Sunday’s elections were an Epic Fail for Greek polling companies. They failed to predict the clear victory of left-wing SYRIZA. Still during the last week, their prediction was up to a neck and neck race between left-wing SYRIZA and right-wing New Democracy. Some of them, even “saw” a slight lead for the conservatives.
Exit Polls right after the closure of the ballot boxes on Sunday, finally managed to ‘see’ SYRIZA as elections winner. Exit Polls had a margin error +/- 2%. However, elections result brought SYRIZA to lead with an impressive difference of 7.36%.
Public opinion polls:
SYRIZA 25.3% -31.6%
ND 25% – 31%
Polls saw also Independent Greeks – Tsipras’ declared and favorite coalition partner – not in Parliament with around 2.5%-2.7% , but Lafazanis Popular Unity in the Parliament.
Indep Greeks 3.69%
Public opinion polls and Exit Polls had failed. Epic. And thus for a second time in just a couple of months, if one reckons the Referendum in July.
Several Referendum polls where showing a ‘derby’ and a ‘neck and neck’ still on Friday before the Referendum on July 4th 2015
NO 40% -43%
YES 42.5% – 47%
Referendum results: NO 61.31% – YES 38.69%
What has happened? How could they be so wrong?
Sunday night, criticism against the pollsters was immense, especially as their predictions appeared to have misleading and manipulating the public – see: voters – but also political parties -see: independent Greeks and Popular Unity -, thus influencing possible Day After coalition government cooperation.
Sunday night, it was only Dimitris Mavros from MRB who made a step forward and apologized for the failure to approach the result.
In a statement, Dimitris Mavros said “I am personally sorry” and stressed “the responsibility is ours. He added “we don’t have the methodological tools […] the problem is methodological, the rates of undecided voters and abstention are not been read.”
Other pollsters whispered live on camera as they were still on television network panels something like “the tools! the tools!” that “there was no intention to mislead the public” as well as “we will talk about it later.”
Pollsters were proven blatantly wrong by the Independent Greeks that managed 3.69% and elect 10 lawmakers. In his well-known ‘waterfall’-style party leader Panos Kammenos attacked pollsters as “paid killers” claiming that they were paid by the “pimps” of the old regime.
The Popular Unity did not make it to the Parliament.
How did the polls influence voters?
A friend who was pending between SYRIZA & Popular Unity decided to cast vote for PU “in order to send a message of disappointment to Tsipras.” She was sure the PU would enter the parliament and enhance the opposition.
Another friend, who was also pending between SYR & PU cast vote for SYRIZA “in order to hinder a ND-led government.”
A neighbor cast vote for PASOK considering “it would join the government coalition.”
I assume, other voters too – mostly ‘undecided’ – cast their vote according to similar considerations.
BTW: I wonder how the UK dealt with the total failure of public opinion polls in the last elections.
It’s surely not about tools. One may search “sample size and error margin” and have a look on several on-line calculators.
Lower margin of error requires a larger sample size.
Higher confidence level requires a larger sample size.
Relative equilibrium of competing options requires larger sample size.
Accurate polls are time-consuming and expensive.
5% error margin and 95% confidence level requires 355 valid answers (in a population greater than 20000)
Just reducing the error margin to 2% and keeping unchanged the confidence level would require 2144 answers.
The first rule in predicting is to predict when your model/poll is useless. Higher margin of errors are also generated by poor statistical techniques (in this case ignoring or failing to test for structural changes ie fundamental changes in voting patterns). The statisticians producing such polls probably (hopefully) realize this. But polls are “manufactured products” that have to be sold on the market – they need to be swallowed by someone.
The rain dance theory of forecasting
The rain dance has something for everyone. The dancer gets paid.
The clients get to watch a good dance. The decision maker gets to shift the problem onto someone else in a socially acceptable way. (Who can blame him? He hired the best dancer in the business.) —. from pp 386-9 A Guide To Econometrics – Kennedy
And we so should always ask who is paying for the dance.