Skip To Content

Go to the Table Of Contents

Click for DHHS Home Page
Click for the SAMHSA Home Page
Click for the OAS Drug Abuse Statistics Home Page
Click for What's New
Click for Recent Reports and HighlightsClick for Information by Topic Click for OAS Data Systems and more Pubs Click for Data on Specific Drugs of Use Click for Short Reports and Facts Click for Frequently Asked Questions Click for Publications Click to send OAS Comments, Questions and Requests Click for OAS Home Page Click for Substance Abuse and Mental Health Services Administration Home Page Click to Search Our Site

 Health Services Utilization by Individuals with Substance Abuse and Mental Disorders

Chapter 7. A Hybrid Cost Function for Outpatient Nonmethadone Substance Abuse Treatment Facilities

Sarah Q. Duffy, Ph.D.
Laura J. Dunlap, M.A.
Moshe Feder, D.Sc.
Gary A. Zarkin, Ph.D.


A substantial literature provides estimates of the average cost of an episode of substance abuse treatment (e.g., see Anderson, Bowland, Cartwright, & Bassin, 1998; McGeary, French, Sacks, McKendrick, & De Leon, 2000), and various methods of estimating these costs have been used in benefit-cost analyses (see Cartwright, 2000, for a review). However, little work has been done to examine the cost structure of the substance abuse treatment industry. Extant work has focused on the association between cost and a measure of size based on point-prevalence client counts of residential treatment programs, a small and declining segment of the industry (Harwood, Kallinis, & Liu, 2001), or client counts and revenue in a convenience sample of outpatient substance abuse treatment facilities (Mark et al., 2000). Neither study employed economic methods commonly used to study cost structures in other health care industries. This chapter expands our understanding of the substance abuse treatment industry by using methods informed by economic theory to examine the cost structure of a national sample of substance abuse rehabilitation facilities that offer outpatient nonmethadone treatment. In 1999, such facilities accounted for approximately 82 percent of all substance abuse treatment facilities nationwide (Office of Applied Studies [OAS], 2001).

This chapter offers empirical evidence on the extent to which economies of scale exist in the outpatient nonmethadone substance abuse treatment industry, holding constant for client characteristics. If there are scale economies, and other studies find outcomes are no better in small programs than in large ones, State substance abuse treatment agencies and other payers may be able to encourage, through their licensing and payment policies, the formation of larger, more efficient programs, as long as the desired level of client access can be maintained. In addition, this research provides national estimates of costs for outpatient substance abuse treatment facilities, and how they vary with different mixes of clients and other characteristics, which could inform benefit-cost analyses. Most existing benefit-cost and cost-effectiveness analyses rely on a simple average cost obtained from data on a small number of purposively sampled facilities (Cartwright, 2000) or from revenue data from a national survey of substance abuse treatment facilities (Harwood, Hubbard, Collins, & Rachal, 1995). Incorporating findings based on nationally representative treatment cost data would improve the generalizability of findings from these studies. Finally, the research presented here includes measures of the characteristics of each facility's clients as a way to examine whether client characteristics are an important factor in facility costs. Such information may be important to policy makers who may wish to compare on the basis of costs programs that serve different types of clients.

Cost Function Specification

Economists often estimate cost functions to determine whether or not economies of scale can be realized in a particular industry. An industry is said to exhibit economies of scale if the average cost of producing a unit of output declines as more output is produced. If scale economies exist, the industry would be more efficient if it consisted of a few large firms rather than many small firms, other things being equal and given other constraints. In the case of substance abuse treatment, such a constraint might be ensuring adequate access. Economists also use cost functions to estimate the effect on costs of firm characteristics, such as the effect of teaching programs on hospital costs. In the substance abuse treatment industry, policy makers may wish to examine, for example, whether for-profit substance abuse treatment facilities have higher costs than nonprofit facilities, or the effect on costs of offering a variety of special programs. The purpose of this study is to explore these issues by specifying and estimating a cost function for a nationally representative sample of outpatient nonmethadone substance abuse treatment facilities.

