Rising autism prevalence: real or displacing other mental disorders? Evidence from demand for auxiliary healthcare workers in California.
Dave, Dhaval M. ; Fernandez, Jose M.
I. INTRODUCTION
Autism is a developmental disorder characterized by impairments in
social interaction, communication, and restricted or repetitive
behaviors (American Psychiatric Association 1994). The previously rare
condition has experienced a dramatic increase among 8-year-old children
with a prevalence of 0.5 in 1,000 children during the 1970s to 14.7 in
1,000 children in 2010 (Baio 2014). (1) Currently, autism is only second
to mental retardation (MR) as the most commonly diagnosed developmental
disability (Bhasin et al. 2006; Yeargin-Allsopp et al. 2003).
Furthermore, the cost associated with caring for individuals with autism
is not trivial. The annual costs of care for a child with autism are
estimated to be 85%-550% higher than the cost for a typically developing
child and the average lifetime public expenditures are approximately
$4.7 million (Guevara et al. 2003; Jacobson and Mulick 2000). This
increase in prevalence, compounded with the cost of care, has prompted
many researchers to investigate various factors that could potentially
explain the cause of such an increase.
The three leading factors are an increase in environmental toxins,
(2) de novo (3) gene mutations caused by older parental age at birth,
and a change in diagnosis determinants. The latter factor has led to
much debate in the medical community as researchers cannot agree on the
true incidence level of autism. The diagnosis of autism poses a
difficult challenge to clinicians in that there are no biological
markers and many of the observable characteristics are shared with other
mental disorders such as MR and attention deficit disorder (ADD). For
example, King and Bearman (2009) found that roughly 25% of the increased
prevalence of autism is associated with a diagnostic change in
determining MR. Bishop et al. (2008) studied adults with a history of
developmental language disorder and found that one-third of adults
previously diagnosed with this disorder would have been diagnosed with
autism using contemporary techniques.
In this study, we use economic theory to help shed some light on
the debate. If the incidence of autism is increasing independently of
other mental disorders (such as MR), then the demand for auxiliary
health providers (e.g., speech pathologist, behavioral therapist,
occupational therapist, etc.) should increase, leading to higher wages
and labor supply for these providers as well as an increase in the
number of providers. However, if the increase in autism diagnosis is
merely displacing other mental disorders, then the effects of the
increase on demand will be mitigated or not present as individuals are
only changing diagnostic labels but maintaining the same level of
services demanded. We construct an econometric model to distinguish
between these two effects.
Using data from the California Department of Developmental
Services, we study how changes in the number of autism cases at each of
the 21 regional development centers affected local wages, labor supply,
and quantity of auxiliary health providers. We focus on this subset of
health providers because, unlike physicians and psychologists who can
diagnose autism, these workers cannot induce their own demand. (4) Using
data from the regional development centers, we estimate that one-third
of the increase in autism stems from displacing MR diagnoses, but not
other mental disorders, which is consistent with previous studies (Coo
et al. 2008; King and Bearman 2009). Using annual wages and provider
counts from the American Community Survey (ACS) from 2005 to 2011
(referencing 2004-2010), we find that a 100% increase in the number of
autism cases at a regional center, which is approximately the average
increase experienced in CA over the sample period, raises the wages of
auxiliary healthcare workers over non-autism healthcare occupations by
8%-11%; additionally, the number of providers increases by 9%-14% over
the following 2 years.
II. BACKGROUND
Autism is a developmental disorder that limits an individual's
ability to form social relations and appropriately respond to
environmental stimuli. The prevalence of autism in the United States has
increased rapidly over the last 30 years. In California, the number of
autism cases increased by 1,148% between 1987 and 2007, which is
remarkable considering that cerebral palsy increased by 73%, epilepsy by
66%, and MR by 95% over the same time period (Cavagnaro 2007). However,
the co-morbidity rate between autism and MR has fallen from 79.6% in
1987 to 35.6% in 2007, which differs from national trends ranging now
between 40% and 55% (American Psychiatric Association 2000; ASD Best
Practice Guidelines 2002; Chakrabarti and Fombonne 2001).
A confounder in distinguishing between autism and mild MR arises
when the child presents speech delay, rendering important IQ
measurements useless. When initial testing does not present a clear
diagnosis, physicians often label the child as having
"developmental delay," which is the same categorization used
for autistic children prior to a diagnosis. MR is disaggregated into
severity levels by IQ: mild (70-50), moderate (49-35), severe (34-25),
and profound (less than 25). According to the American Academy of Family
Physicians, mild MR represents nearly 75% of all MR cases. These
individuals may have no unusual physical signs and may be able to
perform practical tasks, which are two traits shared with most cases of
autism. However, autism is not bounded by IQ restrictions, and an
individual with mild MR is still capable of appropriate social
interaction.
We highlight both the rise in reported autism cases as well as the
trade-off between autism and MR as these relationships are at the center
of much debate. Grinker (2007) argues that the true prevalence of autism
is not rising. Rather, physicians are finally diagnosing autism
correctly. The author cites five factors causing the rise in autism
cases: (1) broader definition of autism; (2) a change in school policy
allowing autistic students to receive special education in 1992; (3) the
decreased stigma associated with an autism diagnosis reducing the number
of under-reported cases; (4) states started to allow families with
autistic children to apply for Medicaid funds regardless of family
income leading more families to seek care and more physicians to
"up-code" patients so they can receive funding (an advantage
not readily available for a diagnosis of MR); and (5) relabeling, where
due to the broadening of the autism diagnosis fewer cases are
misdiagnosed into other mental disorders such as MR and attention
deficit hyperactivity disorder (ADHD).
Mental disorders have a history of underreporting and misdiagnosis.
Elder (2010) uses regression discontinuity methods to compare ADHD
diagnosis between students who are slightly above and below the birthday
enrollment cutoff for kindergarten and finds that approximately 20% of
children who use stimulants intended to treat ADHD are misdiagnosed.
Epilepsy is found to have a misdiagnosis rate of 26.1% in adults (Smith
et al. 1999) and 39% in children (Uldall et al. 2006).
A similar pattern is observed with autism. Autism was first
recognized as a disorder associated with schizophrenia beginning in 1911
(Bleuler 1912). This association continued well through the 1960s.
Autism is later categorized into four individual disorders comprising
the autism spectrum: Rett syndrome, Asperger syndrome, Autistic
disorder, and pervasive developmental delay-not otherwise specified
(PDD-NOS).
PDD-NOS is the primary source of growth within the spectrum and
could be the source of misdiagnoses. Tomanik et al. (2007) run an
experiment where independent physicians are asked to review cases for
autism. There are two groups of patients, 77 autistic patients and 52
non-autistic patients. The authors find that 21.2% of individuals who
are not classified as autistic receive an autism diagnosis and 23.4% of
autistic individuals are not diagnosed with autism. With respect to
under-reporting, Liu et al. (2010) find the reduction in stigma
associated with autism has led to a 16% increase in the number of
reported cases between 2000 and 2005. Furthermore, autism diagnosis
rates can differ based on demographic characteristics. Mandell et al.
(2002) use administrative data for Medicaid patients in Philadelphia and
find that African American children receive an autism spectrum disorder
diagnosis nearly 2 years later than Caucasian children, on average.
The displacement effect of autism has been previously studied. As
noted above, King and Bearman (2009) find a 25% displacement rate of
autism on MR diagnoses. Coo et al. (2008) report that one-third of the
diagnoses of autism from 1996 to 2004 in British Columbia, Canada
resulted in a switch from a diagnostic category other than autism to
autism. Nassar et al. (2009) also report diagnostic substitution. They
note that as incidents of severe intellectual disability decreased by
10% in the state of Washington (USA), rates of autism increased by an
average of 22%. Shattuck (2006) analyzes administrative data from the
U.S. special education system for students between the ages of 6 and 11
years and compares how changes in the administrative prevalence of
autism versus other mental disorders affect the demand for special
education classes. The author finds that the average administrative
prevalence of autism among children increased from 0.6 to 3.1 per 1,000
from 1994 to 2003, whereas the prevalence of MR and learning
disabilities declined by 2.8 and 8.3 per 1,000, respectively, over the
same time period. The increased administrative prevalence in autism is
concluded to be offset by the decrease in prevalence in other mental
disorder categories.
III. METHODS
The primary aim of this study is to assess whether, and to what
extent, the increase in autism diagnoses impacts the demand for
auxiliary healthcare workers. As these workers do not diagnose autism,
in contrast to physicians and psychologists, they cannot induce their
own demand. (5) Any shift in autism diagnoses in the area therefore
represents an exogenous shift in the demand for their services. In the
short term, this potential increase in demand would be reflected in the
form of higher wages and salary among these workers, likely because of
increased work effort and/or a higher price for their services. This may
also induce a higher entry of such providers over time into the market
area experiencing the increase in diagnoses, demand, and wages. However,
to the extent that the increase in autism cases is displacing diagnoses
of other mental disorders, notably mild or moderate MR, the increase in
the demand for auxiliary healthcare workers would be mitigated as one
diagnosis is simply replacing another.
First, we therefore study the association between autism diagnoses
and MR over the sample period to assess whether, and the extent to
which, the increase in autism cases may be displacing MR diagnoses in
CA.
(1) [MR.sup.jt] = [[delta].sup.0] + [LAMBDA] [(Autism).sub.jt] +
[[delta].sub.1] [(Total clients).sub.jt] + [[delta].sub.2] [(Center
demographics).sub.jt] + [A.sub.j][OMEGA] + [Z.sub.t][PHI] + [V.sub.jt]
The parameter [LAMBDA], in Equation (1), represents the association
between Autism and MR, which we measure alternately as percentage of
total clients and as the number of total cases in regional center j in
year t. A negative A would suggest that increases in autism cases are
displacing MR cases. Alternate specifications control for center
demographics, center-specific indicators ([A.sup.j]) to account for
time-invariant factors that may differ across the regions, and year
indicators ([Z.sub.t]) to account for general trends in the state that
may be affecting diagnoses, healthcare coverage, and other unobserved
factors. The disturbance term is represented by v, and we adjust
standard errors for arbitrary correlation in this error term within
centers over time.
