The knowledge gap has been one of those long-standing mass comm research areas, not unlike agenda-setting, that refuses to die (and I wish agenda-setting would, but that's a different issue).
The knowledge gap hypothesis "proposes that the media can increase gaps in knowledge" and as information increases, those of higher SES tend to acquire information more than those lower in SES. A meta-analysis published in the latest issue of Journalism and Mass Communication Quarterly (volume 66, #3, Autumn, 2009, 515-532, for the citationally inclined) confirms the existence of a knowledge gap and reports a consistent moderate correlation between education and level of knowledge.
A meta-analysis is basically a test of lots of tests. In this study, it's 71 effect sizes published in 46 studies. I did a meta-analysis once for a book chapter. I'll never do another. It's statistical hell.
So knowledge levels differ across "social strata," as the study reports. But the study failed to find changes over time, even in increases of publicity, which is rather interesting and "does not offer strong support for the knowledge gap hypothesis," which rests solidly on the idea of changes as time progresses and information becomes more available.
Now for a cool methodological moment.
A smaller gap was found when studies used "belief-type" measures, as "compared to awareness-type and factual-type measures." Fact-based questions are harder, thus increasing the knowledge gap that exists between those of lower and higher SES. This makes sense. "Belief-type" measures are apparently when respondents were asked to list arguments and these were simply counted, regardless of the quality of those responses. "Awareness-type" measures seem to be the ability to answer questions, even evaluations of candidates. If an answer is provides, then you are "aware" of that person regardless of the quality of that response. It's an interesting approach and one necessary given the wide number of studies examined in the meta-analysis. So when people can just say stuff, SES matters less. When they have to be accurate, SES matters more.