The object of market research is to find information that generally refers to people’s opinions about something or someone. It is almost always impossible to poll (ask) everyone who might matter, say every possible voter. In this example, all possible voters would understand the “universe”. Instead, market researchers examine a part of the universe. That something is called the “sample.”

REMEMBER: The sample is the key ingredient in the statistical recipe or process. No matter how advanced the analysis is, without a sample that faithfully reflects the universe you have GIGO.

Perhaps nothing better illustrates this point than the famous literary compendium poll. literary compendium was a successful American magazine. In 1936, he published a poll predicting that Republican presidential candidate Alf Landon would easily defeat incumbent Franklin Delano Roosevelt’s (FDR) bid for re-election as president. Tea literary compendium used the appropriate analysis. But he did not consider that by simply polling (sampling) his readers more people in readily available lists, such as telephone directories, he was not sampling typical American voters.

The reason is that during the depths of the Great Depression, only the wealthiest Americans, who tended to be Republicans, could afford a magazine subscription and a telephone. Simultaneously and independently of literary summary, a generally unknown statistician was sampling a considerably smaller number of truly representative American voters (5,000 versus 2,000,000). He accurately predicted FDR’s landslide victory. Shortly after the elections, literary compendium closed while George Gallup became perhaps the best known of all pollsters.

The Literary Digest sample was what is called “biased.” Gallup’s sample was what is called “random.” The laws of probability dictate that a random sample will better represent the universe than any other type of sample. However, to be random, each member of the universe must have an equal chance (equal probability) of being sampled.

Now you can see where literary compendium had a big problem and the result was GIGO.

The first step in this random sampling process is to define the universe. For example, in politics it is the universe of all those eligible to vote, registered voters, or some other group of people. Much depends on the purpose of the survey. Another point is that many polls pretend to use a random sample or to make readers think that the sample was random, when in fact it was not. That’s one of the tricky things about interpreting survey results, and it often takes some expert questioning to figure out.

Similarly, if you are trying to determine the market potential for a new product, who is likely, not who you want, in your universe? Those are the people you want to randomly sample.

The only improvement on random sampling is when, with high certainty, the universe can be divided into different “strata”. Random sampling from each stratum produces a stratified random sample. But remember that knowing the strata is critical for stratified random samples to be successful predictors.

Sometimes you don’t know what your universe is. One way to try to determine that universe is with focus groups randomly selected from the general population. A future article will discuss focus groups.