The problem with numbers is that there is always that margin of error.

It's like a poison manufacturer having a malfunction in its equipment and a batch of poisons getting the 'un' part of the 'unsafe for human consumption' label omitted.

The machine tells you one thing but deep down you know it's not true.

Adults (sane adults) would never take a sip of the liquid unless they were committing suicide.

The numbers showed that Ben Mulwa was more popular than Martha but if you look at the numbers critically what amount of respondents actually said they would vote for Mulwa? What was the exaxt number using the stats?

I was referring to any bias on the part of Smart Octopus and not the respondents.

All smart Octopus had to do was go to a location where the Kamba vote was overwhelming meaning that splits in different regions would not be enough to offset any Kamba bias in the results. That's why I think Ben Mulwa showed up.

To know where this poll was conducted we need to know from where Ben Mulwa comes from.

Here is my view on the numbers - Ben Mulwa's case:

2000 respondents obtained form all 47 counties implies that each county had 42.5531914893617 respondents. (42 hereafter) i.e. the figure was obtained after dividing 2000 by 47 (2000/47=42.5531914893617)

To be fully representative you need to go farther and have all constituencies represented.

Certain counties have more constituencies than others and the number of registered voters also differs sharply.

Cosmopolitan areas have a good mix of voters from different ethnic groups whereas other areas have an overwhelming majority of voters from a specific ethnic group.

If a pollster is not biased, he will do the poll by the book. He will factor in voter registration patterns and voter turnout patterns.

The reason for this is simple. An indication that 80% of voters in county A would vote for candidate 1 does not necessarily mean that they are more than 50% of voters in county B voting for candidate 2.

Regions such as North Eastern and Coast have lower populations than smaller more densely populated areas like Nairobi and Central.

80% in Coast or North Eastern could turn out to be 2 million people whereas 50% in Nairobi or Central could turn out to be 3 million people.

This was the main anomaly in the 2007 elections where President Kibaki obtained high numbers from small regions.

The Pollster factored this into his poll but to a minimalistic extent.

Back to Ben Mulwa.

If the pollster due to cost factors decided to make the poll minimalistic he would have sent a single representative of the company to each of the 47 counties but directed that the poll should be conducted in a single location in these counties.

Assuming the researchers had gone to a town like Nakuru which is metropolitan the outcome would be very balanced since such an urban area would have a good mix of different ethnic groups in Kenya.

The same case would differ if the pollster had sent a researcher to a county center such a Kitui where an overwhelming majority would have voted for a particular candidate because of ethnic affiliations.

This would actually make a difference since 42 respondents all from a single region claiming they would vote for a particular candidate would imply that such a candidate gets a more than 2% score in the poll i.e. 42/2000 x 100 = 2.127659574468085% (2.13%)

Ben Mulwa's 4% could have been influenced by a single center voting for him overwhelmingly.

I suspect the particular poll was conducted in his home county where he might have marketed himself beforehand to garner enough support to bring about this result.

Smart Octopus like that mollusc that "predicted" World Cup football matches??

HILARIOUS! And we're supposed to give such polls credibility??

Foolishness indeed.