Next, we turn to our main analyses, which examine the impact of the
increase in autism diagnoses on the wages of auxiliary healthcare
providers in the short term and, in alternate specifications, their
labor supply at the intensive margin (annual hours worked), and the
number of providers over the medium term. We use a quasi-experimental
research design--akin to a pre-and post-comparison with treatment and
control groups--in conjunction with multivariate regression methods. The
following specification relates changes in wages to autism diagnoses:
(2) Ln [Wages.sub.ijt] = [[alpha].sub.0] + [pi] (Ln
[Autism.sub.jt]) + [X.sub.jt][beta] + [O.sub.i][PSI] + [A.sub.j][OMEGA]
+ [Z.sub.t][OMEGA] + [[epsilon].sub.ijt].
Equation (2) posits that log wages {Ln Wages), for the ith
auxiliary healthcare occupation in center j during year t, is a function
of the number of autism diagnoses (Ln Autism). The parameter of interest
is [pi], which captures the effects of autism diagnoses on the average
wages of those healthcare providers whose services are complementary to
the treatment of autism. The parameter e represents an error term at the
level of the occupation, center, and year. We use a log transformation
of wages and autism diagnoses, separately controlling for the
county-specific population base and the total number of cases in each
center, and allowing these coefficients to remain unrestricted. The log
adjusts for the skewness of the wage and diagnoses distributions,
facilitates interpretation (in terms of elasticity), and makes the
effect magnitudes comparable across outcomes. (6) We estimate models for
wages and hours worked using ordinary least squares (OLS). For the
number of providers, we use a Poisson regression model for two reasons.
First, the discrete nature of the outcome variable as a count of service
providers makes the Poisson probability distribution especially
suitable. Second, the Poisson framework does not suffer from the
"incidental parameters" problem and can accommodate fixed
effects well (Cameron and Trivedi 1998). We adjust standard errors on
the conservative side to account for arbitrary correlation within
centers, across occupations, and over time.
A challenge in any such analysis relates to disentangling the
effects of autism diagnoses from other unobserved factors that may also
affect the outcome. We account for such confounding factors in various
ways. First, in alternate specifications, we control for a vector of
time-varying center-specific (and county-specific) characteristics (X)
including the total number of clients served within the center's
geographic area, the racial and ethnic composition of the center's
clients and at-need population, total population of the counties
reporting to the center, and the demographic (age, race, and ethnicity)
composition of the served population. It is important to control for the
racial/ethnic constitution of the county population and the clients
served by the center as research indicates substantial disparities in
autism diagnoses and the age of diagnosis across Black and White
families (e.g., Mandell et al. 2002, 2009). (7)
Year fixed effects (Z) account for unobserved trends specific to
the state of California, including changes in public and private
insurance coverage, overall economic conditions, shifts in diagnosis
criteria, and state-level policies enacted over the sample period. (8)
Alternate specifications include center-specific fixed effects (A),
which account for all unobserved time-invariant local factors that may
be differentially affecting autism diagnoses, clients served, and
treatment patterns across centers, and include occupation fixed effects
(O), which account for unobserved time-invariant characteristics
specific to each occupation (such as working conditions, nonmonetary
attributes, and stable labor demand) that may affect wages in these
occupations.
Despite these controls, the possibility remains that there may be
residual unobserved time-variant factors which potentially impact wages.
We address this problem by considering a comparison group of occupations
that should not be directly affected by a shift in autistic disorder
diagnoses. Thus, any correlation between autism diagnoses and wages for
these "control" occupations reflects unobserved
center-specific trends. This correlation can be differenced out from the
effect ([pi]) identified in Equation (2) to arrive at a cleaner estimate
of the effects of autism diagnoses on the wages of impacted providers.
This difference-in-differences (DD) effect can be obtained directly
from estimating the following specification:
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].
In Equation (3), Target represents a dichotomous indicator equal to
one for those healthcare occupations whose services are demanded by
families with autistic children, and zero for occupations in the
comparison group (providers whose services are not directly impacted by
autism diagnoses). The DD estimate of the effect of autism diagnoses is
the coefficient ([[pi].sup.*]) of the interaction term between Ln Autism
and the Target indicator. This effect is identified by comparing changes
in wages associated with autism diagnoses for the target occupations in
relation to changes for the control occupations, accounting for all
other observable factors and the fixed effects.
The choice of the target group of auxiliary healthcare occupations,
whose services are used by families with autistic children, is
straightforward. As noted earlier, nonmedical interventions include
behavioral, educational, sensory, and communication therapy, typically
requiring the services of healthcare providers such as a speech
pathologist, behavioral therapist, or occupational therapist. (9) These
occupations are supported by the Committee on Children with Disabilities
of the American Academy of Pediatrics (2001; Myers and Johnson 2007) and
the American Academy of Family Physicians (Daily et al. 2000) as
accepted therapy for both autism and MR. Data from the 2009-2010
National Survey of Children with Special Health Care Needs show that
almost 75% of children diagnosed with autism use the services of
physical, speech and language, and occupational therapists. We therefore
expect the demand for services for the following providers to be
potentially and directly impacted by a shift in autism diagnoses: (1)
audiologist, (2) occupational therapist, (3) physical therapist, (4)
recreational therapist, (5) respiratory therapist, (6) speech-language
pathologist, and (7) other therapist. (10) Additionally, these same
therapists serve individuals with MR. (11)
We use two alternate control groups to account for unobserved
trends in wages within centers over time: (1) all other healthcare
practitioner and healthcare technical occupations (occupation codes
3000-3540); and (2) all other healthcare practitioner, healthcare
technical, and healthcare support occupations (occupation codes
3000-3650). (12) The former control group includes occupations that
provide direct healthcare services to the patient. As auxiliary
healthcare providers of services to autistic children are classified by
the Bureau of Labor Statistics (BLS) in this grouping, this control
group may represent those occupations that are most similar to those in
the target group but not directly impacted by an increase in autism
diagnoses. The latter control group comprises all healthcare-related
occupations and also includes support occupations such as aides,
assistants, and other support workers. These control groups will account
for trends specific to healthcare occupations within each center and
county. However, to the extent that demand and wages for some of these
other health practitioner (physicians) and health support occupations
(e.g., aides and nurses) may also be impacted by an increase in autism
diagnoses, effects may be potentially understated. We therefore draw
conclusions based on the range of estimates from both sets of control
occupations and interpret these estimates as potentially conservative
effects. (13)
Unconditional means from Table 1 suggest that wages in occupations
whose services are complementary to autism diagnoses increased by 40%
between 2004 and 2010. This compares with wages for other healthcare
practice occupations, which increased 26%, while those among all
healthcare occupations increased 24%. Thus, wages among autism service
providers have increased relative to those among similar healthcare
practice providers as well as all other healthcare providers and support
workers, coinciding with an increase in the number of autism diagnoses
over our sample period, and this increase in wages is statistically
significant.
We extend the model specified in Equation (2) in order to test for
a "dose-response" relation and to assess the plausibility of
the estimates by exploiting the suggestive displacement of other mental
disorders for autism diagnoses. That is, we expect a stronger response
on wages for those centers where the increase in autism represents a
larger increase in demand (larger dose) for the services of the
auxiliary healthcare workers. If the increase in the diagnoses of
autistic disorders is partly displacing other mental disorders, then the
effects of the increase on demand and wages would be mitigated. If one
diagnosis is merely displacing another, then there is no effective
increase in the demand for the services of the auxiliary healthcare
workers as their services are demanded by both individuals with autism
as well as those with other mild or moderate mental disorders. Hence, if
one diagnosis is merely displacing another, then there should be no
observed increase in their wages, conditional on trends. We therefore
test whether the demand effects are larger in those areas where this
displacement is low, in which case the increase in autism diagnoses is
occurring independently of other mental disorders and would lead to a
net increase in demand for the healthcare providers.
Specifically, we estimate a version of Equation (1) for each of the
21 regional centers, as part of a two-step procedure, to quantify the
effects of higher autism diagnoses on the diagnoses of mild MR.
(4) Mild [MR.sup.t] = [[delta].sub.0] + [LAMBDA][(Autism).sup.t] +
[[delta].sub.1] (Total clients), + [[delta].sub.2] (Center
demographics), + [[delta].suhb.3] Trend + [[delta].sub.4] [Trend.sup.2]
+ [v.sub.t]
The parameter A, which is estimated separately for each center,
represents the association between Autism and Mild MR, both measured as
the number of total cases. (14) We multiply A, for each center, by -1 to
obtain the magnitude of the displacement. Thus, if the displacement is
0.5, this signifies that for every two additional diagnoses of autism,
one of these cases is displacing a diagnosis of mild MR in that center,
on average. While Equation (4) controls for linear and quadratic trends
and demographic shifts within each center, we note that these rates
should not be interpreted as causal because they may reflect residual
unobserved trends. Nevertheless, we expect to find weaker effects on the
demand for services in those centers where the displacement rate is
high.
In the second step, we modify the baseline DD model (Equation (3))
to allow an interaction between the effect of autism diagnoses and the
center-specific displacement rate (15):
(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
The parameter [[lambda].sub.1] represents the effect of autism
diagnoses on wages, among those centers which had no displacement of MR
cases for autism; hence, we expect [[lambda].sub.1] to be larger than
the estimated DD effect it [[pi].sub.1] in Equation (3) which
represented a weighted average effect that conflates effects for centers
that have high and low displacement rates. The parameter
[[lambda].sub.2] represents the effect of higher levels of displacement
on the link between autism and wages. We expect [[lambda].sub.2] to be
negative, because centers that have high rates of displacement would see
lower demand and hence lower wages, relative to centers that have no
displacement. This specification tests the proposition that higher
autism diagnoses should only represent an effective increase in the
demand for the services of the auxiliary healthcare workers if these
diagnoses represent a "true" increase rather than a
substitution from MR diagnoses. Standard errors for Equation (5) are
bootstrapped to account for the sampling variance in the estimate of the
displacement rate. (16)
[FIGURE 1 OMITTED]
IV. DATA
The autism data are collected from the California Department of
Developmental Services (DDS). The California DDS is the central
administrating body tasked with diagnosing and reporting mental health
cases in the 58 counties of the state. The 58 counties are serviced by
21 regional centers. The service areas of these regional centers are
shown in Figure 1. The data for this study are collected from the
quarterly Client Evaluation Reports, which are available for each county
or each regional center for the years 2002-2011. (17) We use the
December issue of the Client Evaluation Reports, which includes the
following client characteristics: number of total clients, number of
autism cases, number of MR cases (by severity), number of cases with
physical handicap, primary language, gender, race, and age distribution.
