From monthly reports to county trends

• · CDSS SOC 242: monthly Adult Protective Services reports filed by each of California's 58 counties, restricted here to financial-abuse allegations.
• · U.S. Census Bureau: annual county population age 65+, used as the exposure denominator, plus American Community Survey demographic context (income, poverty, education, broadband, disability, living alone).
• · USDA ERS: rural-urban continuum codes used to describe county setting.

The inferential unit of analysis is the county-month. Fiscal-year rates are descriptive summaries constructed from the monthly records. The core measure is the number of reported financial-abuse allegations per 10,000 residents age 65+. An allegation is a report received by an APS agency. It is not an investigation outcome, a substantiation, or a confirmed case.

Descriptive rates are published as a low, midpoint and high scenario. When a county publishes every month exactly, the three coincide and the uncertainty band collapses to a single line.

Of 4,176 county-months assembled, 2,901 were published as exact counts, 1,195 were published as “*”, and 80 were blank. 28.6% of all county-month observations were suppressed (1,195 of 4,176).

Across the six completed fiscal years, published SOC 242 files consistently showed numeric zeroes, no positive numeric values from 1 through 10, and a minimum published positive value of 11. The Observatory therefore treats “*” observations as interval-censored counts from 1 through 10. This interval is empirically inferred from the publication pattern and is not presented as an officially documented CDSS suppression rule.

A suppressed month indicates a small non-zero count. Substituting zero would bias small rural counties downward, and dropping the month would bias them upward whenever suppression is concentrated in low-volume periods. Interval censoring avoids both by letting the model use the plausible range instead of a single imputed value.

Descriptive rates are shown under low, midpoint, and high suppression scenarios, while inferential county trend models incorporate suppressed observations directly through interval censoring.

Fully blank county-months had no published value across the SOC 242 reporting fields used in the audit. The reason for those blank reports is unknown: a blank report does not mean a county failed to report, and it does not mean the true count was zero.

Each county's monthly series is fitted with a Negative Binomial model used to accommodate overdispersion in monthly allegation counts, using the log of the year-specific county population age 65+ as an offset, so the fitted trend describes change in the reported allegation rate relative to the senior population rather than change in raw counts. Suppressed months enter through an interval-censored likelihood. The time coefficient is reported as an estimated annual percent change in the reported allegation rate.

Because 58 counties are tested simultaneously, raw p-values would produce false positives. Benjamini–Hochberg correction yields the q-values shown on the Trends page, and classification follows the corrected result:

• · Increasing: positive trend, significant after FDR correction (34 counties).
• · Decreasing: negative trend, significant after FDR correction (1 county).
• · No clear trend: not significant after correction (23 counties). This is an absence of evidence, not evidence that the county was flat.

Statewide rates pool county counts and divide by total senior person-years for that fiscal year, again under all three suppression scenarios. The statewide midpoint rate moved from 86.7 in FY2019–20 to 123.8 in FY2024–25 per 10,000 residents age 65+.

This statewide series is descriptive. No statewide trend model is fitted and no statewide significance is claimed; the inferential results on this site are the county-level trend tests.

• · Reported allegations measure agency contact, not prevalence. Cases never reported are invisible to this data.
• · County reporting practices, intake coding, and outreach differ, so direct cross-county comparisons of reported levels require substantial caution.
• · Counties with very small senior populations have volatile annual rates; their trends carry wide uncertainty even when statistically significant.
• · Demographic variables are published as context only. No model on this site predicts an expected reporting rate from demographics, and no county is labelled as reporting above or below expectation.
• · Fiscal-year boundaries and mid-window reporting-system changes can affect comparability within a county.