Post by: luna_lava
Back when a computer took up two floors of the Engineering Faculty (UNSW) one of my favourite books was “The Use and Abuse of Statistics” by W.J. Reichmann (… I know a bit nerdy). Since that time, not a lot has changed with the central theory of Statistics and the concept that when you “sample a population” your sample must be representative of the larger group.
Most of the polls published in the mainstream media (MSM) are based on the assumption that they are sampling the population of Australian voters to draw inferences on their intentions and to gauge a reaction to a particular issue
One of the traps for the guys who design these polls is the belief that:
“As the size of a sample is increased it approaches more closely to the population itself ….there is safety in numbers”
In 1936 an American magazine forecast that the Republicans would sweep all before them in the Presidential elections. A huge sample of 10 million people was selected and, of these over 2 million replied to the questionnaire. The size of the sample, coupled with the fact that the same sample had previously been used successfully in four earlier elections, was taken as a sure guide to the outcome of the 1936 election. The sample returns indicated that Franklin D Roosevelt would be defeated whereas, in fact, he was elected with one of the largest majorities ever recorded.
The apparent reason for failure to make a correct prediction was the fact that the sample chosen was not representative of the total population of American voters. The magazine’s inquiries had been addressed to its own readers and also to telephone subscribers as listed in the telephone directories, so that the sample represented only the population of people who either owned phones or where readers of the magazine.
In 1936, choosing a sample from those who could afford a telephone was in effect selecting people who preferred one party (Republicans).
Fast forward to today and the use of phone polls. This type of poll is relatively cheap and quick however they are suspect because fixed line users are over represented. Many voters from the younger demographic use mobile phones which are very difficult to sample (you can look up my fixed line but even I don’t know my mobile number).
Pollsters can factor in corrections for some of this stuff but it costs more for what maybe considered marginal improvement.
I suspect that the marketing guys have taken control of newspapers polls (they are a good little earner) and statisticians are a little too close to “scientists” to be trusted with important stuff.
Greg Jericho in his brilliant blog has shown how editorial agendas distort how the polls are published and promoted depending on the political message. Pollsters know what the right wing media wants and they are in the business of selling a product with the “correct message”.
So do I think the polls are accurate – no .
Do I trust Newspoll given their News Ltd association. Well send me $100 dollars and I will tell you how to avoid being conned.