QUICK
COUNT Q & A
The Quick Count in Indonesia's April 5 elections showed
accurately who won. The Quick Count in the presidential
elections will do the same.
What is a Quick Count?
A Quick Count is NOT A SURVEY. It is NOT AN EXIT POLL.
Rather, we count around half a million votes from all
provinces in the country. Instead of counting every single
vote, which takes weeks, we count a representative sample of
all the votes actually cast in Indonesia. We get our votes
from 2,500 carefully selected polling stations that form a
representative sample. Because they form a representative
sample, we can use them to accurately predict the result of
the election.
What is a representative sample?
A representative sample is a statistically significant
sample of something what you are trying to measure, whether
it's a country's level of unemployment or, in this case,
votes. Representative samples from the basis of the science
of statistics. They are relied on by Governments and
corporations every day to measure things. When Governments
measure things like unemployment, they don't find every
person who is unemployed before they announce an
unemployment rate. Rather they get take samples of
unemployment in all areas of a country and from those they
calculate the actual level of unemployment in the country.
To be representative, the sample must be designed according
to statistical principles - that means using appropriate
statistical methodology to measure voters behavior.
How will the Quick Count works?
LP3ES and NDI will send 5,000 people to sit in 2,500
separate polling stations spread across all 32 provinces,
which is why it's expensive. We have selected our stations
according to so called multistage cluster sampling design.
When KPU officials count the votes after booths close at
1pm, our volunteers watch and write down the results. That
way we will know on Monday how some 500,000 people voted
across the county.
What happens to this information?
Our teams will phone in their results to our head office in
Jakarta. We process and analyse these data, calculating
various statistical indicators (such as convergence,
variance and deviation, sampling error) that are telling us
about accuracy of our sample. We can then see the percentage
of votes each candidate has received in Indonesia, but also
in separate provinces.
What then?
After checking the figures, we will release the results
publicly on the press conference.
How accurate will the results be?
A Quick Count is designed to be accurate to plus or
minus one per cent with a 95 per cent probability. That
means it is 95 per cent certain we will be accurate to
within one per cent of the final figure. Often our figures
are a lot more accurate than that. Please see the list of
our past results to understand how accurate Quick Counts
really are.
Why should journalists report this?
We believe the public has the right to know the results
of the election. The political uncertainty created after the
Election Day can lead to economic uncertainty. By knowing
the results of the election earlier, it can give more
confidence to the economy and foreign investment. We also
believe it's the job of journalists to provide important and
accurate information as soon as it becomes available. The
Quick Count is not perfect, but it is extremely reliable and
very accurate. Remember, no system of voting and counting
votes is perfect. Some votes are always lost, weather and
other factors can disrupt the polls in some places. A Quick
Count is the best way of getting the picture of what
happened across the country. Journalists report early
results published by the KPU even though these results give
a distorted picture of the eventual result because they are
not a representative sample of the votes cast.
What's the difference between this and other surveys?
This is not a survey. Public Surveys rely on people's
opinions of how they would vote which may or may not be
given honestly and they might never actually cast a vote.
Quick Counts rely on facts taken from the official polling
station vote count. The Quick Count does not measure
preference; it measures voter behaviour. There is no
argument about what people really wanted to do.
How do we get a representative sample?
Sample that is used for projection of election results
follows statistical principles. The sampling base on which
we are designing a sample is a called sampling population.
Sampling population for the Quick Count is registered voters
in Indonesia. However, statisticians observe populations
through sampling frames - meaning a list of our potential
sample points. The Quick Count cannot reach individual
voters because we observe results at the polling station
level. That means that the sampling frame consists of
polling stations in Indonesia. Polling stations are our
sample points, the primary unit of analyses. In statistical
terminology, this kind of sampling unit is called a
"cluster." Voters are called "elements," and are the
ultimate units of analyses. To determine which polling
stations will be used as our sample points, we use
multi-stage processes. The first stage is the stratification
of the sampling points. In our Quick Count, the strata will
be the polling stations in a single province, as we also
design the sample to project provincial results. The next
stage is selecting a subset of polling stations that will be
processed for the projection of the national popular vote.
Both within the first stratification and among the subset
points, sample points are distributed proportionally down to
the district level (Kabupaten). The last stage is selection
of villages. A village can be selected in several ways,
based on data available: simple random, systematic or with
probability proportional to size (which means that a village
with a large number of voters has more probability to be
selected). The described sample is a so-called
"probabilistic" sample design.
How the KPU electronic vote count performed compared with
LP3ES-NDI Quick Count?
The solid lines in the graph represent LP3ES-NDI Quick Count
results for the major parties as announced the day after
elections, 6 April 2004. The doted lines represent our daily
tracking of the KPU electronic vote count and its
convergence of the official results with the Quick Count
over the course of three weeks.
How the KPU manual vote count performed with LP3ES-NDI
Quick Count?
The final KPU manual count was announced one month after the
election. This table compares the 6 April 2004 Quick Count
results against the final KPU results. KPU results are shown
in solid bars, the Quick Count results are shown in
chequered bars. |
|
|
