Well, 23 days until the election. Romney is suddenly awake, having trounced Obama in their first debate, and Biden was horribly rude with Ryan while Ryan was almost scholarly. The White House and the State Department are taking public dumps all over each other in a story where the president is out of touch at best and a public liar at worst. And i was sick for a whole week and did not write anything anywhere.
My issue for today is the manipulation of the Gallup Poll, and polls in general. Historically Gallup has been the most accurate at predicting presidential winners. Rasmussen also has an excellent track record over a shorter period. Polls are samples of the general population; therefore, if the sample mix is wrong the results will not be applicable to the population as a whole.
In dealing with statistics the issue of sample accuracy is a well recognized problem. This applies to random samples of a normally distributed population. But our population is not normally distributed, and the sample to represent it is not random. Indeed, it is carefully constructed to represent our nation as a whole. With 300 million people, +/-, such a constructed sample of 1500 people represents 1 in 20,000 people being used as a sample of the whole. That sample must be divide among Republicans, Democrats, Independents, men, women, blacks, hispanics, etc, etc. Now how do you get the right mix? This is a different problem of statistical accuracy- trying to represent a non-homogeneous population in a single sample. How to do this?
The answer is you do very careful population modeling and many, many samples. And many, many erros are made, and the results then passed off as truth.
One simple error is to poll Americans rather than likely voters, or registered voters as opposed to likely voters. No matter what is said about women, blacks, hispanics, Democrats, etc, there are certain behavioral issues that must be reflected in any poll. Among these- a much lower proportion of black registered voters actually vote than whites; those under 25 vote far less often than those over 45; Democrats as a whole vote much less often than Republicans. These are not opinions, these are repeatedly observed facts. One may argue all day long about the relative merit of British and German battlecruiser design and construction. In action, three British battlecruisers were sunk in 15 minutes, while the Germans lost only a single ship scuttled because of uncontrollable fires. Those are facts.
Thus the mix of a polling sample can have a profound effect on its results. This has been harped on in the blogoshpere and in responsible publications (and I choose that label quite deliberately). One poll after another has a population skewed heavily democratic, or looks only at registered voters, or asks simplistic questions to screen likely voters.
These poorly constructed polls have shown Obama 'winning' the race for months. Rasmussen, who polls only likely voters, has had Romney ahead by 2-5 points for months, and has been dismissed as Republican biased. But Rasmussen models only likely voters. It is not a poll about public opinion or voter distribution. It is about who really votes, and what they like. It is a raft of other polls, with Democratic bias up to +11 percent, that show Obama winning by 2-4 points. Allowing for the fact that democrats are voting 90-10 for Obama, for him to be 'winning' by 4 points in a +11 democratic sample is a terrible underperformance.
This statistical legerdemain was exposed when Gallup converted, as they do every election cycle, from registered voter polling to likely voter polling. Suddenly Romney was up 5 points. Wow. No surprise for those following polls closely, but a shock, a shock! for the Obama team.
The response disgusts me. Obama put pressure on Gallup again to change their polling mix. Obama asserted their mix was too white, too Republican. Well, that's who votes. No one expects a Democratic turnout like 2008, yet a surprising number of polls have made sample assumptions based on just that- a historically high amount of black and youth turnout. One poll even persists in assuming a Democratic turnout even higher than 2008.
Gallup resisted, but last week they folded. Gallup adjusted their polling mix, reducing white voters by 5-6%. The result is that for the first time in two years Obama has an approval rating above 50% and is again 'leading' the Gallup poll.
This is bad polling. It is bad journalism. And it is outrageous politics. If the President can strong-arm the press into printing only favorable stories, how is that any different from a state controlled press? This behavior is no different than Biden's in his 'debate'. Biden simply tried to talk over his opponent rather than counter his arguments. The heckler's veto- shout your opponent down when you have nothing to say yourself. Make up favorable numbers when real data is unfavorable, like our employment report last month, a pure fiction in the eyes of nearly every serious observer.
Barbara Tuchman, in in her book 'The March of Folly', defined folly as policy pursued contrary to self interest. One of the major components of folly was the selective use only of favorable data. This behavior preceded Britain's loss of the American colonies and furthered the debacle in Vietnam.
Now we have a President who is manipulating the press into reporting only stories favorable to him. Who is committing folly here? Him? Or us? Obama is a vacuous and ineffective man undeserving of re-election, but if we fall for this kind of poll and press manipulation and re-elect him, we are the ones guilty of folly. And we are the ones who will suffer.
A real democracy demands of its citizens greater intelligence and responsibility. We will see if we are truly worthy of a democracy, or if we fall into state socialism by approving of a state manipulated media.
Ted
My issue for today is the manipulation of the Gallup Poll, and polls in general. Historically Gallup has been the most accurate at predicting presidential winners. Rasmussen also has an excellent track record over a shorter period. Polls are samples of the general population; therefore, if the sample mix is wrong the results will not be applicable to the population as a whole.
