Why most polls done in Nevada are garbage

A few words about all of these polls on the presidential race in Nevada: Don’t believe them.

Sound familiar? Yes, I was telling you the same thing two years ago when every poll (almost) showed Sharron Angle would be the next U.S. senator from Nevada. That didn’t happen, and all of those polls were wrong for different reasons, which eventually comes down to the same reason:

You hear the cliché all the time that polls are snapshots in time. That is true, but it also highlights the basic problem with most polls. That snapshot may be worth a thousand words, but only when the picture is taken. So the best pollsters – this is the key – know how to weight the results to fit the picture that will exist on Election Day – that is, what the turnout actually will look like.

The reason Harry Reid’s pollster, Mark Mellman, nailed the result in 2010 was that he correctly forecast what the demographics of the turnout would look like – how many Hispanics, how much Clark County would be of the total, the difference between GOP and Democratic turnout. And the better pollsters know how to change their models as the campaign evolves, to adjust for whatever atmospherics require some adjustment.

One of the more obvious factors is voter registration, especially in a high-turnout, presidential year, when 80 percent or so of voters will turn out, making likely voter samples not as important as in off-years, when turnout is about 50 percent here. The other important factor often not employed by pollsters in Nevada – especially those who have not been conducting surveys in the state for very long – is history.

Both of those indicators tell me that a series of recent polls do not take into account either the surge in Democratic registration or recent history – i.e. the last two cycles.

I acknowledge up front that 2012 is not 2008, much less 2010. The Democratic wave of 2008 is unlikely to be duplicated here four years later, with a devastated economy. It is possible that Democratic turnout will be depressed compared to the previous two cycles, which could dramatically change the result. But what Democrats here know – as do good pollsters – is that Republicans traditionally turn out in greater numbers than Democrats --  anywhere from 4 to 6 percent. So you have to adjust your numbers for that fact as well as the registration changes. And the greater the registration advantage the Democrats have, the less the GOP turnout edge affects the share of the vote the Democrats will have in the end.

Four recent polls in this state -- PPPSuffolk, Rasmussen and Survey USA -- have shown the presidential and Senate races close. But are they? And do they take into account the registration surge and history?

For my premium subscribers this Sunday, I am going to delve into the crosstabs of all of these surveys and show you anomalies in each of them that should make you wonder about how accurate they really are and whether we are seeing a repeat of 2010 and 2008 when the polls here were skewed Republican. I will also write about how history indicates just how much trouble the GOP could be in here -- from president on down -- if historical turnout trends hold.

One of my problems with what statistics guru Nate Silver of the New York Times does and what Real Clear Politics does is they average together polls without regard to the quality of the instruments. (I say immodestly that I told Silver in 2010 more than once that his sample was polluted by offal surveys and that Reid was still likely to win here.) If you average bad data with good data – and there rarely is good public data in Nevada – you will have a skewed result. (More on this, too, for my premium subscribers Sunday.)

But my bottom line for now is: Remember 2010. Nearly every poll you saw showed that Angle was going to win -- as did Angle's internals. And she lost by nearly 6 points. Six points!

The raw numbers this cycle are very similar in Clark County to what they were in 2008 -- about a 125,000-voter lead (it actually is going to be slightly larger this time.) The way it works is that the South makes up 70 percent of the vote, and if you don't take that into account in your poll, you won't show the kind of raw number lead that Democratic statewide candidates are likely to have (Obama's will be greater than Rep. Shelley Berkley's) that make Republican candidates chances less and less real.

Despite what all of those polls say, Romney's path to victory in Nevada now is much more problematic than any Republican will acknowledge. I'll revisit the math for premium subscribers on Sunday. But let me give you some history:

In 2008, when Democrats had that 125,000-voter edge, Obama won Clark County by more than 122,000 votes, or 19 percent. John McCain never had a chance after that and lost by 12 points. The edge is similar four years later, and while Mitt Romney has contested the state in a way McCain did not, the math isn't much different. Unless the Democrats turn out in record low numbers relative to Republicans, Romney cannot win unless independents overwhelmingly go for him. And none of those polls show that (indeed, even GOP-leaning Rasmussen shows Obama winning indies by 10).

Any pollster who takes into account all of those factors would come out with a survey showing the president up by a half-dozen points or so. Any poll that doesn't has a turnout model that either doesn't make sense, is partisan-biased or is simply garbage.

Comments: