Polls Skewed By Oversampling One Party, Other Factors

Lancaster Online

Polls are generally skewed by oversampling one party or by using weighting factors ostensibly to adjust the polled results to anticipated turnout. But polls are not good at predicting turnout. Weighting factors based on exit polling have proven to be skewed.

Most polls, including the local Franklin & Marshall College poll, use a large oversample of Democrats. Many undersample independents. Results can be quite sensitive to this. Many polls appear to assume a 2008 turnout model that is unlikely to repeat.

A New York Times poll in September assumed Democrats would outnumber Republicans by 33 percent in Florida and Ohio, and by 40 percent in Pennsylvania. In 2010, Pennsylvania Democrats had just a 3-point turnout advantage. Over eight years that advantage has been 4 points. Yet the F&M poll oversampled Democrats by 11 percent. Rasmussen Reports determined over many years the national voting electorate is 35.4 percent Republican, 34.0 percent Democrats and 30.5 percent unaffiliated.

So contrary to history, many polls are heavily weighted to the left. Pollster John McLaughlin found the last four elections in Florida produced a turnout of 37 percent Democrats and 38 percent Republicans, yet polls there are weighted 6 percent to 9 percent in favor of Democrats. In Ohio, he found that exit polls had a 2 percent Democratic edge, but three polls used Democratic weightings of 8 percent to 10 percent. In Virginia, actual turnout edge was 1 percent Republican, but four polls used 5 percent to 11 percent Democratic weighting regardless.

The only possible conclusion to this level of skewing is the pollsters want a desired outcome to shape opinion rather than to document it. It is simply an extension of the heavily documented Democratic bias of the media.

Consider the comments of F&M’s Dr. Terry Madonna, a college professor and former Democratic Lancaster County commissioner prior to his polling career.

“It’s going to be a huge climb for [Mitt] Romney to win this state,” Madonna said. “It’s going to take probably something extraordinary such as a major shock to the economy or some kind of foreign affairs crisis that hurts the president.” He added, “I don’t know that there is anything [Romney] can do, per se.”

Looks like our local unbiased expert was very wrong.

When polls are “unskewed” by re-weighting them based on the historical turnout from Rasmussen Reports data, the results for Gov. Romney are significantly better. This can be monitored on a daily basis at unskewedpolls.com. When this organization unskewed the F&M poll, we saw Obama’s 9-point lead translated into a tie.

Most polls have had Obama under 50 percent all along, and history shows that undecideds generally break late against the incumbent. This further suggests that it will be difficult for Obama to win. Pollsters know this effect.

As the election nears, polling results must be corrected for the polling organizations to save face and claim accuracy in the end. First, they switch from polling registered voters to polling likely voters. With two-thirds of registered voters not voting, polling registered voters is hardly a scientific method. The September F&M poll reached a paltry 392 likely voters in Pennsylvania with a margin of error of almost 5 percent. By contrast, the recent Pennsylvania poll by Susquehanna Polling & Research included 1,376 likely voters and gave a 4-point edge to Mitt Romney.

Now that the polls have shifted dramatically toward Romney, we are to believe the change results entirely from one debate performance. This is difficult to accept. There have never been that many undecideds in recent months, and debates have not been shown to have this great an effect on polls.

Comparing F&M poll results to Susquehanna Polling results for the U.S. Senate race, you would think Tom Smith turned a 10-point deficit into a 2-point lead. There has been no debate in that race. The reality is somewhere in-between the two explanations, but inappropriate skewing to the Democratic side has been pervasive and extensively documented.

F&M pollster Berwood Yost claims: “In the Electoral College, Obama continues to have an advantage.” But this claim is based on the skewed polls. A state-by-state analysis model by the University of Colorado, accurate since 1980, predicts a comfortable Romney win on the basis of economic data. Its updated model predicts a Romney win of 330 to 208 electoral votes and includes Romney winning Pennsylvania.

When you think about it, how could Obama have been doing so well with an economy worse than Jimmy Carter’s?

Read more: http://lancasteronline.com/article/local/765070_Polls-skewed-by-oversampling-one-party–other-factors.html