Transcript Fixed and Random Effects Models for Understanding Mental Health

Fixed and Random Effects Models for
Understanding Mental Health Treatment
Disparities
Lonnie R. Snowden
Health Policy and Management Program
School of Public Health
University of California at Berkeley
Research Supported by NIMH R01 MH070942
Arriving at Limited English Proficiency as a
Research Interest
Core interests, expertise, competencies
Enlarging the scope of interests and projects:
Reaching out from the core
Cumulative learning from projects
Team membership and building: Complementary
roles among trusted collaborators
Snowden Connected
Center for Drug Abuse Prevention in the Child Welfare
System (J. Landsverk, PI)
Child and Adolescent Services Research Center (J.
Landsverk, PI)
George Warren Brown School of Social Work’s Center for
Mental Health Services Research (E. Proctor, PI).
Genentech MOSAIC Project
Chapin Hall at the University of Chicago
Some Distinguishing Characteristics of Policy
Research
Social indicator measurement: Working to identify data and
to assemble, clean, analyze data sets
Studies natural variation in policy options
Time varying and level-of-political-authority varying
(different nations, states, counties, etc.): “Cross-sectional,
longitudinal”
Statistical and logical control
Policy Research: Have Lower Tax Rates on
Upper Income Earners Stimulated the US
Economy?
Sixty-five year post war economic observation period
What is the association between lower top marginal tax rates
and economic well-being?
Correlating highest marginal tax rate with: savings,
investment, productivity
Thomas L. Hungerford, TL (2012). Taxes and the Economy: An
Economic Analysis of the Top Tax Rates Since 1945. Congressional
Research Service
Have Lower Tax Rates on Upper Income
Earners Stimulated the US Economy?
“There is not conclusive evidence, however, to
substantiate a clear relationship between the 65-year
steady reduction in the top tax rates and economic
growth. Analysis of such data suggests the reduction
in the top tax rates have had little association with
saving, investment, or productivity growth”
Hungerford (2012)
Controlled Policy Research
“Correlation (alone) does not prove causation”: That
events co-occur, or that one follows the other, is
insufficient to establish a causal connection
Joint common causes and reverse causation
Reducing the possibility of coincidence
Controlling in Research Tradition #1:
Eliminating “Confounds”
Followed by Psychologists, Psychiatrists, Nurses,
Social Workers, others
Campbell D & Stanley J (1973). Experimental and
Quasi-Experimental Designs for Research
Catalogue of “confounds”: history, testing,
maturation, etc. Notation: “X O X”
De-confounding via randomized experimentation
Controlling in Research Tradition #2:
Avoiding Bias
Followed by Economists, Epidemiologists, some
Sociologists
Confounds inflate or deflate accurate estimation of
population direct effects
“Endogeneity”=confounded
Policy research is a creature of Tradition #2
LEP and Mental Health: Reviews
Bauer AM, Alegría M (2010). Impact of patient language
proficiency and interpreter service use on the quality of
psychiatric care: A systematic review. Psychiatric Services,
61,765–773.
Snowden LR, Masland M & Guerrero, R (2007). Federal civil
rights policy and mental health treatment access for limited
English proficiency persons. American Psychologist, 62, 109117.
Threshold Language Studies: Publications
Snowden LR, Masland M & Guerrero, R (2007). Federal civil rights
policy and mental health treatment access for limited English proficiency
persons. American Psychologist, 62, 109-117.
Snowden LR, Masland M, Peng CJ, Lou C, Wallace N (2011). Limited
English proficient Asian Americans: Threshold Language Policy and
access to mental health treatment. Social Science and Medicine, 72, 230237.
McClellan, S, Wu, F, Snowden, LR (2012). The Impact of Threshold
language assistance programming on the accessibility of mental health
services for persons with limited English proficiency. Medical Care, 50,
554-558.
Threshold Language Studies: Publications
McClellan S & Snowden LR (Under review). Threshold
language policy’s impact on access to minimally adequate
mental healthcare among persons with limited English
proficiency.
Patel S, Firmender W & Snowden LR (Under review).
Threshold language policy’s language assistance
programming: Mental health consumers’ perspectives
Threshold Language Studies: Publications
Masland MM, Lou C, Snowden LR (2010). Use of
communication technologies to cost effectively increase the
availability of interpretation services in healthcare settings.
Telemedicine & e-Health, 16, 1-7.
Snowden LR & McClellan S (Under review). Spanish
language community-based mental health treatment programs
, policy-required language assistance programming, and
mental health treatment access in Spanish speaking Medicaidinsured population. American Journal of Public Health
LEP and Treatment Access: Results from the
California Health Interview Survey
English Only=29,991
Bilingual=9243 (Speaking a language other than English at
home and speaking English “well or very well”)
No English=2,750 (Speaking a language other than English at
home and speaking English “not well or not at all”)
Controls: Ethnicity, US born, years in US, Insured, poverty,
age, married, good health, education, urban
LEP and Treatment Access: Results from the
California Health Interview Survey
Results from California Health Interview Survey bivariate
analysis: (Confirmed in Multivariate Analysis)
Need for MH Care
English Only
Bilingual
No English
18%
14%
16%
Received Needed Care
51%
42%
8%
Sentell T, Shumway, M, Snowden, LR (2007). Access to mental health treatment by English
language proficiency and race/ethnicity. Journal of General Internal Medicine, 22, 289293.
