Poll holes - Village Magazine

10 February 2016

2016 ELECTION

ou may not be sick of the election yet,

but give it time. One of the more

depressing aspects of elections is

how little coverage is given to dis

cussions of the merits of different

policy options or the plausibility of the teams

on offer, and how much is given to treating the

election as a horse race. The 2011 election was

somewhat different, but that’s because Ireland

was in peculiar straits and people suddenly

cared about Ireland’s bond yields and such

stuff. Now that people aren’t so concerned the

media is also reverting to type.

The big feature of election-as-horse race is

the opinion poll. They are useful for voters who

can use them to consider which potential coali-

tions are likely. They are useful for parties

which can better understand the concerns of

voters and failings in their campaigns. They are

especially useful for newspapers which can

make an opinion poll ﬁll a few pages, even when

nothing at all has happened.

Though there are concerns as to the accuracy

of polls in recent elections across Europe, they

remain the best barometer of public opinion we

are likely to have. But because papers are in the

business of making the ephemeral seem note-

worthy there is a temptation to read too much

into polls. To avoid falling into this trap here are

some pointers to keep in mind:

Polls draw samples, and the size of the

sample matters. Samples only work because

of the probability that they are representa-

tive of the population (which is what we

really care about) though comprising a rela-

tively small number of respondents. A poll

with 1,500 respondents will typically be a

more accurate representation than one with

900. With a sample of about 1,000 we are

pretty certain that the parties’ real support

is in a range around what is reported. This is

often about +/-3%. Movements within this

Margin of Error (MoE) really might be

random, not based on any movement in sup-

port for the parties in the population.

So: look at the sample size. Many constitu-

ency polls have small samples, which

increases the margin of error to rates that

make the poll wholly questionable.

Sometimes we see interesting results and

the media gets very excited. One poll in the

2014 Scottish referendum showed the Yes

side in the lead, though all other polls

showed a narrow but consistent lead for the

No side. There was a frenzy. The likelihood

was that this was a ‘rogue poll’, which isn’t

to say it was a dishonest poll. When we are

pretty certain that a party’s support is in a

range (say +/-3%), we mean we’re conﬁdent

that it is in this range 95% of the time. About

one in 20 polls will be wrong. Unfortunately

we don’t know which ones they are. But if a

poll is very interesting, then it’s likely to be

wrong. To avoid getting excited about poten

tially rogue polls, look at the trends. If a

number of polls show a party going up in

support, and it’s sustained over time, then

we can conclude they’re capturing real

movement.

Just because many opinion polls agree about

a level of support doesn’t mean that they are

right. Polls in the UK election had the Tories

and Labour neck and neck for much of the

campaign. The MoE assumes perfect random

sampling. In practice opinion polls rarely use

pure random sampling. Different companies

identify and approach respondents in differ-

ent ways, but all have some selection bias.

That is certain groups are more likely to

respond than others. This would be ﬁne if

these groups all had similar opinions. But

they don’t. Older, middle-class men are

harder to get to respond than younger, politi-

cally active people. This is what caused the

failure of polls in the UK to detect the Tory

lead. Tory voters weren’t ‘shy’ - embarrassed

to admit they voted Tory; they were just less

likely to say Yes when asked to respond to a

survey. Unfortunately there’s not a lot the lay

person can do about this, except retain

scepticism.

Even when we get accurate poll numbers Ireland

offers other challenges. The nature of the elec-

toral system means that converting percentage

support into seats is difﬁcult. It is even harder

now because the fragmentation of the party

system means looking at past trends isn’t all

that useful. Also, partly because of the gender

quota legislation, many parties have more can-

didates than they would like. This splitting of

the vote means that while we can be reasonably

conﬁdent of the ﬁrst two or three seats in most

constituencies, after that the large number of

competitive candidates makes predicting the

ﬁnal seats little more than a coin toss.

Predictions

Dr Eoin O’Malley is senior lecturer in the

School of Law and Government, Dublin City

University.

If we assume polls in Ireland have the same

problem as those in the UK and under-

sample people who are "content", and if we

assume that larger parties get a seat

bonus, and Sinn Féin struggles to attract

transfers, then on the basis of recent poll-

ing numbers, here are my predicted seat

ranges for the parties.

Fine Gael 59 ±5 seats

Fianna Fáil 34 ±4 seats

Sinn Féin 25 ±4 seats

Labour 14 ±3 seats

We can be even less certain for smaller par-

ties, especially for the Greens and Renua,

who may return no TDs.

Green 1 seat, Renua 2 seats,

Social Democrats 3 seats, AAA/PBP 5

seats, Independents 15 seats

Poll holes

There are systemic

risks of errors

by Eoin O’Malley