Student Article High School Science Journal Author: Timofey Titov

On the Reliability of Internet Polls

July 12, 2006

Everyone has at least once participated in an internet poll. It is an easy way for a specific group of people on a forum or an internet portal to see the general opinion of members or visitors in that community. In this article the reliability of such polls will be examined from the statistical point of view.

90% of polls are using what is called “convenience sampling” in statistics which is based on “voluntary response.” This is considered one of the least reliable ways of collecting data compared to SRS and census sampling methods. SRS stands for Simple Random Sample. In this method inferences are made from results of a random sample from a population. A census covers almost all of the population. The reason why data collected through convenience sampling is considered to be poor and unreliable, is because it results in bias. Strongly opinionated people are more likely to respond to a voluntary survey. Such bias makes it impossible to make any reasonable inferences about the population.

Poll results could be much more truthful, if the SRS sampling method were used. This could be done by letting only some people to fill out a survey. However, first, a distinction between “permanent” members and visitors must be made. This could be made by simply saying that registered users are members of the community, while visitors are the rest of the population. Another way of doing this is considering only registered users with a significant amount of posts as members of the community. A census should be also considered in internet polls. In real life applications a census is not found, because of its cost. Since most internet polls are not very big compared to populations of countries, a census would not require huge computer resources.

The following restrictions in the alternative sampling methods still apply: undercoverage and nonresponse. However, both can be systematically removed by reminding the subject to take the survey. For example registered members could receive private messages with reminders in their accounts. In the case of visitors, their IP or ISP address could be saved and used as an identification.

The design of internet surveys could greatly benefit from alternative methods of sampling. Since the data would be processed electronically, there will be no limiting resources in processing data. In return to the cost of development of an additional server system the owner of an internet portal would get much more reliable results from surveys.

 
Back to the Apple Blossom