.
California’s Threshold Language Policy Response to
Limited English Proficiency
A widely implemented response to Title VI: Threshold
language policy
Trigger: “Surpassing threshold”: 3,000 persons or 5% of
Medical enrollees, whichever is lower, speakers of nonEnglish language
Threshold Language Policy-required programming in
California: Notification translators, translated written
materials, 24-hour hotline
Qualitative Methods
Administering and interpreting the SEMI with Spanishspeaking and Vietnamese-speaking consumers
Ten speakers of each language
Personal experience of requirements in context of personal
distress and assistance seeking
Purposes: Flesh-out interpretation of findings, generate
hypotheses
The Quantitative Threshold Language Study
Public mental health services consolidated, decentralized to
county level
Wide variation among plans and county environments (e.g.
LA vs. many counties with population <30,000)
Quasi-experimental research via cross sectional-longitudinal
econometrics
Unit of Analysis: County x quarter- County penetration rate
as observed during a quarter (57 x 36 = 2052): No individual
people
The Quantitative Threshold Language Study
Dependent Variables: Medicaid (Medi-Cal) Penetration Rates
and Quality Indicators
County Administrators Survey: 66% response rate
Key independent variable #1: notification
Key independent variable #2: threshold language required
programming at minimally adequate levels (24 hour crisis
line, translation of materials, use of translators, notification)
The Quantitative Threshold Language Study
Quasi-experimental time series with staggered onset of
Independent Variable:
1. Intervention counties serve as “own controls” preintervention
2. Counties with no threshold languages serve as
counties
control
3. Counties with other-than target threshold languages
serve as control counties
The Quantitative Threshold Language Study
Other controls:
1. Fixed effects (preferred) rule out static inter-county
differences, whether observed or unobserved
2. Linear time trend
3.English penetration rate
4. Selected variables (time varying covariates): e.g. percapita income
Notification’s Impact on Access to Care for Speakers
of Threshold Asian Languages
Short term: Thirteen quarters (July 1998 to June, 2001)
Aggregated Asian Languages: Vietnamese, Cantonese,
Hmong, Cambodian
Competing or Complementary Services Controlled: Cultural
Competence Training, Bilingual Staff, Asian LanguageSpecific clinics and service program.
Snowden LR, Masland M, Peng CJ, Lou C, Wallace N (2011). Limited English proficient
Asian Americans: Threshold Language Policy and access to mental health treatment.
Social Science and Medicine, 72, 230-237.
Notification’s Impact on Asian-Language
Penetration Rates Over Three Years
(Partial Results)
B
SE
Time
0.71
0.22
Threshold × time
0.85*
0.34
Threshold
14.6* 3.90
Independent variable
Language Assistance Programming’s Impact
Thirty-nine quarters over 10 years
Russian, Vietnamese, Spanish
Full implementation of threshold language assistance
programming package directly assessed
Measures programming package’s implementation whenever
it occurred, not in relation to trigger
McClelland S, Wu F& Snowden LR (2012). The Impact of threshold assistance programming on the
accessibility of mental health services for persons with Limited English Proficiency in the Medi-Cal
Setting. Medical Care, 50, 554-558.
Threshold Language Programming’s Impact on Penetration
Rates Over 10 Years
Predictor
Russian
(N=390)
Coefficient
Time (Quarters)
-0.025
Programing
8.173**
Programing * Time
0.296 **
Threshold notification
0.787
English Penetration
Rate
0.535**
Eligible Benes
(1,000)
-3.439**
Per Capita Income
(1,000)
0.260 **
Republican Party
-0.397
Spanish
(N=1326)
Coefficient
Vietnamese
(N=663)
Coefficient
0.006
0.110
-0.001
-0.105
-0.009
3.266**
0.018
-0.117
0.096**
0.444**
0.000
-0.196
-0.018*
-0.005
0.117*
0.343
Spanish-Language Treatment Programs, Policy- Required
Language assistance programming, and mental health
treatment access among Spanish-speaking clients
Spanish-speaking LEP and threshold language policy’s
language assistance programming: No global increase in
penetration rates
Specialized, Spanish language clinics: County operated vs.
Community-Based Organization-operated under contract
Closing English-speakers vs. Spanish speaking LEP access
disparities state wide?
Spanish-Language Treatment Programs, Policy- Required
Language assistance programming, and mental health
treatment access among Spanish-speaking clients
Fixed county effects
Quarterly time tend
LEP programming
Programming*Qtr
No Use of
Contractor
Programming*
Contractor
English Pen Rate
Eligible (1,000)
Per capita income
Votes republican
Programming*Cont
ractor*English Pen
Rate
0.007 (0.004)*
0.053 (0.075)
-0.002 (0.003)
-0.137 (0.246)
0.280 (0.115)**
0.090 (0.008)**
0.000 (0.001)
-0.034 (0.009)**
0.033 (0.080)
0.034 (0.021)