How the six-component ERDI is calculated
Page updated October 26, 2026: monthly and weekly reference wording now distinguishes the current observations from the preserved October 5 calculation example. Historical published values are unchanged.
This October 5, 2026 method starts a new series. Each of six components receives equal weight, one-sixth. For search, existing-home turnover, new-housing permits, weather exposure and matched review counts, the component index equals 100 × current measure / its stated reference. Mortgage financing reverses the ratio: 100 × reference mortgage rate / current mortgage rate, so a higher mortgage rate lowers its score. The composite is the arithmetic mean of all six unrounded component indices, rounded once to one decimal for display. It is not an estimate of job growth, lead volume or future revenue.
A component’s 100 means equality with its own reference, not that all six share a single seasonal norm. Search is the mean of ten national Google Trends trade-search series for the latest complete Sunday–Saturday week compared with the preceding 52 weekly points’ median in the same retrieval vintage. These are relative search-interest scores, not search volumes or booked leads. Existing-home turnover compares the qualified month’s original National Association of Realtors release, expressed as a seasonally adjusted annual rate, with the mean of the original releases for the same month in the preceding three years. The October 26 edition uses September 2026 against September 2023, 2024 and 2025; the October 5 example below retains its August observations. Original-release values are kept separate from later revised series. Permits are Census/HUD monthly new-private-housing authorizations in thousands of units at a seasonally adjusted annual rate, compared with the mean for the same month in the preceding three years. The October 26 edition uses September observations; the October 5 example retains August. Mortgage financing compares Freddie Mac’s qualified weekly 30-year fixed mortgage average with the mean of the corresponding weekly observations in the preceding three years, then inverts the ratio. The October 26 edition uses the October 22 observation; the October 5 example retains its early-October reference. Those monthly and weekly observations retain their real dates rather than being called October weekly events.
Weather uses ten fixed NOAA stations. The latest fully observed Sunday–Saturday week must contain 70 distinct station-days; if the target week is incomplete, only the immediately preceding fully observed week may be used. For each station and day, calculate the absolute difference between the mean of its daily maximum and minimum Fahrenheit temperatures and 65°F. Average those departures across the ten stations, then sum seven days. This station-first measure is compared with the nearest-week median across three earlier years. A one-week lag is a freshness policy, not statistical validation or a claim of local representation. For October 5 the September 27–October 3 target lacked 40 station-days; September 20–26 was complete and is the displayed window.
The review-count component compares net signed changes among the same 929 business identities at both ends of each interval. The frozen starting identities are matched to provider identities; missing or changed identifiers are excluded rather than filled with zero. In the preserved October 5 example, the September 13–October 5 interval covers 22 actual elapsed days: 3,986 positive count changes and 1,028 downward count changes, net 2,958. Its rate is 2,958 / 929 / 22 × 7 = 1.013 net reviews per business per seven-day equivalent. The one available September 6–13 reference interval covers seven days: 1,337 positive and 123 negative, net 1,214, or 1,214 / 929 / 7 × 7 = 1.307. The component is 100 × 1.013112829… / 1.306781485… = 77.5. The cohort is fixed across the two displayed intervals; the retained cohort identity digest is 8bacd8f0741bc163082937d98812a8467edd942104a1b3263a7a3f63a91aa046. One reference interval supports a provisional description, not a seasonal norm. The 22-day total does not reconstruct unobserved weeks. Review counts include late posts, corrections and removals; they are not jobs or revenue.
For the October 5 edition, the component indices are 74.5981 search, 100.3361 turnover, 94.9042 permits, 91.3462 financing, 102.0896 weather and 77.5273 reviews. Their unrounded mean is 90.1336, displayed as 90.1. Method version: erdi-six-interval-v1-20261005. The review-count baseline remains provisional. Source windows and retrieval dates remain distinct. Future revisions must state the affected source vintage, method version and whether a print is comparable.
Historical boundary
This method does not retroactively certify, reweight or replace prior ERDI publications. The previous public methodology describes unequal weights and a shared three-year same-week seasonal norm. Those earlier claims do not describe the current method. Older published values retain their original dates, scope and method labels; do not chart their levels beside the new 90.1 as one continuous series or compute a week-over-week change. Correction dated October 5, 2026: this page previously described unequal weights, a universal three-year seasonal reference and eight regional indices. This edition uses equal weights and source-specific references; it does not publish validated regional indices. Earlier published values are preserved in the report archive; this notice does not recertify them.