Don Anderson | How come I have never been polled?

People often ask me, “how can polls be right if I have never been interviewed?”. This is a reasonable question to ask.

Ideally, the best way to get the opinion of persons within a country is to do a census, meaning interviewing everyone. That way, all views are recorded. Censuses are generally done every 10 years in countries all around the world. There are two major limitations to more frequent conduct of censuses, time and cost. Here we have been trying to complete our latest census for close to three years now and we have still not done so.

The alternative to conducting a census is a sample of the population.

The most important factor in designing a sample is that it must be representative of the population from which it is chosen. That is, it must mirror the population in terms of the various groups within the population, males, females and various age groups, for example.

That way the results can be accurately projected to the overall population.

When a sample of 1000 persons is interviewed here, it means that 2,899,000 persons don’t get interviewed. Therefore hundreds of national surveys could be conducted before one is confronted by an interviewer.

A follow up question normally comes. “So how can you interview 1000 persons and project from that number that this is how the population thinks?”

Well, As long as the sample is truly representative of the population, that is, it is a mirror of the population from which it is chosen you can use that base of 1000 persons to project that this is how the overall population thinks.

What is most critical in designing a sample then is how representative it is and not its size. But there are times when it will be necessary to conduct a survey of 3,000 persons for example. When? If the objective is to do a detail analysis of one of the sub-units of the population, say men in Kingston, then this will necessitate using a large sample so that prominence can be given to this smaller sub-unit that one needs to analyse. One would need then a large enough number of men who live in Kingston to do any meaningful statistical analysis. This will in turn impact the ultimate size of the sample.

HOW SAMPLING WORKS

My company Market Research Services Limited, now in its 50th year, normally uses a sample ranging from 1,000 to 1,300 persons aged 18 years and over for national political polls. This sample is generally nationally representative of the population, being a mirror of the population from which it is chosen.

The easiest way to explain sampling is by using the example of a bowl of vegetable soup. Imagine you have prepared a nice bowl of vegetable soup, including a bit of pumpkin, cho cho, carrot, etc, nicely cut up.

As you about to drink it, an important telephone call comes in for you. You return 15 minutes later and take up a spoonful. This spoonful will not give you an accurate account of the taste of this soup, until it is properly stirred, as some ingredients would have settled to the bottom.

If however you properly stir this soup then take a spoonful, you are guaranteed to know exactly how the soup tastes. The ingredients in that spoonful will be represented in line with their numerical proportions in the bowl. From that spoonful then, properly stirred, you will get an accurate assessment as to how the rest of the soup tastes. The more of this soup you drink, properly stirred, the better feel you have of the soup, but there is a point at which each additional spoonful merely serves to fill your stomach, not give you more confidence in how the soup tastes. So with a few spoonsful, you can accurately report on the taste of the soup. Sampling theory 123. The ingredients in the soup represent the mix of people in the population,

In the US, very few polls done on the presidential elections use a sample size larger than 1,500 persons. The population of the US is over 350 million, not three million.

MARGIN OF ERROR/NO PERFECT WORLD

The margin of error is an important statistic in reporting on surveys and polls. Having designed a sample efficiently, one cannot avoid other challenges in the execution of the survey work such as non-sampling errors and other human errors. Detailed statistical calculations are done to measure the extent to which the data is truly accurate and the extent to which it varies from actual. We have statistically established that a sample of 1,000-1,300 persons designed in the most efficient way, will result in data which we are 95 per cent sure is within plus or minus three per cent points of actual. This is the concept of the margin of error.

Major Private sector and corporate bodies have been benefiting from our high level of professionalism for the past 50 years.

Don Anderson is the executive chairman of Market Research Services Limited. Send feedback to columns@gleanerjm.com