With the American elections entering their final stage, we become witnesses to somewhat contradictory opinion polls regarding the final winner. One day we hear that Clinton won the debate according to voters and the next we see another poll stating that Trump was the clear winner of the same debate. So what is going on? Are research companies so bad at actually doing their job or is something wrong with the data collection?
It all comes down the latter. We say in our sex life that it is not the size that matters but how you use it… well it is more or less the same for a research’s sample. True the larger the sample size the smallest the margin of error but what is most important is how close the sample’s demographics or special characteristics are to the actual population in question. For example we can ask 4000 women ages 18-34 who they will vote in the US election and 1000 voters whose demographics match that of the US population. The first sample size is 4 times bigger than the second one, which can lead to almost 1% less margin of error. But most likely the second sample will give us an answer closer to the final outcome than the first since it will reflect the actual population in question.
So research companies that tend to collect their own data and keep tight control on every quota, participant and data entry, have a better chance of predicting results than companies who simply rely on users sharing a questionnaire with each other to get answers. Imagine the scenario of FB users sharing a link to a questionnaire regarding the American elections. Trump supporters will share it with some of their friends who also support Trump so as to increase the chances of him actually winning. The same goes for Clinton supporters.
People will ask others who share the same beliefs, to participate in researches just to “get their numbers up” and have more to join their cause. Unfortunately this manages to mess the data collected and give biased results. And this is the reason why we get so many different outcomes especially from online questionnaire which have no control as to who really participates.