Next, we identify the target group of auxiliary healthcare
providers by nonmedical treatment therapy used. The National Institute
of Mental Health (NIMH) provides separate guidance manuals for autism
and MR. (18) Both manuals encourage the use of behavioral therapy (i.e.,
applied behavioral analysis [ABA]), psychotherapy (i.e., the
developmental, individual difference, relationship-based model), as well
as speech and language therapy for autism and MR. (19) From these
treatments, we identified the following providers: audiologists,
occupational therapists, physical therapists, and speech-language
pathologists. We expand this list of providers to include recreational
and respiratory therapists. The Bureau of Labor and Statistics defines
recreational therapists as individuals who plan, direct, and coordinate
recreation programs for people with disabilities or illnesses. We
include respiratory therapists as some individuals with poor brain
activity report using hyperbaric oxygen therapy. (20)
Information on annual provider income from wages and salary and
their total earnings is matched to these 21 regional centers for each
year. Specifically, we use data from the 2005-2011 ACS Public Use
Microdata Sample (PUMS) to compute occupation-specific wages and
earnings in California. The ACS is administered monthly by the U.S.
Bureau of Census. It is nationally representative, and annual estimates
are also representative of geographic units with a population of at
least 65,000 individuals; representative estimates of smaller areas may
be obtained by combining multiple years from the ACS-PUMS (3-and 5-year
estimates). The PUMS comprises approximately a 1% random sample of the
total number of housing units in the nation. It thus contains
information on about 1.34 million housing units and 3 million person
records in total, and yields about 368,809 persons sampled in California
in 2011.
The smallest geographic unit identified in the ACS-PUMS is the
Census-defined Public Use Microdata Area (PUMA). PUMAs are statistical
geographic areas nested within states, which have a population of at
least 100,000 individuals. (21) The 21 regional centers that report the
provision of services for developmental disabilities in California are
matched to the PUMAs, based on the counties that each center serves, for
2005-2011 (22) (we are unable to use ACS information prior to 2005 as
local-area identifiers within each state are not available in the PUMS).
Each individual respondent reports their average annual total
income from wages and salary, which represents earned income from
employment within an establishment, as well as their average annual
total personal earnings, which reflects all earned income and
additionally includes income from self-employment. (23) Respondents also
report on their hours worked in the past year and their occupation
(4-digit occupation code based on the 2002 Census). We compute mean
annual wages, annual earnings, and annual hours worked, and the total
number of individuals working within each occupation, for each regional
center and year. (24) As annual wages and hours worked reference the
past year, we match the ACS information at time period t (referencing
period t - 1) with the DDS information at time period t - 1, to assess
the contemporaneous effects on wages of the demand shock stemming from
the increase in autism diagnoses.
In supplementary analyses, we also assess lagged effects on wage
income and labor supply in the subsequent period by matching ACS wage
income and hours worked information at time period t (referencing period
t - 1) with the DDS information at time period t - 2. As provider counts
reference the year of interview, we match ACS information on the number
of providers in each occupation at time period t with DDS information at
time periods t - 1 and t - 2 to assess 1- and 2-year lagged effects on
provider entry. In addition, we append population figures and the
demographic (age, race, and ethnicity) composition of the population for
each center and year, derived from the U.S. Census Bureau.
[FIGURE 2 OMITTED]
Table 1 presents the means for key variables across various periods
spanning our analysis sample. Over 2004-2010, the average regional
center reported 9,229 clients served for mental health disorders, 19.4%
of whom were diagnosed with an autistic disorder and 54.4% were
diagnosed with mild or moderate MR. Most of the clients served are
minorities, with Whites representing only about 47.6% of the cases in
the average center.
As illustrated in Figure 2, the percentage of autism cases has
increased from 12% to 25% between 2002 and 2011. However, the percentage
of total MR cases has decreased from 80% to 70% and the percentage of
mild and moderate MR cases decreased from 58% to 52%. Moreover, the
total number of both autism and MR cases has grown over this time
period, but the ratio of autism cases to MR cases has increased from
0.16 to 0.35, and specifically the ratio of autism to mild MR cases has
increased from 0.32 to 0.69.
There is substantial variation in autism growth within centers
(ranging from an average of 4% to over 14% on an annualized basis). This
within-center variation is significantly positively correlated, even
without controlling for other factors, with increases in the growth rate
of annual wages for auxiliary healthcare workers (Figure 3); and the
differential growth rate of annual wages for auxiliary healthcare
workers compared with the control group of other healthcare practitioner
occupations (Figure 4). We exploit this source of year-to-year
within-center variation in the DD models, adding two alternate control
groups and adding a further source of variation based on the
center-specific displacement rates in the DDD models, after accounting
for various center-specific demographic factors along with occupation,
center, and time fixed effects--all of which further help to reduce the
sampling variance and maximize precision.
V. RESULTS
Table 2 shows the association between autism diagnoses and MR cases
over the sample period. Model 1 indicates that a 1% increase in autism
cases is associated with a statistically significant 0.40% decrease in
mild MR cases, suggestive of a displacement of MR diagnoses for autistic
disorder diagnoses. This displacement rate is robust to nonparametric
(inclusion of year indicators) controls for trends in California (Model
2) and decreases slightly in magnitude to -0.34% with the addition of
center-level fixed effects in Model 3. Thus, the negative association
between autism and MR diagnoses is present both across centers
(cross-sectionally) and within centers over time. Model 4 suggests a
similar displacement rate (-0.38%) when moderate MR cases are considered
in addition to the mild cases.
[FIGURE 3 OMITTED]
Displacement is also evident in levels (Models 5-8). (25) These
specifications suggest that about one of every three additional autism
cases is a shift from mild or moderate MR to an autism diagnosis.
Models 9 and 10 present a placebo test and confirm that with
respect to cerebral palsy and epilepsy there is no significant or
substantial displacement coinciding with an increase in autism
diagnoses. Individuals with these conditions also use occupational and
behavioral therapists, but unlike MR, these conditions are easier to
diagnose. Furthermore, as these conditions are also served by the
developmental centers, it is possible that autism is potentially
crowding out other conditions besides MR or there is an unobserved
effort by administrators to increase autism diagnosis. However, we find
no discernible displacement effect between these conditions and autism
cases at the center level.
Table 3 presents the effects of an increase in the autism caseload
on annual wages, based on the DD framework specified in Equation (3).
Models 1-4 use non-autism healthcare practitioner occupations as a
control group, whereas Models 5-8 use an alternate control group that
includes all non-autism healthcare occupations (practitioner and support
occupations). Model 1 suggests that a 100% increase in autism diagnoses,
which is approximately the increase that occurred over the sample
period, raises wages among the target occupations (auxiliary healthcare
providers of services to autistic clients) by 10.8%, relative to all
other healthcare practice occupations. This effect remains robust and
statistically significant with the inclusion of centerlevel fixed
effects (Model 2) and center- and county-level demographics (Model 3).
The effect magnitude decreases slightly to 8.2%, although it still
remains statistically significant, when the full set of
occupation-specific fixed effects is added to the regression (Model 4).
The coefficient of Ln Autism is generally insignificant although not
necessarily insubstantial in magnitude; this suggests that the control
group is indeed picking up remaining (noisy) trends in healthcare wages
within centers coinciding with the increase in autism diagnoses. The DD
estimates purge these residual trends by differencing the effects for
the control group. The coefficients of the Target indicator are
expectedly negative, consistent with the patterns suggested by the
unconditional means (Table 1) that providers of healthcare services to
autistic clients in general have lower wages relative to other
healthcare practitioners.
[FIGURE 4 OMITTED]
Models 5-8 indicate that the effect magnitudes of the DD estimates
are highly similar when the control group is broadened to include
healthcare support workers. Specifically, these estimates suggest that a
100% increase in autism diagnoses is associated with a 7.1%-10.2%
increase in the annual wages of the auxiliary healthcare providers
(relative to all other healthcare occupations).
The DD effects are understated as part of the increase in autism
diagnoses represents a displacement of MR cases, in which case the
increase in the demand for the services of the target healthcare
occupations is also mitigated. Table 2 suggests displacement rates of
about one-third to one-half, in which case the DD effects reported above
are also understated by a similar factor. If some of the increase in
autism cases is merely displacing MR cases, and if an effective increase
in autism cases causally raises the demand for services of certain
healthcare providers, then we expect that accounting for the level of
displacement or directly controlling for MR cases should raise the
magnitude of the wage effects. The models reported in Table 4 implement
this plausibility check.
Models 1-3 present estimates of Equation (5), which interact the
displacement rate with the DD effect to assess whether the impact on
wages is larger in those centers where displacement of MR for autism is
lower. The coefficient of the interaction between Ln Total Autism and
Target increases in magnitude to 17% (Model 1); it can be interpreted as
the impact of an effective 100% increase in autism diagnoses,
independent of MR cases, on annual wages. The effect is robust to the
inclusion of area and center demographics (Model 2) and declines
somewhat in magnitude to 12.7% with the inclusion of occupation fixed
effects (Model 3). It is validating that these effects (12.7%-17%) are
larger, as hypothesized, relative to the effects in Table 3 that do not
account for displacement (8.1%-11%). Furthermore, the coefficient of the
triple interaction term is negative and significant (-0.064 to -0.086),
suggesting that the effect on wages varies inversely with displacement;
in those centers where the increase in autism cases is mostly reflecting
a shift from MR to autism, we would not expect large increases in the
demand for services.
Model 4 explicitly controls for mild or moderate MR diagnoses at
the center-year level. As expected, the DD effect of a 100% increase in
autism cases (which now represents a pure demand shock for related
auxiliary healthcare providers) increases in magnitude, suggesting a
16.1% higher annual income from wages and salary. It is validating that
this estimate of the wage increase from a "pure" 100% increase
in autism diagnoses is consistent with the effects emerging from the DDD
models (12.7%-17.1%) noted above. It is further validating that the
inflation of the effect magnitudes in Models 1-4 in Table 4 relative to
similar models in Table 3 (that do not account for displacement) is
consistent with the displacement ratios estimated in Table 2. That is,
the effect magnitudes with respect to a pure 100% increase in autism
diagnoses are about 40% larger, corresponding to a displacement rate of
autism for mild or moderate MR of about 30%-40%. Models 5-8 suggest very
similar patterns and effect sizes based on the alternative control group
of all healthcare occupations.
