representative of likely voters on Election Day.
From poll to poll, the composition of the half or less of those chosen randomly who actually respond fluctuate. Thus virtually all polling organizations except Gallup re-weight their sample statistically so it looks like the projected electorate on next Election Day on "party ID", ethnicity, gender, income, and other factors known to be strongly correlated with "candidate preference".
Gallup does not weight by "party ID", thosgh it does weight by other factors. And its own results from this weekend show a dramatic jump in "enthusiasm" for one's party's candidates. This is a big problem and possible source of BIAS for Gallup's methodology, since Republicans temporarily "energized" by the theatrics at St. Paul may be much less likely to refuse a Gallup interview, though no more likely to show up at the polls in November. Also, during R convention week, Republicans are more likely than Democrats and Independents to be sitting home in front of the TV and thus be available to get the call from Gallup.
Exactly the same thing happened four years ago after the Republican convention. Gallup and Rasmussen (then using the Gallup methodology) showed big jumps for Republican candidate preference, while other organizations weighting by party ID showed tiny gains for the Rs. Rasmussen then blamed their error on the methodology they used and switched to the "weight by party ID" camp. Don't be alarmed; the next Gallup poll should show a dramtic swing back toward Democrats.
See
http://www.depts.ttu.edu/hs/hdfs3390/weighting.htm .