Greek Referendum Polls: What Happened?
This past Sunday, Greek voters overwhelmingly rejected the EC/ECB/IMF debt agreement, with 61.3% responding “no” to the referendum. Media outlets across the US and Greece reported that the final decision would be ‘too close to call,’ with polls indicating that the ‘yes’ tally had increased over the past week. Obviously, the polls were wrong; so what happened?
Background: 2015 Greek Legislative Election & Party Policies
In January of this year, voters across Greece elected the left-wing Syriza party (originally a coalition of leftist parties) to government with 36% of the popular vote. The previous New Democracy government garnered only 28% of the popular vote, with smaller parties making up the difference.
On June 27, Prime Minister Tsipras called for a referendum to determine public levels of support for accepting bailout conditions set by the EC/ECB/IMF. While Tsipras campaigned for “no” votes, ND leader Samaras urged voters to accept the austerity conditions.
While there wasn’t much time for polls to be completed, the press managed to report several surveys of the Greek electorate. In the final days before the referendum, most outlets were reporting that the vote was ‘too close to call,’ with roughly a 42%-42% split.
Interestingly, we’ve had a very hard time tracking down the actual cross tabulations and methodology from any of the Greek polls, most of which were completed by Greek firms on behalf of larger news organizations. This might have something to do with the fact that no one on staff here speaks Greek, but I would imagine that Bloomberg would have the resources to secure a translator for reporting purposes.
At any rate, we were able to find one such poll conducted by Ipsos with a methodological summary and topline data (click here to view the PDF). We can see that the poll has the following attributes:
- Landline and mobile collection
- Conducted June 30th-July 3rd
- 1001 total respondents
- Weighted according to 2015 voter turnout
- Demos: age, gender, region (presumably the only variables weighted)
- MOE of 3.1%
Now, none of the above specs are outright problematic, apart from one glaring omission: party ID. As mentioned, the parties were split between “yes” and “no” camps, and while it isn’t universal that voters cast their ballot along party lines, ideology should have been considered.
This omission is all the more jarring when we look at the maps.
Here’s a map of the January 2015 election results vs. the July 2015 referendum:
The similarity of the two maps is striking. This suggests that, contrary to the polls, the Greek electorate voted along party lines, plain and simple. It isn’t often that the simplest explanation is necessarily the best, but in this case, it fits.
What did we learn?
Simple: collect and weight for party ID.
We’ve seen this omission to some degree in the United Sates, but most pollsters realize the importance and analytical power of political affiliation. Absent of any polls, if we simply take the proportion of Syriza vs ND voters in the January election, we would predict a win for the “no” camp of rougly 57%. That’s still beyond the [usual] margin of error, but it is at least a correct prediction of the coin-flip.
We can only surmise that other polls made this same error, and perhaps there is some legitimate reason that prevented firms from collecting party identification. One certainty is this: absent of party ID, we would have been better off predicting the results without polling data. If you can’t collect this data, don’t bother publishing the poll — it makes the industry look even worse.