Table 5 presents supplementary DD and DDD models similar to
Equations (3) and (5), which inform whether an increase in autism
prevalence affects the supply of auxiliary healthcare providers in the
subsequent period. Specifications (2) and (6), using alternate control
groups, indicate that a 100% increase in the prevalence of autism raises
the number of auxiliary healthcare workers in the area in the next year
by a statistically significant 8.9%-10.9%, relative to the supply of
other healthcare providers and support workers.
To the extent that the increase in autism diagnoses is supplanting
alternate diagnoses of MR. these effects underestimate the impact of an
effective 100% increase in autism cases. Specifications (3) and (7)
estimate the DDD model shown in Equation (5) for alternate control
groups, including the explicit interactions with the center-specific
displacement rate. The coefficient of the interaction between Ln Total
Autism and Target suggests that, among centers that do not experience
any displacement of MR for autism and for whom the increase in autism
diagnoses reflects a real increase in the demand for services, the
elasticity of provider supply with respect to autism cases is 0.09-0.13.
The latter effect magnitude is about 20% larger relative to the effects
that do not account for displacement (Model 6), consistent with the
25%-35% displacement rates noted in Table 2. It further adds to the
plausibility of this estimate that the coefficient of the triple
interaction is negative (-0.064 to -0.092), suggestive of a
dose-response relation--although they are imprecisely estimated. The
effect of autism diagnoses on the supply of auxiliary healthcare workers
is diminished with larger displacement rates; for those centers where
the increase in autism is crowding out MR diagnoses, this increase would
not be expected to raise substantially the demand for services of
workers in the target occupations. Models 4 and 8 explicitly control for
mild or moderate MR cases to isolate the full effect in a different way.
After accounting for MR cases, the estimates indicate a robust
10.4%-13.2% increase in auxiliary healthcare workers in the target
group, relative to the control group.
Table 6 presents estimates for the current and lagged effects on
annual wages, labor supply at the intensive margin as measured by annual
hours worked, and provider counts to shed more light on the adjustment
mechanisms in the labor market for these workers. We do not expect the
supply of new auxiliary healthcare workers to adjust substantially over
a short span of a few years given that the requirements for these
auxiliary healthcare occupations typically comprise a specialized
Master's degree and practical training spanning 2-3 years. Any
increase in supply in the medium term is likely to come from mobility of
providers from surrounding areas; this may include full relocation,
working in offices/hospitals in more than one area, and/or home visits
by auxiliary healthcare providers to surrounding areas experiencing an
increase in demand. It should be noted that the regional centers
servicing CA are fairly large and represent a collection of counties.
Hence, we do not expect strong cross-center mobility over the very short
term given that doing so would necessitate traveling through several
counties. Thus, for the very short term, provider supply is expected to
be fairly inelastic and thus any adjustment in the labor market for
these providers would occur through wages.
Estimates in Table 6 suggest that the largest contemporaneous
response to an increase in autism occurs through wages (Model I). There
is also a slight lagged effect on wages as well, as evidenced by the
larger 1-year lagged elasticity (0.103) relative to the contemporaneous
elasticity (0.082). Part of this increase in annual wages occurs through
an increase in labor supply at the intensive margin, that is an increase
in hours worked. This effect on labor supply is somewhat larger in the
following year. Models 3 and 4 suggest that a 100% increase in autism
cases leads to a 5.4% and 7.1% increase in hours worked in the current
year and the subsequent year, respectively. Thus, about two-thirds of
the increase in annual wages is due to an increase in work effort, and
the remainder representing an increase in the price of these services.
There is essentially no effect on provider supply in the very short
run, which is validating given that the supply of new providers to the
region is expected to be inelastic in the short term. There is a
positive and significant elasticity for provider supply (0.089) in the
following year, and this effect increases to 0.140 after 2 years
following the increase in autism diagnoses. The average center had about
25 autism-supporting auxiliary healthcare providers per 100,000
population in 2005. These estimates suggest that a 100% increase in
autism cases would increase provider supply in a region by about 2-4
extra providers (per 100,000 population) over the next 2 years. The
pattern of results in Table 6 is suggestive of an adjustment mechanism
favoring higher wages and labor supply at the intensive margin (hours
worked) in the short and medium term, and higher provider supply
(potentially through mobility from surrounding areas) over the medium to
longer term. (26)
VI. COMPETING HYPOTHESES AND ROBUSTNESS CHECKS
A potential caveat to our estimates is the effect of
"labeling" on the demand for services. An underlying
assumption of our identification strategy is that the price of an autism
hour of therapy is equivalent to that for a mentally retarded patient.
However, there may be differences in intensity of use between the two
diagnoses or differences in willingness to pay clue to the labeling of
autism instead of an alternative mental disorder. If this is indeed the
case, then we should expect a significant difference in the elasticity
of wages for autism-related occupations with respect to a 1% increase in
either autism or mild MR cases. We find the elasticities of 0.20 for
autism and 0.10 for mild-to-moderate MR cases, but we cannot reject the
null hypothesis of equal elasticities. While this suggests somewhat
higher intensity of use among autism cases than mild-to-moderate MR,
these effects are not sufficiently large to be the primary driver of our
results.
Next, we consider sources of external validity to the
"labeling" hypothesis. The first external source is from the
2005-2006 and 2009-2010 waves of the National Survey of Children with
Special Health Care Needs (NS-CSHCN), a nationally representative study
of all non-institutionalized children with special health care needs
between the ages of 0 and 17 in the United States conducted by the
National Center for Health Statistics (NCHS). We calculate the
percentage of individuals with autism and mild or moderate MR that use
these auxiliary healthcare providers for each wave. In the 2005-2006
wave, 70.1% of autistic children and 71.4% of mild or moderate MR
children used these services. Likewise, in the 2009-2010 wave, 71.7% of
autistic children and 71.6% of mild or moderate MR children used these
services. The differences in percentages by disability are neither
significant within nor across waves.
The second external source is the Special Education Expenditure
Project report: Total Expenditures for Students with Disabilities,
1999-2000, by the U.S. Department of Education (Chambers, Shkolnik, and
Perez 2003). The report documents federal expenditure per pupil by
disability type. The average expenditure for MR is $15,040 and that for
autism is $18,790, but the 95% confidence intervals surrounding these
means overlap. Therefore, we cannot reject the null hypothesis of equal
mean expenditure per pupil for these two diagnoses at the 5% level of
significance. In this same report, expenditures are disaggregated by the
type of auxiliary healthcare provider. Again, we cannot reject the null
hypothesis of equal expenditures with respect to occupational therapists
and other related services (e.g., school psychologists, social workers,
school nurses, audiologists, vision specialists, other therapists, and
personal health aides). However, there is a significant difference
associated with speech-language pathologists. We have estimated the
primary model dropping speech pathologists and find that the effects are
fully robust both in terms of magnitude and significance to excluding
speech-language pathologists from the analyses. We report similar
results of excluding other auxiliary healthcare providers in Table 7.
Along these lines, we may be concerned that although these
auxiliary healthcare workers cannot induce their own demand they may
work for physicians or units that can. (27) Based on the BLS
Occupational Outlook Handbook (2012-2013 edition), about 37% of physical
therapists, 29% of recreational therapists, 27% of occupational
therapists, 17% of recreational therapists, and 13% of speech-language
pathologists worked in hospitals or large medical practices. The
remainder worked in offices and clinics of physical, occupational and
speech therapists and audiologists, home healthcare services, nursing
care facilities, schools, community care facilities, and individual and
family services. Hence, while spillovers from induced demand by primary
providers to auxiliary healthcare providers is a possibility, the
majority of these auxiliary healthcare providers do not work in
hospitals or medical practices where one might expect such spillovers to
be the strongest.
The fraction working in hospitals and medical practices is highest
for respiratory therapists (81%). We therefore assess whether our
estimates are driven by potential spillovers in induced demand for these
auxiliary healthcare workers who are most likely to work in medical
practices/hospital settings. Table 7 reports estimates excluding
respiratory therapists and physical therapists in turn, who are far more
likely to work in medical facilities that may also diagnose autism. The
estimate magnitudes are not substantially affected by this exclusion;
the wage elasticities range from 0.08 to 0.13 compared with 0.07 to 0.11
for the full sample. (28)
One concern, which cannot be directly tested, is that there may be
residual confounding trends that differentially affect the target and
control groups leading to biased DD estimates. Table 8 indirectly
assesses this possibility by implementing a placebo test based on a
pseudo target group comprising healthcare occupations whose services
(and hence wages or provider supply) should not be directly affected by
an increase in autism cases. In order to define this pseudo target
group, we chose the first seven occupations within the BLS
classification of healthcare practitioner and technical occupations that
service human patients and are not complementary to autism diagnoses.
Specifically, the pseudo target group comprises the following seven
occupations in sequence in the BLS classification: chiropractors,
dentists, dieticians and nutritionists, optometrists, pharmacists,
radiation therapists, and dental hygienists. The control group
represents all other healthcare practice occupations (excluding the
autism-related providers). If our DD estimates are effectively purging
residual center-specific trends, then we would not expect an increase in
autism cases to affect the wages or supply of providers in these pseudo
occupational categories (relative to the control group). Indeed, Table 8
confirms that there are no significant or substantial effects on either
wages or provider supply for occupations in this pseudo target group.
This falsification check is further validating in that it also adds a
degree of confidence to the use of our control groups as a
counterfactual for the target autism-related occupations.
Next, the identification strategy relies on year-to-year variation
in growth rates to estimate the parameters. If the year-to-year growth
rates are highly correlated, then we cannot treat each observation as
independent. We address this issue in two ways. First, we have adjusted
all reported standard errors to account for arbitrary correlation across
years within each center. If the year-to-year fluctuations in growth
rates are mostly noise, then this would reduce the precision of our
estimates. Second, we re-estimate the models only considering 2004 and
2010 and exploiting the long-span variation in autism growth rates
(Figures 3 and 4). This step removes concerns of arbitrary correlations
across years driving the significant results although it somewhat
weakens the statistical power of our estimates as we have fewer
observations. Although the estimates are less precise, we still find
wage effects between 6.6% and 12% without displacement, and 9% and 14%
with displacement where most estimates are significant at the 10% level
or better.
