By Jon Frandsen, Senior Editor August 18, 2008 The time to start taking polls seriously is drawing near. It usually begins with the conventions. It's always interesting to see how high a candidate's post-convention bounce will go and how long it will last. By mid-September, the race should start to take clearer shape as voters pay closer attention and the number of undecided voters begins shrinking. But be careful this year -- polling is unusually tricky. There are huge numbers of newly registered voters and it's hard for pollsters to know how many are "likely" to show up on Election Day. Voting by minorities, who tend to distrust pollsters, was way up in the primaries and will be again in November. And pollsters are concerned that some white voters who don't intend to vote for Obama will say they do so they don't appear to be racists. All these factors can throw polls off -- sometimes way off. A little caution and knowledge can keep you from feeling whiplashed on days that results go every which way: Pay closer attention to tracking polls (polls taken daily and averaged for the past three days) than others because they tend to be less volatile -- and when there is a surprising spike or drop you can tell pretty quickly whether it's a polling anomaly or a sign of a trend. Gallup's is well respected. Skip the real outliers. If a poll is way off from others -- well, it's probably way off. But if it's by a respected company and keeps getting similar results, they may be on to something. Look at the actual results to see if there's something in the way the question is phrased that might be making the outcome more accurate -- or less so. Pay more attention to registered voters than likely voters until we get much closer to the election. Trying to determine now who is truly likely to vote in November is an art, not a science, and not an easy one. Even when the election is closer, picking an accurate sample of likely voters could be especially difficult this year. A Gallup-USA TODAY poll in late July illustrated the confusion that can be caused when it showed, like most polls, Obama with a slight lead over McCain among registered voters, but McCain leading Obama among likely voters. That led to a thoughtful debate among pollsters about the difficulties in selecting likely voters. Respected pollster Mark Blumenthal at Pollster.com suspects that that particular poll left out too many younger and new voters since one of the screening questions for likely voters involves whether you voted in 2004. For a different example of how likely voters can be selected, here's how one respected pollster explained his technique. Finally, remember that while national polls are interesting and can raise important questions (why is the race so close when the national mood is so anti-Republican, for example), they are ultimately meaningless. Instead, look at polls in individual states. Pollster.com and RealClearPolitics.com update state polls whenever new ones are available and show trend lines in state. Of course keep an eye on battleground states such as Ohio and Florida, but look now and then at states that historically vote for one party or the other to see how the candidate from that party is doing. The fact that McCain is going into his convention with narrow leads in genuine Republican Country (states such as North Carolina and Alaska) doubtless alarms him and the party. Polls in toss-up states are also the best place to try to gauge whether third party candidates can cause much mischief. They won't dent national totals, but they could tilt a key state or so in one direction or another.