a web site and send email. The cost difference is night and day.

I know my comrades here at Crosstabs.org regularly dismiss the Zogby Interactive numbers, and well they should. While they may be interesting, they have nothing to do with statistically projectable survey research.