In dealing with statistics the issue of sample accuracy is a well recognized problem. This applies to random samples of a normally distributed population. But our population is not normally distributed, and the sample to represent it is not random. Indeed, it is carefully constructed to represent our nation as a whole. With 300 million people, +/-, such a constructed sample of 1500 people represents 1 in 20,000 people being used as a sample of the whole. That sample must be divide among Republicans, Democrats, Independents, men, women, blacks, hispanics, etc, etc. Now how do you get the right mix? This is a different problem of statistical accuracy- trying to represent a non-homogeneous population in a single sample. How to do this?
The answer is you do very careful population modeling and many, many samples. And many, many erros are made, and the results then passed off as truth.
One simple error is to poll Americans rather than likely voters, or registered voters as opposed to likely voters. No matter what is said about women, blacks, hispanics, Democrats, etc, there are certain behavioral issues that must be reflected in any poll. Among these- a much lower proportion of black registered voters actually vote than whites; those under 25 vote far less often than those over 45; Democrats as a whole vote much less often than Republicans. These are not opinions, these are repeatedly observed facts. One may argue all day long about the relative merit of British and German battlecruiser design and construction. In action, three British battlecruisers were sunk in 15 minutes, while the Germans lost only a single ship scuttled because of uncontrollable fires. Those are facts.
Thus the mix of a polling sample can have a profound effect on its results. This has been harped on in the blogoshpere and in responsible publications (and I choose that label quite deliberately). One poll after another has a population skewed heavily democratic, or looks only at registered voters, or asks simplistic questions to screen likely voters.
These poorly constructed polls have shown Obama 'winning' the race for months. Rasmussen, who polls only likely voters, has had Romney ahead by 2-5 points for months, and has been dismissed as Republican biased. But Rasmussen models only likely voters. It is not a poll about public opinion or voter distribution. It is about who really votes, and what they like. It is a raft of other polls, with Democratic bias up to +11 percent, that show Obama winning by 2-4 points. Allowing for the fact that democrats are voting 90-10 for Obama, for him to be 'winning' by 4 points in a +11 democratic sample is a terrible underperformance.
This statistical legerdemain was exposed when Gallup converted, as they do every election cycle, from registered voter polling to likely voter polling. Suddenly Romney was up 5 points. Wow. No surprise for those following polls closely, but a shock, a shock! for the Obama team.
The response disgusts me. Obama put pressure on Gallup again to change their polling mix. Obama asserted their mix was too white, too Republican. Well, that's who votes. No one expects a Democratic turnout like 2008, yet a surprising number of polls have made sample assumptions based on just that- a historically high amount of black and youth turnout. One poll even persists in assuming a Democratic turnout even higher than 2008.
Gallup resisted, but last week they folded. Gallup adjusted their polling mix, reducing white voters by 5-6%. The result is that for the first time in two years Obama has an approval rating above 50% and is again 'leading' the Gallup poll.
This is bad polling. It is bad journalism. And it is outrageous politics. If the President can strong-arm the press into printing only favorable stories, how is that any different from a state controlled press? This behavior is no different than Biden's in his 'debate'. Biden simply tried to talk over his opponent rather than counter his arguments. The heckler's veto- shout your opponent down when you have nothing to say yourself. Make up favorable numbers when real data is unfavorable, like our employment report last month, a pure fiction in the eyes of nearly every serious observer.
Barbara Tuchman, in in her book 'The March of Folly', defined folly as policy pursued contrary to self interest. One of the major components of folly was the selective use only of favorable data. This behavior preceded Britain's loss of the American colonies and furthered the debacle in Vietnam.
Now we have a President who is manipulating the press into reporting only stories favorable to him. Who is committing folly here? Him? Or us? Obama is a vacuous and ineffective man undeserving of re-election, but if we fall for this kind of poll and press manipulation and re-elect him, we are the ones guilty of folly. And we are the ones who will suffer.
A real democracy demands of its citizens greater intelligence and responsibility. We will see if we are truly worthy of a democracy, or if we fall into state socialism by approving of a state manipulated media.
Ted
I was able to watch the Presidential and Vice-Presidential debates via youtube. My perception was that Romney came out with better arguments in the first debate, but Biden had better arguments than Ryan in the second. Much of the debate seems to center around competing economic theories. And both sides claim the other misrepresents what they say.
ReplyDeleteI do wonder how much these debates affect peoples' vote? One thing is for sure: prediction is a tricky business. The Venezuelan presidential election, for example, was a week ago and supposedly Capriles was really going to give Chavez a run for his money. But in the final polls, Chavez trounced him, with 55+% of the vote. Now, whether the vote was manipulated is another matter, but the bottom line is that Chavez is looking at another 6 years in office (assuming his health holds up). So what happens in the U.S. next month remains to be seen...though afterwards, I'm sure more than a few people will look at the results and claim, "I knew it would turn out like that..."
DG