How ERA report distorts statistics

[Published in the Cambodia Herald on 29 January 2014.]

There is an old proverb in English, “Figures don’t lie”. Perhaps because of the increasing use of statistics by politicians and their agents, the proverb was long ago modified: “Figures don’t lie, but liars can figure”. A later version of this saying is “How to lie with statistics”.

For some reason, these modified proverbs kept coming to mind as I read the report of the Electoral Reform Association released in early December. Among its many accusations against the conduct of the 28 July general election, the ERA report includes what it considers some very suspicious numbers emerging from voter registration and the voting.

The report’s authors claim to discern “unusual patterns” in the additions and deletions to the voter list in the 2012 annual updating of the list. They don’t mention the fact that the appropriate and legally specified time to raise objections to anything that seemed illegitimate about the voter list was then, when it was being updated. The (at that time two) opposition parties boycotted the whole process, and the organisations that make up the ERA seem to have followed them. Now they want to convince us that there was something wrong in the procedure that they refused to observe a year earlier – based on figures.

Not so ‛unusual’

The “unusual patterns” of the ERA are the fact that, in the updating, the number of voters in some polling stations increased or decreased more than at other stations. A moment’s thought makes it obvious that this is not really unusual. It would be unusual, bordering on the impossible, for all polling station voter numbers to increase by the same percentage. Social science developed averages and other statistical tools because social processes involving more than about three people nearly always vary.

The ERA presents a chart that “shows the number of polling stations in each province that increased in size by at least 50%” in the 2012 update. Such an increase, it asserts, implies that “either half the population in the community turned 18 recently or there was large migration to these areas”. The report’s authors have a great ability to imagine misdeeds by the electoral authorities, but when it comes to looking for simple explanations, their imaginations fail them. Instead of either large numbers turning 18 or large migration, isn’t it possible that both phenomena happened in the same commune? Anything else? It seems that NGOs that are continually complaining that eligible citizens aren’t registered are unable to imagine unregistered citizens deciding to register.

From the report’s chart (Figure 21), it appears that there were 1274 such polling stations. That is 7.0% of the 18,109 stations that existed at the time of the 2012 commune elections. That is not a large number for Cambodia, which has a highly mobile population.

What they didn’t notice

The figures in the ERA report also point out something else, which seems not to have been noticed by its authors. The law sets a maximum of 700 voters per polling station. As a result of the 2012 voter list update causing some stations to exceed the maximum, 691 polling stations were “split” into two, and 209 completely new stations were created. The report singles out Kampong Cham, Kandal and Takeo as having especially high numbers of stations in which the number of voters increased by 50% or more. For instance, there were 188 such stations in Kampong Cham. But the number of split and totally new stations in that province, according to the report’s Figure 23, was only 97. So at least 91 of the 188 stations with “unusual” increases in the number of voters did not increase sufficiently to exceed the 700 maximum. (There could be more than 91, because one or more polling stations close to the maximum might have gone over the limit with only a small increase in numbers.)

What this means is that at least 91 of those Kampong Cham polling stations with large percentage increases had relatively small numbers of voters prior to the 2012 update. In order not to go over the 700 limit with a 50% increase, they must have had fewer than 467 registered voters. The significance of this is that percentage increases of course appear much larger if you start from a low base. If a polling station increases by 75 voters, that is a 75% increase if it previously had 100 voters but a 15% increase if it previously had 500 voters.

In all three provinces singled out by the report, around half of the polling stations with “unusual” increases were stations in which the percentage seems large because the stations had a relatively small number of registered voters prior to the update (at least 91 of 188 stations in Kampong Cham, at least 66 of 135 in Kandal, at least 69 of 112 in Takeo). For the whole country, there were 1274 stations with a 50% or more increase and only 900 new stations, so in at least 374 stations, and probably many more, the increase appears unusual only because the the report’s authors chose to present the figures as percentages in which the base of the fraction is hidden from the readers.

Turning to polling stations in which there was a large percentage drop in voters registered, the report is even sillier – and not as complicated to dismiss. The ERA managed to find a grand total of – wait for it – 79 polling stations in which half or more of registered voters’ names were deleted in the 2012 update. Of the total 18,109 pre-update polling stations, that represents 0.4%. In an honest statistical treatment, they would probably not even be mentioned except as “other”.

But the ERA has not concluded its statistical war. “New polling stations had unusually high turnout rates”. The authors leave us to develop our own suspicions from the fact that, once again, some stations deviate from the average. Clearly, the increases in registered voter numbers that the report has been complaining about didn’t happen or are supposed to be forgotten about while reading this part of the report, so there was no real need to create the new stations, where all the names entered were those of the Vietnamese troops that the ERA saw marching into Cambodia to vote on 28 July.

It is true that the percentages of registered voters actually voting on 28 July was higher in new stations than in pre-existing ones. This is what one would expect if commune authorities were doing their jobs. All of the voters registered in the new stations would be voting in a different place than where they last voted. Therefore, local officials would be calling this to their attention, pointing out to them where their new polling station was located and maybe even reminding them to find their name on the voter list before election day. They would be much more conscious of voting in the upcoming election. Meanwhile, some of those registered at the old station had moved away or forgotten its location or forgotten the election date. Of course new polling stations had a higher turnout.

Last gasp statistics

The ERA authors must have been commissioned to quote the opposition interpretation of every statistic. For example: “In polling stations with turnout well higher than the country’s median, [the] CPP performed above its national average.” What more proof do Sam Rainsy or Kem Sokha need? Where the voter turnout was higher, it must have been because the CPP dragged military personnel and civil servants to the polling stations and forced them to vote for the CPP.*

What the ERA report authors don’t say is that it is also true that: “In polling stations with turnout well lower than the country’s median, [the] CPP performed above its national average.”

That’s right: both of the seemingly contradictory statements are true. Once again, the ERA authors are illustrating those proverbs. This is because, according to the report’s Figure 30, the CPP’s vote was higher than its national average in eight out of 10 voter turnout deciles. In only two of the deciles – 60-70 and 70-80 percent voter turnout – was the CPP vote below 50%.

As in all things where averages are necessary to understand what is going on, some figures are above the average and some below it; that is obvious to everyone except the authors of the ERA report. And so, in the polling stations with an over 90% turnout, the CPP vote exceeded the CNRP vote by 18%. But in the 30-40% turnout decile, the CPP outvoted the CNRP by 17%. Is the one figure really more significant than the other? Only according to the ERA.

Furthermore, according to the report’s Figure 30, in all deciles, across the board, the CPP’s percentage of the vote was higher than the CNRP’s. How, then, can the distribution of these votes among the different percentages of voter turnout make a case that the CNRP has been illegitimately deprived of anything? I await the ERA’s answer; I cannot even begin to imagine what it might be.

The other arithmetical mystery that arouses great suspicions in the minds of the ERA report’s authors is the fact that the CPP got a higher percentage vote in polling stations where there was a high turnout than in stations where there was a low turnout.

If the ERA electoral experts looked at almost any other country in the world where there are free elections, they would discover that, when a governing party loses some of its supporters, some of those disaffected voters vote for the major opposition parties, but varying numbers of them vote for small parties or don’t vote. Yes, that is a well-known phenomenon: people unhappy with the party they normally vote for, but not trusting the opposition, just don’t vote. (The percentage of registered voters who voted on 28 July was 5 percentage points less than in 2008.) End of mystery.

* At one of the CNRP rallies after the election, Kem Sokha said that 70% of the civil service, police and military had voted for the opposition. Oops!

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s