An owner list goes stale at a rate set by its own sources, its market and the dates its fields were last confirmed, and a dated sample measures that rate in a few weeks. The Census Bureau's American Community Survey puts the share of people in the United States who lived in a different home a year earlier at 11.8 percent for 2024, down from 12.1 percent in 2023 (Census Bureau). A list of landowners or small-company owners moves at its own rate.
Mail that could not be delivered
In July 2014 the Postal Service's Office of Inspector General reported that of more than 158 billion mailpieces handled in fiscal year 2013, almost 6.8 billion, or 4.3 percent, could not be delivered to the address on them (USPS Office of Inspector General, MS-AR-14-006). The report named people moving, and incorrect or illegible addresses, as key causes, and put the cost of forwarding, returning or discarding that mail at $1.5 billion for the year. The Postal Service's Move Update standard at the time tested whether a mailing's addresses had been updated within the previous 95 days. A mailing in which fewer than 70 percent passed paid a 7-cent surcharge on each piece in the shortfall. The inspector general found that only 3 percent of mail was checked against the standard.
What the public record shows
Public sources show what drives decay, field by field.
| What changes | What the public record shows |
|---|---|
| People move | 11.8 percent of people in the United States lived in a different home in 2024 than a year earlier (American Community Survey) |
| Phone numbers change hands | In December 2018 the Federal Communications Commission estimated that about 35 million numbers were disconnected and made available for reassignment each year (FCC 18-177) |
| A disconnected number can go to someone new within weeks | The same order set a minimum aging period of 45 days before any disconnected number is reassigned |
| Movers leave forwarding records | The Postal Service's NCOALink data holds about 160 million permanent change-of-address records (NCOALink) |
| Businesses open and close | The Census Bureau's Business Dynamics Statistics count establishment births and deaths each year, with tables running from 1978 to 2023 (Business Dynamics Statistics) |
None of these rows is a list's decay rate.
A date on every field
Each phone number and mailing address carries its origin and two dates:
- where it came from (a county record, a skip trace provider, the owner on a call);
- the date it arrived;
- the date it was last confirmed, and how: a person reached the owner at that number, a letter to that address was not returned, or a verification check passed.
A field with no confirmation date goes into its own group, labeled never confirmed. Call logs and returned mail can backfill some of those dates.
Age groups
Every field is sorted by the time since it was last confirmed:
- 0 to 3 months
- 3 to 6 months
- 6 to 12 months
- 12 to 24 months
- over 24 months
- never confirmed
Phones and mailing addresses are measured apart. A phone fails when it is disconnected or reassigned, and an address fails when the owner moves, so one blended rate can hide both.
A sample from each group
- Pull 50 fields at random from each group, or all of them if the group has fewer.
- Check each one the same way: a call for a phone, and a verification check or the returned-mail history for an address.
- Record one of four outcomes for each: reached the owner; reached someone else, or the owner no longer lives or works there; dead, meaning disconnected or the letter came back; unknown, meaning no answer or nothing returned yet.
- Leave unknown out of the rate until it resolves, since counting it as good makes old records look healthier than they are.
An invented example: 10 of 50 sampled numbers fail, a rate of 20 percent. At 95 percent confidence the Wilson method puts the range at about 11 to 33 percent. With 3 failures in 50, Wilson gives about 2 to 16 percent. The two ranges overlap only between 11 and 16 percent.
The usual normal approximation gives about 9 to 31 percent for 10 of 50, and for 3 of 50 a range whose lower end falls below zero. The NIST/SEMATECH e-Handbook of Statistical Methods cites Agresti and Coull's recommendation of the Wilson method for virtually all combinations of sample size and proportion, and it counts a method that yields an impossible lower limit as inferior (NIST/SEMATECH e-Handbook, 7.2.4.1).
Reading a curve
An invented example: 50 numbers sampled from each age group of one owner list give the curve below.
| Time since last confirmed | Sampled | Dead or wrong person | Share failed |
|---|---|---|---|
| 0 to 3 months | 50 | 3 | 6% |
| 3 to 6 months | 50 | 5 | 10% |
| 6 to 12 months | 50 | 8 | 16% |
| 12 to 24 months | 50 | 14 | 28% |
| over 24 months | 50 | 21 | 42% |
| never confirmed | 50 | 19 | 38% |
In the invented list, the share that fails climbs with age, and the never-confirmed group fails almost as often as the oldest group. The same comparison on a real sample shows whether numbers nobody ever confirmed behave like old ones.
The re-check age
The re-check age comes from the curve and a line set in advance.
- Pick the failure share that justifies paying for a re-check, and write down why; the invented list uses 15 percent.
- Find the first age group whose share passes that line; in the invented list, 6 to 12 months.
- Re-verify each field before it reaches that age, starting with the never-confirmed group.
- Sample again three months later, and move the line if the curve moves.
The line depends on what a failed attempt costs. A dead number costs a caller a couple of minutes, and a returned letter costs postage and printing, so a call list and a mail list can carry different lines.
For mailing addresses, a licensed provider can match a list against the Postal Service's change-of-address data and return the new address for anyone who filed a change. Full-service providers hold 48 months of those records, and other licensees 18 months (NCOALink). The data holds only filed changes of address, so an address the match leaves alone is still unconfirmed and keeps its own date.
Limits of a sampled curve
- It measures whether a contact point still reaches the owner, and says nothing about whether the owner wants to talk.
- Records nobody tries go unmeasured. A curve built from calls to the highest-scored leads describes those leads and no others.
- A small group carries a wide range. This site's working rule merges neighboring groups until each holds at least 30 fields.
Sources
- U.S. Census Bureau, "United States Migration/Geographic Mobility At A Glance: American Community Survey 1-Year Estimates": census.gov
- Federal Communications Commission, Second Report and Order FCC 18-177, "Advanced Methods to Target and Eliminate Unlawful Robocalls," adopted December 12, 2018: docs.fcc.gov
- United States Postal Service, PostalPro, "NCOALink": postalpro.usps.com
- U.S. Census Bureau, "Business Dynamics Statistics": census.gov
- United States Postal Service, Office of Inspector General, "Undeliverable as Addressed Mail," Audit Report MS-AR-14-006, July 14, 2014: oversight.gov
- NIST/SEMATECH e-Handbook of Statistical Methods, section 7.2.4.1, "Confidence intervals": itl.nist.gov