NAIROBI (Reuters) - They've been kidnapped by bandits, accused of witchcraft, arrested, robbed, and swept away by flash floods. They've also been mobbed by angry voters and confronted by men offended that their wives are being interviewed instead of them.
The work of political pollsters is still widely misunderstood in Kenya, with international companies like Ipsos accused of everything from corruption to making children sick.
"We have had to airlift people out before," said Hilda Kiritu, head of public relations for Ipsos's Kenya office, as she listed her colleagues' travails.
Kenyans are preparing to elect their next president, lawmakers and local representatives on Aug. 8, with incumbent President Uhuru Kenyatta seeking a second five-year term and veteran opposition leader Raila Odinga determined to unseat him.
As campaign rhetoric heats up, hostility towards pollsters has erupted again. The latest Ipsos poll showing the presidential race may go to a second round sparked angry denunciations and threats on social media.
"There's a lot of people who believe polling is not scientific, that everything has a price tag," Aggrey Oriwo, managing director at Ipsos Kenya, told Reuters. "They don't understand there are a lot of checks and balances."
Kenyan elections are traditionally tight but there has never been a run-off. President Mwai Kibaki won in 2007 by 232,000 votes; in 2013, Kenyatta avoided a run-off by just 8,100 votes.
Both times, irregularities marred the elections and the opposition alleged rigging. In 2007, protests sparked widespread ethnic violence that killed around 1,200 people. In 2013, Odinga took his challenge to court. Protests were mostly peaceful.
A close race and eroding trust in public institutions are two of the most closely-watched indicators of potential violence. Ipsos polls show worrying trends for both.
The latest Ipsos survey of 2,026 Kenyans, released last month, showed 48 percent supported incumbent President Uhuru Kenyatta, seeking a second and final five-year term. Forty-two percent supported veteran opposition leader Raila Odinga. Two percent refused to say and eight percent were undecided.
The sampling error was plus or minus 2.18 percent.
The winner needs one vote more than 50 percent of the national total, with at least a quarter of the vote in 24 of the 47 counties.
Trust in institutions has increased overall, but it's split along partisan lines: 79 percent of Kenyatta supporters trust the electoral board, but only 43 percent of Odinga's supporters.
Less than half of all Kenyans said they would trust the presidential result and only 39 percent said they trust the courts. That may be dip further when the next survey comes out this month. On Sunday, Kenyatta angrily attacked a high court decision he said favoured the opposition.
Voters' faith in pollsters has slipped globally after polls failed to predict a majority of people in Britain would vote to leave the European Union or that Donald Trump would win the presidential election in the United States.
"Who are these aliens you claim to take opinions from?" is one of the more polite comments on the Ipsos Kenya Facebook page.
Ipsos made a Swahili-speaking American, Tom Wolf, its public face in Kenya to dampen accusations of ethnic bias. The price of fame means that he has twice had to flee angry mobs and is often surrounded by voters demanding their views be heard.
Politicians have called for his deportation, threatened to sue Ipsos over defamation and complained to its Paris headquarters, saying their work is biased.
It's not just Kenyans who have a problem with pollsters, Wolf said. After Trump denounced polls during his presidential campaign, some of his supporters stopped responding to them, possibly helping skew results in the United States, he said.
"America is becoming more like Kenya every day," he said with a wry smile.
Ipsos tries to counter doubters by explaining how its methodology is designed to ensure representative sampling.
For its quarterly survey, computers send pollsters to 200 locations around the country, randomly chosen to reflect population density. They visit urban slums and remote deserts bordering Somalia.
A computer program ensures randomness in the selection of households by telling pollsters to pass a certain number of homes before stopping. After the pollster lists everyone inside over the age of 18, the person to be surveyed is chosen using a kish grid, a table widely used in international research.
A smartphone records the interview, answers, time and exact location. For questions on presidential candidates, interviewees enter answers on the phone so the interviewer can't see them.
"We are trying to remove the element of human error," Oriwo said, and added the most important thing is to accurately reflect voters' concerns, so their leaders can listen.
Editing by Sonya Hepinstall