Zeeto’s political opinion polls draw nonprobability samples from the millions of internet users who navigate through our network of partners every month. Our survey samples include people of all demographics and from every state across the U.S. With nearly every American now on the internet, our polls are able to accurately capture public opinion on important current events, ranging from hot-button issues to presidential and congressional races.
Zeeto takes two broad steps to ensure survey samples accurately represent targeted populations. The first step is through a set of qualifying questions on the front end of our polls. For example, before inviting internet users to participate in an election poll, we ask them if they are registered to vote. Then we ask about their likeliness to vote. Since the goal of an election poll is to predict an election outcome, only people who are registered and likely to vote are invited to participate in the survey. If the poll is on a primary election, we ask about party registration to ensure only eligible voters are included in the survey.
The second step we take involves the application of weighting adjustments to the data on the back end of our polls. We employ a variation of “proportional weight adjustment” to account for underrepresentation and overrepresentation of population cells in our polls. For each poll, we identify the population cell frequencies for the population in question. Then, each response receives a weight in accordance with the strata it belongs to given by:
where nk/n is the frequency of the kth strata in our poll’s sample and Nk/N is the kth strata’s population frequency. This results in reported answer rates of the form:
where ak is the count of positive responses to a given poll question for the kth strata. Presently, we use strata derived from separating the population by gender and by age.