Surveys are the lifeblood of modern electoral campaign coverage, and nearly every major media organisation is publishing them this year. But with so many different numbers bouncing around, making sense of them becomes hard. Each survey is also conducted using slightly different techniques, resulting in systemic biases from the true result. But aggregating all these data points together makes the situation easier to understand and potentially increases accuracy.
To build our aggregate, we’ve collected 18 telephone surveys from MaltaToday, The Independent, Xarabank and Times of Malta spanning a period of one and a half years; from January 2016 until now.
Each dot on the graph above is a data point from one of those surveys. The curved line making its way through the points is a LOESS curve; the best way to think of it would be a modified version of those lines of best fit from your Form 5 days, but which can deal with data that might not be accurately represented by a straight line.
LOESS curves smooth noisy or variable data and attempt to plot a single or more representative line – you can see this in practice most clearly to the right of the graph, with the lines fitting through a cluster of dots representing all the surveys over the past month.
The two fitted LOESS lines currently give PL 53.1% of the vote and Forza Nazzjonali 46.3% of the vote. (Minor note, PN’s numbers were used up until the declaration of the single ballot, after which PD’s numbers included with PN’s).
Now that only means that when looking at all available polls, PL leads PN by 6.8%, it doesn’t necessarily mean that PL is set to win the election by that margin, for a variety of reasons.
Firstly, each one of those dots that specify a single poll has an associated margin of error, typically in the 3% to 5% range. This margin is an attempt to account for the inevitable degree of error that is encountered when a sample of 500 people or so is used to make guesses on an entire population.
Secondly, other sources of error, which are harder to estimate can also exist; like for instance members of one party being reluctant to participate in surveys. This appears to be true for at least one datapoint of ours; the Times survey released on Sunday had a larger share than usual of “I don’t knows”, and a disproportionately higher number of people saying they voted Labour in the previous election. If a sufficiently balanced snapshot of the electorate was obtained, it would be reasonable to assume that reported voting in 2013 matched the actual election result. But this didn’t seem to happen despite the fact that the survey was excellently crafted, with 1,826 randomly selected individuals being asked to participate, and a reported response rate of 43%.
How accurate have Maltese polls been historically?
Probably the best online repository of election polls is MaltaToday, whose James Debono has been in the polling business for close to a decade. MaltaToday accurately predicted both 2008 and 2013 elections, both of which were polled to within 1.5% accuracy.
MaltaToday’s 2008 election polling; Labour started with a lead, before PN managed to turn the tables in that campaign.
MISCO, which conducted this year’s Times of Malta poll also has a great track record, and managed to accurately predict the results for all parties within less than 1% for the 2013 elections.
Interestingly however, when looking at MaltaToday’s 2013 result, it does appear that PN support was also underrepresented there when extrapolating the numbers and getting rid of voters who didn’t know how they’d vote or didn’t reveal their intentions.
To illustrate this, here are MaltaToday’s numbers for the 2013 elections.
Malta Today’s polls for the 2013 election. While PL and AD support was overestimated, PN support was underestimated by around 3%.
While both PL and AD were over represented in MaltaToday’s 2013 surveys, PN voters were underrepresented by around 3%. It’s important to note that this is well within the margin of error –the point being made here is that there might be a historical precedent for PN voters to be under represented in a way that’s similar to the UK’s Shy Tory factor.
This, coupled with participants who don’t know how they will vote yet – a range spanning from 21% to 33% – depending whose numbers you quote, may ultimately sway the election in PN’s favour.
How are the small parties doing?
Both are likely to capture around 3% of the vote, with AD currently polling at 1% and PD at 1.7%. AD peaked in popularity over the summer of 2016, with around 3% of voters expressing support, before dipping to hover around 1%. While PD’s performance has been more variable, it does seem to be the better doing of the small parties.
All of the above interpolations discarded any replies that did not reveal voting intentions. This means we essentially threw them out and assumed a perfectly even and uniform split. But that segment is currently worth 21% according to MaltaToday and 33% according to Times of Malta. Just to get an idea of how influential it could be, let’s take MaltaToday’s numbers published on Sunday 28th, which gives PL 41.1%, FN 36.9% and AD 0.7% of the vote respectively. This totals out to 78.7%, or around 21.3% of the vote still to be determined.
Now let’s assume a 2013-esque turnout of 93%. 2.2% of voters surveyed have already said they plan not to vote, so we’d have to subtract the resultant 4.8% (7-2.2) from our 21.3%, leaving 16.5% of the vote. If all of this entire 16.5% goes to PL, it would win with a landslide 57.6% of the vote compared with FN’s meagre 36.9%. If the entire 16.5% goes to PN, it too would win with a landslide 53.4% compared to PL’s 41.1%.
But, more crucially, for PN to win by a wafer thin 0.2% margin, it would need at least 10.45% of this vote. In other words, if nearly two out of three undecideds vote FN, FN would just win with 47.35% of the vote to PL’s 47.15%.
Change in percentage of undecided voters across weeks to election in 2013 and 2017. Both these numbers are courtesy of MaltaToday.