How Effective are Public
Health Departments at
Improving Health Status
and Preventing Mortality?
Evidence from California County Departments
of Public Health
Timothy T. Brown, PhD
School of Public Health
University of California, Berkeley
10th Annual Public Health Finance Roundtable
November 16, 2014
Funding
Robert Wood Johnson Foundation
Public Health Services and System Research
ID Number 67617
Roadmap
 What do Departments of Public Health do?
 Prevention takes time
 How do we figure this out?
 Results
 Limitations
Prevention Takes Time
 Prevention will usually affect health status
before mortality with very short or no lag
 It generally takes at least a decade for
changes in health status to fully impact
mortality rates
 Models need to incorporate lag times to
account for the overall impact of public health
spending on mortality
Prevention Takes Time
How do we figure this out?
 Short-term and long-term relationships
 Lags are likely, possibly extended lags
 Requires panel data
Koyck Distributed Lag Model
 Flexible lag structure – based on fit to data
 Instrumental variables to obtain causal
estimates (correct reverse causation,
measurement error, omitted variable bias).
Koyck Distributed Lag Model
Results – Overall Pattern
 Model analyzing self-rated health finds
approximately 200,000 improve their health
immediately – CAUSAL EFFECT
 Over a decade 26,937 lives per year are
saved (about 14%) – CAUSAL EFFECT
 Thus, with every round of funding,
approximately 200,000 improve their health
status. Of these 200,000, approximately
27,000 do not die who otherwise would have.
Results – Overall Pattern
 Average long-run impact
 9.1 lives saved per 100,000 for every
$10 per capita invested.
 Cost per life saved: $109,514
(limited societal perspective of public
health agencies – does not include costs
to individuals using programs – e.g., cost
of any lifestyle changes)
 26,937 lives per year
Discussion - Comparisons
 Cost per life saved: $109,514
 Flu-vaccine for adults over age 50
$35,000 per life saved
 Mammography
$100,000 per life saved
 Higher nurse-to-patient ratio
$136,000 to $449,000 per life saved
 Mandated mental health insurance
$1.3 million per life saved
Discussion – Overall Pattern
 Results can be expressed differently
- Can value statistical lives ($7.9 million – EPA)
- Can include the value of change in health status
from other research
- Using the above information and adding additional
information, results can be expressed in costbenefit terms or “societal ROI” terms
Limitations
 Only valued mortality, value of improved health status is
not included (societal benefits are underestimated)
 Longer panel of data may yield different results
 California population is racially diverse and culturally
distinct, which may limit external validity
Publications
Brown, TT. (2014). How Effective are Health
Departments at Preventing Mortality? Economics
and Human Biology 13, 34-45. (Released online
in 2013). PHSR Article of the Year
Brown TT, Martinez-Gutierrez MS, Navab B.
(2014). The Impact of Changes in County Public
Health Expenditures on General Health in the
Population. Health Economics, Policy and Law 9,
251-269.
Thank You!
Econometric Estimation
 Generalized method of moments
 Unit root test
 Clustered standard errors (by county)
 Lewbel instrumental variables
 Weak instrument test
 Underidentification test
 Overidentification test
Data (2001-2008)
 California Department of Health Services
 California State Controller’s Office: Counties
Annual Report
 U.S. Census (estimates)
 U.S. Bureau of Economic Analysis
 RAND
 HealthLeaders-InterStudy
 San Francisco County, Alpine County
omitted
 56 counties x 8 years = 448 observations
Plenty of within-county variation
 All-cause mortality per 100,000
 Within-county standard deviation: 12.11 to 190.97
(median: 34.04)
 Public health expenditures per capita
 Within-county standard deviation: $2.10 to $92.10
(median: $8.17)
Koyck Distributed Lag Model
y = all-cause mortality per 100,000
x = public health expenditures per capita
k = vector of private insurance, Medicare, Medicaid, proportion of
population by age, proportion of population by race/ethnicity,
crime index, relative per capita income, unemployment, education
proxies, population density
f = year fixed effects
Koyck Distributed Lag Model
Koyck Distributed Lag Model