Lastly, the estimates may potentially suffer from cross-county
border effects. As illustrated in Figure 1, each regional center is
responsible for several counties, thereby, requiring auxiliary
healthcare workers to travel through potentially several counties before
affecting wages in the service area of a different regional center. The
primary exception is Los Angeles County. Table 9 presents estimates for
a sample of 15 regional centers where the seven centers that service Los
Angeles County are aggregated together. We also estimate all models for
annual personal earnings as reported in the ACS (not shown). All of
these estimates remain robust and highly similar, both in terms of
magnitudes and statistical significance, to those discussed above. (29)
VII. DISCUSSION
As the prevalence of autism has expanded dramatically over the past
three decades, a central debate relates to the various factors that
could potentially explain this increase. This study directly assesses
whether the increase in autism diagnoses in regional developmental
disability service centers in California is associated with a
displacement of MR diagnoses. We also test this proposition indirectly
by examining the impact of the higher number of autism cases on the
demand for auxiliary healthcare workers--occupations whose services are
complementary to the diagnoses of autism--within a DD multivariate
regression framework.
If the incidence of autism is increasing independently of other
mental disorders, then the demand for auxiliary health providers (i.e.,
speech pathologist, behavioral therapist, occupational therapist, etc.)
should also increase, leading to higher wages for these providers in the
short term and possibly a higher supply of these providers over time.
If, however, the increase in autism diagnosis is merely displacing other
mental disorders, then the effects of the increase on demand will be
moderated or not present.
We find robust evidence that the higher prevalence of autism in
California has raised wages and earnings among those target occupations
whose services would be potentially impacted, conditional on confounding
trends. The elasticity of wages and earnings with respect to autism
diagnoses controlling for displacement effects is estimated to be
between 0.13 and 0.16. About two-thirds of the resulting increase in
wages is due to an increase in work effort as measured by annual hours
worked.
The wage elasticity diminishes in magnitude with a higher
displacement rate of MR for autism diagnoses, consistent with mitigated
demand. We find that the average displacement rate is on the order of
one-third, suggesting that one of every three new autism diagnoses is
merely supplanting MR diagnoses and does not represent a true increase
in autistic disorders. The estimated displacement effect is similar in
magnitude to those found using administrated data of one-fourth for
autism versus MR (King and Bearman 2009) and one-third (Coo et al. 2008)
for autism versus non-autism mental disorders. The total number of
autism diagnoses in California increased by 86%. This suggests a 57.3%
increase (assuming a displacement of one-third) in effectively
"new" autism diagnoses. Combined with the above elasticity
estimates, this increase in prevalence would raise the wages and
earnings among auxiliary healthcare providers who provide behavioral
intervention services for children with autistic disorders by 7.4%-9.2%.
Actual wages for these workers increased by 40% (as shown in Table 1)
over the sample period. Our estimates therefore suggest that about
19%-23% of this observed increase (7.4-9.2 percentage points out of the
40% observed increase in wages) was due to increased demand stemming
from the increase in autism cases. The remainder of the increase is then
due to general economic conditions and factors affecting all workers in
the healthcare practice and technical occupations.
The short span of the data set precludes a comprehensive analysis
of the effects on the number of providers in the long run. Nonetheless,
we find suggestive evidence that expanded prevalence of autistic
disorders does raise the number of auxiliary healthcare workers over the
subsequent 2 years, with elasticity estimates ranging from 0.09 to 0.14.
(30) As with wages, we would expect this effect to be mitigated if the
increase in autism is substituting for MR cases, which is indeed what
the estimates suggest.
These results further confirm that provider income and their hours
worked are responsive to an increase in the demand for their services.
We note that these wage effects are not reflective of a direct
"supplier-induced demand" as these healthcare workers do not
themselves diagnose autism and thus cannot induce their own demand.
Thus, at least part of the increase in the autism caseload represents an
effective increase in their demand given that we observe an increase in
their wages. This further suggests that at least part of the increase in
autism diagnoses, about one-half to two-thirds based on the direct and
indirect estimates of displacement, reflects an increase in the true
prevalence of the disorder.
ABBREVIATIONS
ABA: Applied Behavioral Analysis
ACS: American Community Survey
ADD: Attention Deficit Disorder
ADHD: Attention Deficit Hyperactivity Disorder
BLS: Bureau of Labor Statistics
DD: Difference-In-Differences
DDD: Difference-In-Difference-In-Differences
DDS: Department of Developmental Services
MR: Mental Retardation
NCHS: National Center for Health Statistics
NIMH: National Institute of Mental Health
NS-CSHCN: National Survey of Children with Special Health Care
Needs
OLS: Ordinary Least Squares
PDD-NOS: Pervasive Developmental Delay-Not Otherwise Specified
PUMA: Public Use Microdata Area
PUMS: Public Use Microdata Sample
doi: 10.1111/ecin.12137
Online Early publication August 25, 2014
REFERENCES
American Psychiatric Association. Diagnostic and Statistical Manual
of Mental Disorders. 4th ed. Washington, DC: APA, 1994.
--. Diagnostic and Statistical Manual of Mental Disorders
DSM-IV-TR. 1st ed. Washington, DC: APA, 2000.
ASD Best Practice Guidelines. "California Department of
Developmental Services. California Health and Human Services
Agency." 2002. Accessed October 4, 2010. www.dds.ca.gov (pdf).
Baio, J. "Prevalence of Autism Spectrum Disorder Among
Children Aged 8 Years--Autism and Developmental Disabilities Monitoring
Network. 11 Sites, United States, 2010." CDC-Morbidity and
Mortality Weekly Report Surveillance Summaries March 28. 2014, 631SS02);
1-21.
Bhasin, T. K., S. Brocksen, R. N. Avchen, and B. K. Van Naarden.
"Prevalence of Four Developmental Disabilities Among Children Aged
8 Years--Metropolitan Atlanta Developmental Disabilities Surveillance
Program, 1996 and 2000." MMWR Surveillance Summaries, 55, 2006,
1-9.
Bishop, D. V., A. J. Whitehouse, H. J. Watt, and E. A. Line.
"Autism and Diagnostic Substitution: Evidence from a Study of
Adults with a History of Developmental Language Disorder."
Developmental Medicine and Child Neurology, 50. 2008, 341 -45.
Bleuler, E. The Theory of Schizophrenic Negativism (W. A. White,
trans.). New York: Journal of Nervous and Mental Disease Publishing
Company. (Reprinted 1970 by New York: Johnson Reprint Corporation)
(Original work published 1910/1911). 1912.
Blumberg, S. J.. M. D. Bramlett. M. D. Kogan, L. A. Schieve, J. R.
Jones, and M. C. Lu, "Changes in Prevalence of Parent-reported
Autism Spectrum Disorder in School-aged U.S. Children: 2007 to
2011-2012." National Health Statistics Reports No. 65, March 20,
2013.
Cameron, C., and P. Trivedi. Regression Analysis of Count Data.
Cambridge: Cambridge University Press, 1998.
Cavagnaro, A. T. "Autistic Spectrum Disorders: Changes in the
California Caseload: An Update: 1987-2007." Sacramento, CA:
California Department of Developmental Services. California Health and
Human Services Agency, 2007. Accessed October 4, 2010. www.dds.ca.gov
(pdf).
Chakrabarti, S., and E. Fombonne. "Pervasive Developmental
Disorders in Preschool Children." JAMA, 285. 2001, 3093-99.
Chambers, J., J. Shkolnik, and M. Perez. "Total Expenditures
for Students with Disabilities, 1999-2000: Spending Variation by
Disability." Palo Alto, CA: American Institutes for Research,
Center for Special Education Finance, Report 5, 2003
Committee on Children with Disabilities. "The
Pediatrician's Role in the Diagnosis and Management of Autistic
Spectrum Disorder in Children." Pediatrics, 107(5), 2001, 1221-26.
Coo, H., H. Ouellette-Kuntz, J. E. V. Lloyd, L. Kasmara, J. J. A.
Holden, and M. E. S. Lewis. "Trends in Autism Prevalence:
Diagnostic Substitution Revisited." Journal of Autism and
Developmental Disorders, 38, 2008, 1036-46.
Daily, D., H. Ardinger, and G. Holmes. "Identification and
Evaluation of Mental Retardation." American Family Physician,
61(4), 2000, 1059-67.
Elder, T. "The Importance of Relative Standards in ADHD
Diagnoses: Evidence Based on Exact Birth Dates." Journal of Health
Economics, 29(5), 2010, 641-56.
Eldevik, S., S. Eikeseth, E. Jahr, and T. Smith. "Effects of
Low-Intensity Behavioral Treatment for Children with Autism and Mental
Retardation." Journal of Autism and Developmental Disorders, 36(2),
2006, 211-24.
Fuchs, V. R. "The Supply of Surgeons and the Demand for
Operations." The Economics of Physician and Patient Behavior, XIII,
1978, 35-56. (Supplement to Journal of Human Resources, edited by Victor
R. Fuchs and Joseph P. Newhouse.)
Grinker, R. R. Unstrange Minds. New York: Basic Books, 2007.
Guevara, J. P., D. S. Mandell, A. L. Rostain, H. Zhao, and T. R.
Hadley. "National Estimates of Health Services Expenditures for
Children with Behavioral Disorders: An Analysis of the Medical
Expenditure Panel Survey." Pediatrics, 112, 2003, e440.
Jacobson, J. W., and J. A. Mulick. "System and Cost Research
Issues in Treatments for People with Autistic Disorders." Journal
of Autism and Developmental Disorders, 30, 2000, 585-93.
King, M., and P. Bearman. "Diagnostic Change and the Increased
Prevalence of Autism." International Journal of Epidemiology, 38,
2009, 1224-34.
Liu, K., N. Zerubavel, and P. Bearman. "Social Demographic
Change and Autism." Demography, 47(2), 2010, 327-43.
Mandell, D. S., J. Listerud, S. E. Levy, and J. A. Pinto-Martin.
"Race Differences in the Age at Diagnosis among Medicaid-Eligible
Children with Autism." Journal of the American Academy of Child and
Adolescent Psychiatry, 41, 2002, 1447-53.
Mandell, D. S.. L. D. Wiggins, L. A. Carpenter, J. Daniels, C.
DiGuiseppi, M. S. Durkin, E. Giarelli, M. J. Morrier, J. S. Nicholas, J.
