Evidence Pack: The Accuracy of Address-Based Sampling Versus Random Digit Dialing in Election Polling

Survey methodologists designing national polls must decide how to reach a representative sample of adults, a choice constrained by the collapse of telephone response rates. Their next major opportunity to test these sampling frames at scale comes during the upcoming midterm election cycle. For decades, the mathematical foundation of public opinion research was Random Digit Dialing (RDD). By randomly generating numbers within active telephone exchanges, pollsters could reach nearly every household in the United States. But the proliferation of cell phones and the rise of caller identification destroyed that foundation. Cell phones severed the link between a phone number and a geographic location, making state-level polling prohibitively expensive, while response rates plummeted into the single digits.[1][3]

To restore probability-based sampling, the industry is abandoning the telephone network for the postal system. The new standard is Address-Based Sampling (ABS). Instead of dialing numbers, survey organizations draw random samples from the United States Postal Service (USPS) Computerized Delivery Sequence (CDS) file. This database contains approximately 151 million residential and business addresses, providing near-universal coverage of the American population. By mailing invitations to these addresses, pollsters can push respondents to complete surveys online or via paper questionnaires, bypassing the telecom grid entirely.[1]

The transition has been swift and comprehensive. According to the Pew Research Center, 61% of pollsters who released national surveys in both 2016 and 2022 used different methods in the latter cycle, with a massive shift toward probability-based panels recruited via ABS. Pew itself transitioned its American Trends Panel from telephone recruitment to ABS recruitment in 2018. The shift acknowledges a mathematical reality: dual-frame RDD (mixing landlines and cell phones) is no longer cost-effective for the coverage it provides.[2]

The primary mechanical advantage of ABS is geographic precision. Because every address is tied to a physical location, researchers can append Census data to the sampling frame before a single invitation is mailed. This allows methodologists to geocode addresses down to the block group or tract level. If a pollster needs to oversample a specific congressional district or a geographically clustered demographic, they simply draw more addresses from those specific Census tracts. RDD cannot do this; a cell phone with a New York area code might belong to a resident of Texas.[4]

However, the evidence shows that ABS is not a perfect solution for response rates. In comparative pilot studies evaluating ABS against RDD, researchers found that overall response rates remain challenging across both modes. One study recorded a 7.9% response rate for landline RDD and an 8.9% response rate when ABS addresses were matched to phone numbers. When ABS relies purely on mail invitations pushing respondents to a web URL, the initial response rate can drop as low as 2.1% without aggressive follow-up mailings and financial incentives.[4]

The true value of ABS lies in coverage, not raw response volume. RDD systematically misses "cell-phone-only" households unless expensive dual-frame designs are used, and even then, cell phone samples suffer from a 46.7% rate of unknown eligibility—meaning pollsters cannot tell if a ringing phone belongs to an eligible voter or a disconnected line. ABS reaches these households by default. A physical letter arrives regardless of the resident's telecom setup, ensuring that younger, mobile populations have a known probability of selection.[3][4]

But the data also reveals that ABS introduces its own demographic skews. While RDD historically skewed toward older, wealthier populations who retained landlines, ABS respondents are disproportionately likely to be college graduates and less likely to be non-White compared to the general population. The friction of opening a letter, typing a URL into a browser, and completing a survey screens out different demographics than the friction of answering an unknown phone call.[3]

To correct this, methodologists exploit the auxiliary data attached to the ABS frame. Because the USPS file can be linked to third-party consumer databases and Census records, pollsters can predict which addresses are likely to ignore the survey. Marketing Systems Group, a major provider of ABS frames, notes that "mitigation efforts have made real progress due to Census data appends that can be used to predict areas of high nonresponse." By intentionally oversampling addresses in low-response tracts, pollsters force the raw sample closer to the population baseline before statistical weighting is even applied.

The ABS methodology also facilitates mixed-mode data collection, which the evidence suggests is the most accurate way to survey the modern electorate. A standard ABS protocol begins with a mailed letter containing a $2 or $5 pre-incentive and a link to a web survey. If the household does not respond, a reminder postcard is sent. Finally, a paper questionnaire is mailed to the non-respondents. This sequential mixed-mode design captures tech-savvy voters online while retaining older or lower-income voters via paper.[1]

Despite its advantages, ABS struggles with specific structural limitations. The USPS CDS file contains "drop points"—single delivery locations shared by multiple households, such as a mailroom in a large apartment building. In dense urban areas, a survey invitation addressed to a drop point may never reach the intended respondent. Similarly, rural households that rely on P.O. boxes present a challenge, as the physical residence cannot always be linked to the mailing address, complicating geographic stratification.[1]

The cost structure of ABS also differs fundamentally from RDD. While dialing a phone number is practically free, printing and mailing physical letters with enclosed cash incentives requires significant upfront capital. However, because the web-based completion phase requires no human interviewers, the cost per completed interview in an ABS mail-to-web design is often lower than the cost of paying a call center to dial thousands of dead cell phone numbers.[3][4]

As the polling industry prepares for the next election cycle, the transition from RDD to ABS represents a fundamental shift in how public opinion is quantified. The accuracy of a poll no longer depends on whether a respondent will answer the phone, but on whether the statistical models used to append Census data and weight postal responses can accurately reconstruct the American electorate from the people willing to open their mail.[2][5]