Political Polling – Understanding the Numbers

According to a poll conducted by the Vision Group, if the election were held the day of the poll, President Yoweri Museveni would win by 71%, Dr. Kizza Besigye would receive 19% of the vote, and former Prime Minister John Patrick Amama Mbabazi only 6%. Meanwhile, The Monitor Publication Ltd. poll shows Museveni at 60%, Besigye at 21%, and Mbabazi at 6%.  Research World International (RWI) has Museveni’s support at 51%, Besigye at 32%, and Mbabazi at 12%. Each party has also done internal polling and claims their candidate will not only win, but win by a landslide.

With such conflicting results, what should Ugandans believe?

History is full of inaccurate polling. Sometimes a few get it right, but at times almost everyone gets it wrong. Just last year in the United Kingdom, polling companies all predicted a hung parliament split between Labor and Conservatives. The reality: a clear Conservative win.

Often, mistakes in polling are due to voters changing their minds at the last minute, social pressure inhibiting people from answering honestly, or the sample group not being truly representative of the population.

Polling companies like to emphasis the science that goes into polling, but polling is also somewhat of an art. There may be precision in how they choose the sample population to target, calculated margins of error and levels of confidence, and weighted results according to population statistics, but there is also a lot of guesswork.

One question asked in each of the major Ugandan polls and a bit of history tells why. “Will you vote in the upcoming elections?” asked the polls. Of the Vision Group respondents, 97% said they would vote, and RWI was just slightly less at 96%. It’s wonderful to see an electorate so keen to vote, but a look at history tells another story. In the presidential elections of 2001, 70% of registered voters cast their ballots. In 2006, 69% voted, and in 2011, the number dropped to 59%, about 38% fewer than the more than 96% who told pollsters they were going to vote this time. Here is the art performed by the best pollsters: determining which respondents really will vote as opposed to those who say they will, but then stay home.

Most of the major surveys released information on methodology, such as the number of people polled, the margin of error and confidence level for the results, whether it was proportional to the population, and more. What is often more closely guarded is the response rate. The number of people who agree to take part in a poll are going down, and the lower that number goes, the more results become questionable.

Polling aims to show what a larger group thinks by selecting a representative group, but it is essential that every member of that group has an equal chance of being selected. To find out what 1000 people think of a government initiative, a pollster could randomly choose every 50th person to ask, but if many of those people opt out, the final sample of 50 would be made up of people who are more motivated to answer. The final result could be representative of the population or it could be completely skewed.

RWI’s poll offered the most detail, but within that detail, several questions arise. Reporting that 90% of all people had verified they were on the voter register and 96% knew where their polling station was, it appears the electorate is quite engaged. And yet only 43% knew the date of the election, 12% knew the month but not the date, and a whopping 45% didn’t know when Uganda would hold its elections. It is surprising that people would make the effort to verify being listed on a register and location of polling stations, yet not know when the election would take place.

There are questions as to whether traditional polling methods developed in the United States can be transferred to Africa without modification. Political polling in America took off in the 1930’s. At that time there was very little diversity among the population. A person on the East Coast was not all that unlike a person on the West Coast. Even today, with a population of more than 300 million, there are only a handful of major ethnicities. Uganda, with its population of 37 million has 40 ethnic groups, and many with their own language, culture, and customs. According to the Harvard Institute of Economic Research, Uganda is the most ethnically diverse nation in the world.

Why does that matter for polling? Imagine a bowl of matoke. A small taste could be scooped up from any side of the plate with a very high probability that it would taste like any other bite.  Now imagine a plate of beef stew. A bite taken at random might have some delicious beef, or it could be just the greasy fat at the top. To get a proper sample of that stew would take a larger sample.

Using international standards, the entire population of Uganda can be surveyed with 95% certainty and a margin of error of 5 with a sample size as small as 384 people. However, in a country as diverse as Uganda, that small sample, no matter how well it is randomized, will still have a risk of not being truly representative. By those same standards, to poll with the same margin in South Korea, which has a population a bit bigger than Uganda’s, but is one of the least diverse countries on earth, the sample size would be exactly the same, 384.

There is always excitement and anxiety around elections. The candidates and the people would all like to know what the future holds, and pollsters do their best to give as accurate a picture as possible. Like predicting the weather, scientists can only give their best educated guess based on research, history, and patterns, and like the weather, the results from an election can only be certain once the day comes.