In this series we have looked at accuracy and precision with a continuing question of how much do we need. We started with some definitions and explored randomness and the value of Pi, the ratio of a circle to its diameter. Then we looked at standardiztion of how we measure physical things, primarily in the scientific-engineering domains but also relating to commerce, and again explored randomness. This was expanded in the last column, introducing the value of money and differences in perception of individuals and groups. Now let us consider how we measure the collective perception and preferences of individuals as insights into group preferences. It’s called polling.
Polling is sampling a small number of people or things in order to draw an inference about the collective body. It requires accuracy and precesion in the process but demands randomness in the sampling. It also assumes, for the most part, that things, positions, preferences and so on are distributed normally. That is, for the item being measured there is an averagevalue for the entire group of items and that statistical mathematics can be used. As an example, in 2010 in the United States the average height of an adult female was 63.8 inches according to the Centers for Disease Control and Prevention. Was this determined by measuring the height of every adult female in the United States? No. A sample was taken and it was assumed that the height measurement of adult females in the United States is normally distributed. Is that a good assumption? Math and science say yes. But was the sample taken to measure the value truly random or was it biased?
In 1936, The Literay Digest conducted a poll of the Roosevelt-Landon presidential race and predicted Landon would win. The publication did not realize its sample, even though very large, was biased in that it targeted individuals who were generally more affluent. Remember that this was in the Great Depression. George Gallup conducted another poll which was more scientifically based in sample selection. The Gallup poll correctly predicted that Roosevelt would win. But the vote, not the poll, determines the outcome. Or does it?
In the recent presidential election, most polls got it wrong because they focused on the popular vote, not the electoral college that really determines the outcome. The polls were focused on the wrong thing. Plus, there may have other biases caused by technology and our higher paced lifestyles.
First, there is a non-response bias. Your telephone or cell phone rings in the fourth quarter and your team is behind by two points. Do you anser the phone? If you do and a recorded vioce tells you that you have been selected to participate in a poll, how many of your explitives are deleted? Today, only about 10 percent of the people called choose to participate.
Your team is ahead by 50 point and the game is getting boring, so you choose to participate. The pollster asks you if you prefer Donald Duck, Bugs Bunney, Elmer Fudd, Mickey Mouse, or The Road Runner for the next cartoon mayor. The order of the listing biases the response. For the position of cartoon mayor, it probably is not important, but it might be if a truly important question is being asked. Of course, then results must be determined by vote. But does the announcement of polling results influence the vote?
Another aspect is that polls cost money. Do you have the pollster walk door to door? Do you set up a table at Neiman Marcus or Walmart? Do you visit the senior center or the local college campus? Or do you call people on the telephone or cell phone? From the money standpoint, calls to phone numbers RANDOMLY selected is probably less biased and more efficient. Or, it used to be.
Does every household have a telephone? No, but if they do, the pollster can call and ask to speak to someone of a certain age or a certain sex. But can the pollster be certain the person they poll is really the right person? Moreover, today, more and more households only use cellphones, which are not commonly shared. Further, is the distribution of calls, commonly controlled by area codes, valid? I know of two people with 505 area codes that live in other states.
Polling depends on randomness in selecting members of the population being polled and a lack of bias in presenting the poll. How do we insure the integrity of the process? How do we insure the process measurement does not affect the population itself? How do we insure the announcement of results does not influence the population or do we measure the effect of influence with another poll?
In this series of columns on accuracy and precision we have explored the measurements things, material, ephemeral, and random. We have repeatedly asked how accurate and precise we need to be. As humanity moves into the future and earth continues to receive energy from the sun, accuracy and precision rises in importance as we make choices. How much do we need?