A. Pinto-Martin, P. T. Shattuck, K. C. Thomas, M. Yeargin-Allsopp, and
R. S. Kirby. "Racial/Ethnic Disparities in the Identification of
Children with Autism Spectrum Disorders." American Journal of
Public Health, 99(3), 2009, 493-98.
Myers, S., and C. Johnson. "Management of Children with Autism
Spectrum Disorders." Pediatrics, 120, 2007, 1162.
Nassar, N" G. Dixon, J. Bourke, C. Bower, E. Glasson, N.
deKlerk, and H. Leonard. "Autism Spectrum Disorders in Young
Children: Effect of Changes in Diagnostic Practices." International
Journal of Epidemiology, 38(5), 2009. 1245-54.
Shattuck, P. T. "The Contribution of Diagnostic Substitution
to the Growing Administrative Prevalence of Autism in U.S. Special
Education." Pediatrics, 117, 2006, 1028-37.
Smith, D" B. A. Defallal, and D. W. Chadwick. "The
Misdiagnosis of Epilepsy and the Management of Refractory Epilepsy in a
Specialist Clinic." Quarterly Journal of Medicine, 92, 1999, 15-23.
Tomanik, S., D. A. Pearson, K. A. Loveland, D. M. Lane, and J.
Bryant Shaw. "Improving the Reliability of Autism Diagnoses:
Examining the Utility of Adaptive Behavior." Journal of Autism and
Developmental Disorders, 37,2007, 921-28.
Uldall, P., J. Alving, L. K. Hansen, M. Kibtek, and J. Buchholt.
"The Misdiagnosis of Epilepsy in Children Admitted to a Tertiary
Epilepsy Centre with Paroxysmal Events." Archives of Disease in
Childhood, 91, 2006, 219-21.
Waldman, M., S. Nicholson, N. Adilov, and J. Williams. "Autism
Prevalence and Precipitation Rates in California, Oregon, and Washington
Counties." Archives of Pediatrics and Adolescent Medicine, 162(11),
2008. 1026-34.
Yeargin-Allsopp, M" C. Rice, T. Karapurkar, N. Doemberg, C.
Boyle, and C. Murphy. "Prevalence of Autism in a US Metropolitan
Area." JAMA, 289, 2003, 49-55.
(1.) Self-reported parent surveys place the prevalence rate at 2%
for school-age children in 2012 (Blumberget al. 2013).
(2.) Waldman et al. (2008) find an association between increases in
precipitation and increases in autism prevalence.
(3.) De novo mutations are deletions, insertions, and duplications
of DNA in the germ cells (sperm or egg) that are not present in the
parents' DNA.
(4.) See Fuchs (1978) for a theory of physician-induced demand.
(5.) Parents seek out some of these alternative health workers for
other symptoms such as delayed speech. The parents take the child to a
speech pathologist, and the speech pathologist may recommend that the
child be tested for autism. Thus, it might appear that the demand is
induced, but if anything this would bias the results towards zero
because the child would not yet be counted as an autism case. Therefore,
this would be an increase in demand for the services of a speech
pathologist in the absence of an increase in autism within the data.
(6.) Elasticity estimates are not sensitive to alternate functional
forms: (1) non-logged wages, (2) non-logged autism diagnoses, and (3)
rate of autism diagnoses relative to various population bases.
(7.) Mandell et al. (2002) show that among Medicaid-eligible
children. White children on average receive an autistic disorder
diagnosis at 6.3 years compared with 7.9 years for Black children.
(8.) Insurance status is unavailable in the ACS data prior to 2008.
We construct center-specific rates of public coverage, private coverage,
and uninsured for 2008-2011. Including these center-specific measures of
insurance status in models estimated over 2008-2011 does not materially
change the point estimates, although standard errors are inflated in
these models due to the reduced sample size.
(9.) See for instance: http://www.autism-society.org/
living-with-autism/treatment-options/ and http://www.
webmd.com/brain/autism/autism-treatment-overview. To see medical
treatments used by autistic parents, see
http://www.autism.com/pdf/providers/ParentRatings2009.pdf
(10.) Occupations in the ACS are identified by the 2002 Census
codes. We therefore include codes 3140 (audiologists), 3150
(occupational therapists), 3160 (physical therapists), 3210
(recreational therapists), 3220 (respiratory therapists), 3230
(speech-language pathologists), and 3240 (other therapists) in the
target group. Estimates are robust to excluding other therapists (code
3240).
(11.) (http://children.webmd.com/intellectual-disabilitymental-retardation?page=2
(12.) Specifically, we define control healthcare practitioner and
support occupations based on 2002 Census codes 3000 to 3650, excluding
those defined above for the target group.
(13.) In supplementary analyses, as a robustness check, we also
considered all other occupations (healthcare and non-healthcare) as a
global control group. Expectedly, these estimates were of a larger
magnitude as healthcare occupations (including the auxiliary service
providers for autistic children) generally enjoy stronger wage growth
relative to other jobs, and effects are likely overstated, partly
reflecting just an overall positive trend in the wages of healthcare
occupations.
(14.) Estimates are not sensitive to alternate functions forms: (1)
percent of total cases and (2) log of total cases.
(15.) This becomes a difference-in-difference-indifferences (DDD)
model, which we estimate as a fully flexible specification by also
allowing interactions between Displacement and Ln Autism, and between
Displacement and Target.
(16.) Specifically, we report cluster bootstrapped standard errors
based on 200 replications at the center-year cluster level. Center-level
clusters provide too few clusters for the bootstrap. We note, however,
that non-bootstrapped standard errors clustered at the center-level or
at the center-year level in our DD models yield highly similar variance
estimates and inferences.
(17.) These reports are available from the California DDS at
http://www.dds.ca.gov/FactsStats/Diagnostic_Main.cfm.
(18.) The NIMH "A Parent's Guide to Autism Spectrum
Disorder" http://www.nimh.nih.gov/health/publications/a-parents-guide-to-autism-spectrum-disorder/parent-guide-to- autism.pdf and the NIMH
"Mental Retardation: A Manual for Psychologists"
http://www.nimhindia.org/ A%20Manual%20for%20Psychologists.pdf.
(19.) Eldevik et al. (2006) demonstrates the effectiveness of ABA
therapy to improve intellectual functionality among autistic and
mentally retarded children.
(20.) Medscape.com also provides a list of pro posed treatments for
autism and mental retardation. See
(http://emedicine.medscape.eom/article/912781-treatment# showall) for
autism treatments and (http://emedicine.
medscape.com/article/1180709-treatment#showall)
(21.) There are 233 PUMAs defined for California. See the Census
ACS webpage (http://www.census.gov/acs/www/) for information on these
geographic areas and the sampling schemes of the ACS and the ACS-PUMS.
(22.) There are seven regional centers that service Los Angeles
County. Aggregating these centers into a single regional center
servicing LA does not alter results (reported in Table 9).
(23.) For instance, a therapist may work for an employer (school,
hospital, or practice) and supplement their income by billing via a sole
proprietorship.
(24.) We compute annual hours worked as the product of number of
weeks worked and usual hours worked per week.
(25.) While the population-weighted mean found in Table 2 from the
individual center displacement rates is about 0.3-0.4, there is a wide
range across centers suggesting that some regional centers had
essentially no displacement (e.g., Central Valley, Golden Gate, Valley
Mountain, and Tri Counties) and others had very high or close to full
displacement (Far Northern, Inland, Kern, Orange County, and Kern). Some
of these differences may be due to variation in provider practice and
diagnosis criteria as actually applied, public and private coverage, and
other region-specific observable and unobservable factors. It is
validating that the displacement rates of 0.3-0.4 suggested by the
pooled regression model (reported in Table 2) accord with the average
displacement rate from the individual center regressions (estimated from
Equation (4)).
(26.) There could also be a response to an increase in the demand
for the services of these healthcare workers through mobility at the
level of the parents/patient. That is, parents of autistic children may
seek out auxiliary healthcare workers in surrounding areas if they are
unable to obtain timely services in their own area due to an increase in
demand. This points to some spatial spillovers such that an increase in
autism diagnoses may spill over to higher demand for auxiliary workers
and their wages in surrounding areas.
We note that the regional centers servicing CA are fairly large,
with an average population of 2.7 million (2.1 million, excluding LA).
Hence, our estimates already capture any mobility of patients and
providers across counties within each center.
(27.) We thank an anonymous reviewer for raising this point
regarding spillover-induced demand.
(28.) If respiratory therapists and physical therapists are
excluded jointly then the elasticity range is 0.064-0.122 and remains
significant at the 10% level in all but one specification.
(29.) We also use the number of autism cases in surrounding
regional centers as a method to control for spatial effects and find
similar results. Additionally, the number of autism cases in surrounding
regional centers is not found to have a statistically significant effect
on own wage elasticity at conventional levels.
(30.) Based on our estimates, we can back-out an implied labor
supply-wage elasticity for auxiliary healthcare workers as the ratio of
the labor-supply/autism elasticity to the wage/autism elasticity. We
used two measures of labor supply capturing the intensive and extensive
margins, respectively: (1) annual hours worked (contemporaneous) and (2)
provider counts (1-year lag). Our estimates imply a labor supply-wage
elasticity for auxiliary healthcare workers of about 0.55-0.70 at these
margins. To place these estimate in context, the labor supply elasticity
with respect to wages has been estimated to be 0.33 for physicians and
0.61 for solo proprietor physicians in the literature (e.g., Showalter
and Thurston 1997).
DHAVAL M. DAVE and JOSE M. FERNANDEZ *
* The authors are grateful to Pinka Chatterji, Robin McKnight,
Valerie Rodriguez, Joseph Sabia, two anonymous reviewers, and session
participants at the 2012 conferences of the Southern Economic
Association and the American Society of Health Economists for helpful
comments and feedback on earlier versions of this study.
