The landscape of political polling is changing, and so too is the way CNN carries out its polls. Beginning with Friday’s newly released CNN poll conducted by SSRS, the network is implementing a new methodology for measuring public opinion.
The new approach relies on both online and telephone interviews, allowing respondents to participate in whichever way is more comfortable for them, and will use different methods for measuring long-term trends and for reactions to breaking news.
The new methodology allows for the researchers behind the CNN poll to have a better understanding of who is and who is not responding to the poll, and to better account for political differences between the two groups, addressing one of the most likely causes of 2020 polling errors.
Election polling in 2020 broadly missed the mark on measuring support for Republican candidates up and down the ballot. Assessments of what went wrong suggest that the problem derived from differential non-response – that is, that the people who took surveys were not the same politically as those who did not.
Under the new methodology, CNN and SSRS will carry out polls several times a year using a larger sample and a longer field period than is typically used for news polling. Respondents across the country will be randomly selected based on their mailing address – a process known in polling as address-based sampling – and will initially be reached by mail. Surveys based on such samples will be used to measure long-established trends, such as presidential approval or views on the national economy, as well as to take in-depth looks at critical issues.
These surveys will also be used to establish benchmark measurements of Americans’ political traits, such as party identification, which will then be used in weighting when CNN conducts other types of polling on a shorter time frame.
The sample for the new poll released Friday was selected from a list of all residential addresses in the United States. Rather than randomly generating a geographically balanced sample of phone numbers, potential respondents were selected randomly based on their address and some known characteristics about the area where they live.
Starting from a sample of addresses means researchers know more about who they are trying to reach – and therefore who might not be responding – than they would with just a phone number. There is more complete voter file information available for addresses than there is for phone numbers, for example, and the geographic precision of an address allows for use of data from the Census Bureau to know more about the type of neighborhood a home is in and the demographics of the people who live in that area.
To invite potential respondents to take the new poll, researchers mailed letters to each selected address asking for an adult in the household to take part in the survey. Those receiving the letters had the choice of taking the survey online or calling a dedicated phone number to take the poll with a live interviewer. For a small subset of those addresses, follow-up phone calls were made toward the end of the field period to recruit some of those who had not yet gone online nor called in to take the survey.
This methodology does not necessarily improve the rate at which people respond to a survey, but it allows for researchers to make better adjustments should differential non-response occur, such as by increasing outreach to underperforming segments of the sample.
CNN’s polling had been conducted almost exclusively by telephone from the beginning of its polling operation in the early 1990s through earlier this year. CNN has incorporated online interviews into its speech reaction polling since 2018 and its first text message poll was released in April 2021.