Dave: Department of Economics, Bentley University & National
Bureau of Economic Research (NBER), Waltham, MA 02452-4705. Phone
781-891-2268, Fax 917-426-7015, E-mail ddave@bentley.edu
Fernandez: Department of Economics, College of Business, University
of Louisville, Louisville, KY 40292. Phone 502-852-4861, Fax
502-852-7672, E-mail jose.fernandez@louisville.edu
TABLE 1
Descriptive Statistics: 21 California Regional Centers
Time Period 2002 2004 2010
Total clients 7,799.6 8,403.1 10,123.8
Total autism cases 970.4 1,265.4 2,359.9
Total mild MR cases 3,067.0 3,290.4 3,694.2
Total mild or moderate MR cases 4,499.4 4,783.8 5,276.9
% Clients White 0.4603 0.4527 0.4082
% Clients Black 0.1029 0.1043 0.0986
% Clients Hispanic 0.2805 0.2951 0.3263
Total population 2,595,893 2,669.724
% Population Black 0.0607 0.0620
% Population Other Race
(non-White & non-Black) 0.1479 0.1726
% Population Hispanic 0.2986 0.3294
% Population ages
[less than or equal to] 13 0.1944 0.1813
Annual wages (autism occupations) 42,776.4 59,897.4
Annual earnings
(autism occupations) 49,504.5 62,893.6
Annual wages (other healthcare
practitioner/technical
occupations) 56,737.9 71,292.7
Annual earnings (other healthcare
practitioner/technical
occupations) 66,909.8 77,847.9
Annual wages (all non-autism
healthcare occupations) 44,103.3 54,619.9
Annual earnings (all non-autism
healthcare occupations) 51,296.4 59,284.3
Annual wages (non-healthcare
occupations) 34,208.8 35,242.6
Annual earnings (non-healthcare
occupations) 38,259.7 38,282.7
Time Period 2011 2004-2010
Total clients 10,376.1 9,228.8
Total autism cases 2,566.0 1,786.0
Total mild MR cases 3,740.3 3,479.1
Total mild or moderate MR cases 5,370.2 5,017.9
% Clients White 0.3998 0.4306
% Clients Black 0.0976 0.1012
% Clients Hispanic 0.3322 0.3103
Total population 2,759,920 2,587,315
% Population Black 0.0627 0.0605
% Population Other Race
(non-White & non-Black) 0.1755 0.1534
% Population Hispanic 0.3353 0.3097
% Population ages
[less than or equal to] 13 0.1800 0.1901
Annual wages (autism occupations) 49,775.7
Annual earnings
(autism occupations) 55,032.3
Annual wages (other healthcare
practitioner/technical
occupations) 65,408.9
Annual earnings (other healthcare
practitioner/technical
occupations) 73,192.1
Annual wages (all non-autism
healthcare occupations) 50,517.5
Annual earnings (all non-autism
healthcare occupations) 56,097.7
Annual wages (non-healthcare
occupations) 35,506.1
Annual earnings (non-healthcare
occupations) 39,160.1
Notes: Means are reported across 21 regional centers in
California for the noted periods. Information on average wages
and earnings are from the American Community Surveys (2005-2011)
referring to years 2004-2010; means are weighted by occupation
counts. Information on population demographics are from the U.S.
Census Bureau (2005-2011).
TABLE 2
Displacement of Mental Retardation (MR) Diagnoses for Autism
Diagnoses: 21 California Regional Centers 2002-2011
Model (1) (2) (3)
Mild MR
Outcome (% Clients)
Autism -0.40264 ** -0.40247 ** -0.33529 ***
(% of clients) (0.14454) (0.14712) (0.08693)
Total autism cases -- -- --
Controls for trends Linear + Year Year
Quadratic indicators indicators
Center indicators and No No Yes
client demographics
[R.sup.2] 0.23625 0.23660 0.99173
Observations 210 210 210
Model (4) (5) (6)
Mild or
Moderate Total Mild
Outcome (% Clients) MR Cases
Autism -0.38060 ** -- --
(% of clients) (0.14167)
Total autism cases -- -0.26503 ** -0.26488 **
(0.10402) (0.10581)
Controls for trends Year Linear + Year
indicators Quadratic indicators
Center indicators and Yes No No
client demographics
[R.sup.2] 0.98628 0.89712 0.89717
Observations 210 210 210
Model (7) (8)
Total Mild
Total Mild or Moderate
Outcome MR Cases MR Cases
Autism -- --
(% of clients)
Total autism cases -0.25238 ** -0.18031
(0.09653) (0.16275)
Controls for trends Year Year
indicators indicators
Center indicators and Yes Yes
client demographics
[R.sup.2] 0.99809 0.99815
Observations 210 210
Model (9) (10)
Total Total
Cerebral Palsy Epilepsy
Outcome Cases Cases
Autism -- --
(% of clients)
Total autism cases -0.04404 -0.08555
(0.06559) (0.05369)
Controls for trends Year Year
indicators indicators
Center indicators and Yes Yes
client demographics
[R.sup.2] 0.99930 0.99922
Observations 126 126
Notes: Coefficient estimates from OLS models are presented.
Standard errors are adjusted for arbitrary correlation within
center cells. Models 9 and 10 are based on data from 2002 to
2007. All models also control for the total number of clients
served at the center.
Asterisks denote statistical significance as follows: *** p [less
than or equal to]. 01; **.01 < [less than or equal to] .05; *.05
< p [less than or equal to] .10.
TABLE 3
Ln Annual Wages Difference-in-Differences (DD) Estimates American
Community Surveys 2005-2011: 21 California Regional Centers
Model (1) (2)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism 0.04962 0.12562
(0.03458) (0.14506)
Ln total autism * 0.10784# *** 0.11042# ***
Target (0.03753)# (0.03721)#
Target -0.78875 ** -0.81160 ***
(autism occupations) (0.27905) (0.27696)
Year indicators Yes Yes
Center indicators No Yes
Area demographics No No
Center demographics No No
Occupation indicators No No
[R.sup.2] 0.90397 0.90506
Observations 3,630 3,630
Model (3) (4)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism -0.05549 0.01649
(0.25729) (0.26153)
Ln total autism * 0.10807# *** 0.08162# ***
Target (0.03649)# (0.02840)#
Target -0.79400 *** --
(autism occupations) (0.27137)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators No Yes
[R.sup.2] 0.90531 0.93844
Observations 3,630 3,630
Model (5) (6)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism 0.07283 * 0.03893
(0.03686) (0.12864)
Ln total autism * 0.09801# ** 0.10188# ***
Target (0.03548)# (0.03568)#
Target -0.50971 * -0.54210 *
(autism occupations) (0.26515) (0.26722)
Year indicators Yes Yes
Center indicators No Yes
Area demographics No No
Center demographics No No
Occupation indicators No No
[R.sup.2] 0.87469 0.87596
Observations 4,535 4,535
Model (7) (8)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism -0.20856 -0.14517
(0.16589) (0.17974)
Ln total autism * 0.10019# *** 0.07132# **
Target (0.03480)# (0.02814)#
Target -0.52894 * --
(autism occupations) (0.26036)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators No Yes
[R.sup.2] 0.87623 0.93541
Observations 4,535 4,535
Notes: Coefficient estimates from OLS models are presented.
Standard errors are adjusted for arbitrary correlation within
center cells. Area demographics include total county population:
percent of the population that is Black; percent other race:
percent Hispanic; and percent ages <13, 15-44, 45-64, >65 years.
Center demographics include total number of clients served,
percent of the total clients who are White, percent Black, and
percent Hispanic. Estimated DD effects are presented in bold.
Asterisks denote statistical significance as follows: *** p [less
than or equal to] .01; **.01 <p [less than or equal to] .05; *.05
< [less than or equal to] .10.
Note: Estimated DD effects are presented in bold is indicated with
#.
TABLE 4
Ln Annual Wages DDD Estimates Differential Effects by
Displacement American Community Surveys 2005-2011:21 California
Regional Centers
(1) (2)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism 0.12771 -0.06015
(0.17179) (0.41487)
Ln total autism x Target 0.17044# *** 0.16778# ***
(0.06150)# (0.06096)#
Ln total autism x Target x -0.08591# *** -0.08567# ***
Displacement (0.03258)# (0.03236)#
Target -1.24608 *** -1.22627 ***
(autism occupations) (0.45757) (0.45378)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics No Yes
Center demographics No Yes
Occupation indicators No No
Mental retardation (MR) and No No
mild/moderate MR x
Target
[R.sup.2] 0.90522 0.90546
Observations 3,630 3,630
(3) (4)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism 0.03039 -0.06241
(0.40254) (0.26571)
Ln total autism x Target 0.12713# ** 0.16099# ***
(0.05937)# (0.05558)#
Ln total autism x Target x -0.06401# * -
Displacement (0.03270)#
Target -1.37265 *** -0.21048
(autism occupations) (0.41622) (0.31432)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes No
Mental retardation (MR) and No Yes
mild/moderate MR x
Target
[R.sup.2] 0.93853 0.90537
Observations 3,630 3,630
(5) (6)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism 0.04617 -0.21192
(0.14978) (0.31590)
Ln total autism x Target 0.16648 ** 0.16419 **
(0.06703)# (0.06683)#
Ln total autism x Target x -0.09206# *** -0.09148# ***
Displacement (0.03302)# (0.03279)#
Target -1.00918 ** -0.99167 **
(autism occupations) (0.49954) (0.49817)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics No Yes
Center demographics No Yes
Occupation indicators No No
Mental retardation (MR) and No No
mild/moderate MR x
Target
[R.sup.2] 0.87612 0.87638
Observations 4,535 4,535
(7) (8)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism -0.13880 -0.22081
(0.31789) (0.17001)
Ln total autism x Target 0.11747# * 0.15181# **
(0.06491)# (0.05616)#
Ln total autism x Target x -0.06472# ** _
Displacement (0.03232)#
Target -0.59070 0.03711
(autism occupations) (0.45626) (0.32967)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes No
Mental retardation (MR) and No Yes
mild/moderate MR x
Target
[R.sup.2] 0.93549 0.87628
Observations 4,535 4,535
Notes: Coefficient estimates from OLS models are presented. For
Models 1 -3 and 5-7, bootstrapped standard errors clustered
within center- year cells based on 200 replications are reported.
For Models 4 and 8, standard errors are adjusted for arbitrary
correlation within center cells. Area demographics include total
county population; percent of the population that is Black;
percent other race; percent Hispanic; and percent ages < 13,
15-44, 45-64, >65 years. Center demographics include total number
of clients served, percent of the total clients who are White,
percent Black, and percent Hispanic. In Models 1 -3 and 5-7,
displacement represents the percent displacement of mild MR cases
for autism diagnoses, obtained from estimating Model 5 in Table 1
separately for each center and controlling for center
demographics. These models also control for interactions between
displacement and Ln total autism, and displacement and autism
occupations. Estimated DD and DDD effects are presented in bold.
Asterisks denote statistical significance as follows: *** p [less
than or equal to] .01; ** .01 <p [less than or equal to] .05;
*.05 < p [less than or equal to] 10.
Note: Estimated DD and DDD effects are presented in bold is
indicated with #.
TABLE 5
Provider Counts (1-Year Lag) Poisson Regression American
Community Surveys 2005-2011:21 California Regional Centers
Model (1) (2)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism -0.08940 -0.08896
(0.11183) (0.10020)
Ln total autism x Target 0.04825 0.08903 ***
(0.03183) (0.03316)
Ln total autism x Target x -- --
Displacement
Target -1.54562 *** --
(autism occupations) (0.24036)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics No Yes
Center demographics No Yes
Occupation indicators No Yes
Mental retardation (MR) and No No
Mild/moderate MR x
Target
Observations 4,147 4,147
Model (3) (4)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism -0.15050 -0.08831
(0.21288) (0.09659)
Ln total autism x Target 0.09257 ** 0.10418 **
(0.03874) (0.04708)
Ln total autism x Target x -0.06380 --
Displacement (0.07205)
Target -- --
(autism occupations)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes Yes
Mental retardation (MR) and No Yes
Mild/moderate MR x
Target
Observations 4,147 4,147
Model (5) (6)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism -0.19599 * -0.17378 **
(0.10337) (0.08434)
Ln total autism x Target 0.11675 *** 0.10864 ***
(0.04206) (0.03999)
Ln total autism x Target x -- --
Displacement
Target -2.18229 *** --
(autism occupations) (0.31672)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics No Yes
Center demographics No Yes
Occupation indicators No Yes
Mental retardation (MR) and No No
Mild/moderate MR x
Target
Observations 5,233 5,233
Model (7) (8)
Non-Autism All
Control Group Healthcare Occupations
Ln total autism -0.19718 -0.17778 **
(0.17682) (0.08799)
Ln total autism x Target 0.12555 *** 0.13226 **
(0.03624) (0.05692)
Ln total autism x Target x -0.09171 --
Displacement (0.06746)
Target -- --
(autism occupations)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes Yes
Mental retardation (MR) and No Yes
Mild/moderate MR x
Target
Observations 5,233 5,233
Notes: See Tables 3 and 4.
TABLE 6
Adjustment of Annual Wages, Hours Worked, Provider Counts
American Community Surveys 2005-2011: 21 California Regional
Centers
Model (1) (2) (3)
Control Group Non-Autism Healthcare Practitioner
& Technical Occupations
Ln Annual Ln Annual Ln Annual
Outcome Wages Wages Hours
Estimation OLS
Effect Current 1-year lag Current
Ln total autism x Target 0.08162 *** 0.10279 *** 0.05356 ***
(0.02840) (0.03218) (0.01814)
Year indicators Yes Yes Yes
Center indicators Yes Yes Yes
Area demographics Yes Yes Yes
Center demographics Yes Yes Yes
Occupation indicators Yes Yes Yes
[R.sup.2] 0.93844 0.93533 0.94918
Observations 3,630 3,123 3,590
Model (4) (5)
Control Group Non-Autism Healthcare Practitioner
& Technical Occupations
Ln Annual Provider
Outcome Hours Count
Estimation OLS Poisson
Effect 1-year lag Current
Ln total autism x Target 0.07070 ** 0.00843
(0.02614) (0.02784)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes Yes
[R.sup.2] 0.94153 --
Observations 3,086 4,142
Model (6) (7)
Control Group Non-Autism Healthcare Practitioner
& Technical Occupations
Provider Provider
Outcome Count Count
Estimation Poisson
Effect 1-year lag 2-year lag
Ln total autism x Target 0.08903 *** 0.13983 **
(0.03316) (0.05794)
Year indicators Yes Yes
Center indicators Yes Yes
Area demographics Yes Yes
Center demographics Yes Yes
Occupation indicators Yes Yes
[R.sup.2] -- --
Observations 4,147 3,572
Notes: See Tables 3 and 4.
TABLE 7
Ln Annual Wages: Sensitivity of DD Estimates to Excluding Each
Auxiliary Healthcare Occupation in Turn American Community
Surveys 2005-2011
Model (1) (2) (3)
Occupation Occupational Physical
Excluded Audiologists Therapists Therapists
Control Group Non-Autism Healthcare Practitioner and
Technical Occupations
Ln total autism x Target 0.11677 *** 0.09455 ** 0.09438 **
(0.03955) (0.03480) (0.03767)
[R.sup.2] 0.90539 0.90622 0.90502
Observations 3,575 3.504 3.483
Control Group Non-Autism All Healthcare Occupations
Ln total autism x Target 0.10738 ** 0.08693 ** 0.08531 **
(0.03962) (0.03450) (0.03775)
[R.sup.2] 0.87576 0.87672 0.87536
Observations 4,480 4.409 4,388
Model (4) (5)
Occupation Recreational Respiratory
Excluded Therapists Therapists
Control Group Non-Autism Healthcare Practitioner
and Technical Occupations
Ln total autism x Target 0.08359 ** 0.13066 ***
(0.03423) (0.04304)
[R.sup.2] 0.90753 0.90526
Observations 3,570 3,489
Control Group Non-Autism All Healthcare Occupations
Ln total autism x Target 0.07610 ** 0.12286 ***
(0.03004) (0.04137)
[R.sup.2] 0.87753 0.87568
Observations 4.475 4.394
Model (6) (7)
Speech-
Occupation Language All Other
Excluded Pathologists Therapists
Control Group Non-Autism Healthcare Practitioner
and Technical Occupations
Ln total autism x Target 0.13028 *** 0.10611 **
(0.04001) (0.04070)
[R.sup.2] 0.90606 0.90730
Observations 3,488 3,531
Control Group Non-Autism All Healthcare Occupations
Ln total autism x Target 0.12309 *** 0.09889 **
(0.03810) (0.03706)
[R.sup.2] 0.87656 0.87770
Observations 4,393 4,436
Notes: See Tables 3 and 4.
TABLE 8
Ln Annual Wages and Provider Count Falsification: Pseudo
Occupations American Community Surveys 2005-2011
Model (1) (2) (3)
Outcome Ln Wages (OLS)
Control Group Non-Autism Healthcare Practitioner
and Technical Occupations
Ln total autism 0.12210 -0.01644 -0.01175
(0.15324) (0.26983) (0.27795)
Ln total autism x
Pseudo target 0.01004 0.01007 0.01560
(0.02225) (0.02257) (0.01898)
Pseudo target 0.10369 0.10330 --
(0.16599) (0.16820)
Year indicators Yes Yes Yes
Center indicators Yes Yes Yes
Area demographics No Yes Yes
Center demographics No Yes Yes
Occupation indicators No No Yes
[R.sup.2] 0.91013 0.91034 0.94553
Observations 3,012 3,012 3,012
Model (4) (5) (6)
Outcome Provider Count (1-Year Lag) (Poisson)
Control Group Non-Autism Healthcare Practitioner
and Technical Occupations
Ln total autism 0.00741 -0.41967 ** -0.14335
(0.12372) (0.18884) (0.11209)
Ln total autism x
Pseudo target -0.01770 -0.01786 0.00110
(0.02066) (0.02087) (0.02547)
Pseudo target -1.01131 *** -1.01022 *** --
(0.17156) (0.17342)
Year indicators Yes Yes Yes
Center indicators Yes Yes Yes
Area demographics No Yes Yes
Center demographics No Yes Yes
Occupation indicators No No Yes
[R.sup.2] -- -- --
Observations 3,223 3,223 3,223
Notes: See Tables 3 and 5. Pseudo target represents the following
non-autism healthcare practitioner occupation codes: 3,000
(chiropractors); 3,010 (dentists); 3,030 (dieticians and
nutritionists); 3,040 (optometrists); 3,050 (pharmacists); 3,200
(radiation therapists); and 3,310 (dental hygienists). Autism
occupations are excluded from all models.
TABLE 9
Ln Annual Wages American Community Surveys 2005-2011: 15 California
Regional Centers (Aggregated Centers in Los Angeles County)
Model (1) (2) (3) (4)
Non-Autism Healthcare Practitioner
Control Group and Technical Occupations
Ln total autism -0.18686 -0.10847 -0.22723 -0.21509
(0.30589) (0.29951) (0.51682) (0.28619)
Ln total autism x 0.07278 *** 0.06294 *** 0.13465 ** 0.13951
Target (0.02438) (0.01799) (0.06649) (0.08302)
Ln total autism x -- -- -0.08349 ** --
Target x (0.03281)
Displacement
Target -0.55956 ** -- -1.00565 ** -0.17869
(autism (0.19121) (0.48927) (0.30506)
occupations)
Year indicators Yes Yes Yes Yes
Center indicators Yes Yes Yes Yes
Area demographics Yes Yes Yes Yes
Center Yes Yes Yes Yes
demographics
Occupation No Yes No No
indicators
Mental retardation No No No Yes
(MR) and mild/
moderate MR x
Target
[R.sup.2] 0.90672 0.94339 0.90894 0.90676
Observations 2,692 2,692 2,692 2,692
Model (5) (6) (7) (8)
Non-Autism All Healthcare
Control Group Occupations
Ln total autism -0.36153 * -0.25550 -0.38154 -0.38626 *
(0.20394) (0.20242) (0.40327) (0.18537)
Ln total autism x 0.06304 ** 0.05457 ** 0.13582 * 0.14142
Target (0.02446) (0.01871) (0.07283) (0.08847)
Ln total autism x -- -- -0.08914 ** --
Target x (0.03474)
Displacement
Target -0.27606 -- -0.79927 0.17179
(autism (0.19117) (0.53760) (0.35507)
occupations)
Year indicators Yes Yes Yes Yes
Center indicators Yes Yes Yes Yes
Area demographics Yes Yes Yes Yes
Center Yes Yes Yes Yes
demographics
Occupation No Yes No No
indicators
Mental retardation No No No Yes
(MR) and mild/
moderate MR x
Target
[R.sup.2] 0.87493 0.94102 0.87781 0.87497
Observations 3,352 3,352 3,352 3,352
Notes: See Tables 3 and 4.