Although, to our knowledge, no one has estimated an economic cost function for substance abuse treatment facilities, economists have estimated cost functions for other health care facilities, such as hospitals, nursing homes, and physician practices.1 These studies usually have used a functional form that is less restrictive than the standard textbook economic cost function. The standard textbook cost function, which is derived from the economic theory of the firm, models cost as a function of the volume of output and input prices only (e.g., see Silberberg, 1978, pp. 173–213). For an industry with N outputs (Y) and K input prices (w), the cost function would be

C = C (Y1, Y2, Y3, ... YN; w 1, w2, w 3, ..., wK),     D   (1)

where C is the cost of production. To obtain unbiased and efficient parameter estimates from a textbook cost function requires, among other things, that all firms in an industry use identical inputs and produce identical outputs, and that data on the quantity of all outputs produced and the prices of all inputs used by each firm are entered into the model. Economists use textbook cost functions to determine the economic properties of an industry's cost function, such as economies of scale and scope and the degree of input substitutability.2 When using data on industries that conform to the required assumptions, coefficient restrictions implied by the economic theory of the firm can be imposed on the estimation to improve statistical efficiency.

Economists estimating cost functions for health care facilities rarely use the textbook cost function for a variety of reasons, many of which apply to the outpatient drug-free substance abuse treatment industry. First, and perhaps most important, health care firms within an industry generally are not identical in terms of their inputs or outputs. Second, complete data outputs or input prices are rarely available. Finally, the focus of these studies often is the effect on costs of facility characteristics beyond output and input prices, such as whether for-profit providers have higher costs than nonprofit providers, or rural providers have lower costs than urban providers.

Instead, health economists often estimate what has been called a "hybrid" cost function (Rosko & Broyles, 1988). A hybrid cost function models costs as a function of measurable outputs and input prices, as well as other facility characteristics that may affect costs. We have

C = C(Y, w, X, F, A),     D   (2)

where, as before, Y is a vector of outputs and w is a vector of input prices. To these standard cost function variables we add a vector of measures of the case mix or severity of the facility's clients (X), a vector of facility characteristics (F), such as the number of services it offers, and a vector of characteristics describing the facility's location (A). In this chapter, we estimate a model similar to equation (2). The one difference is that because we have only one output, Y will be a scalar rather than a vector.

To estimate a cost function, the model must be written as a specific functional form that allows for the possibility for economies of scale. A linear function does not allow for economies of scale because it constrains the relationship between cost and the quantity of output to be the same over the entire output range, rather than allowing it to decline, as would be the case under economies of scale, or increase, as in the case of diseconomies of scale. We follow Vitaliano (1987) who used the following logarithmic estimating equation derived from a theoretical model proposed by Nerlove:

     D   (3)

where C, Y, w, X, F, and A are as defined before; image representing alpha, image representing beta, image representing delta, image representing mu, and image representing lambda are coefficients to be estimated; and image representing epsilon is a random, normally, and identically distributed error term. Like Vitaliano (1987), we used Ramsey's Regression Specification Error Test (RESET) to test the suitability of this functional form, as well as other popular functional forms, including the quadradic and translog specifications. Our testing revealed that the logarithmic model was the only functional form for which the null hypothesis of no specification error could not be rejected at conventional levels of significance.


We used data from the Alcohol and Drug Services Study (ADSS) on 222 outpatient nonmethadone treatment facilities. ADSS was conducted under the auspices of the Office of Applied Studies (OAS) at the Substance Abuse and Mental Health Services Administration (SAMHSA) (OAS, 2000a). Among the objectives of ADSS was the collection of detailed information on the characteristics of a random, nationally representative sample of substance abuse treatment facilities and of clients discharged from those facilities. Such data allow development of better estimates of the costs of treatment than had been available using previously existing datasets. The study consisted of three phases, two of which we draw from for this analysis.

Phase I of ADSS was a telephone interview with a nationally representative, stratified probability proportionate to size (pps) sample of 2,395 substance abuse treatment facilities, representing a 91.4 percent response rate from the 2,621 eligible facilities. Facility administrators were asked about the characteristics of the facility and its clients. Phase I missing data were imputed according to standard statistical procedures (OAS, 2000b). Final facility weights included a nonresponse adjustment, and methods used to impute missing values included logical imputation, imputation from external sources, regression, and random within-class hot-deck procedures.

Phase II facilities were selected from among Phase I facilities located in 1 of 62 primary sampling units (PSUs) again according to a stratified pps design. Excluded from Phase II were hospital inpatient facilities and facilities that treated only those with alcohol use disorders. These facilities were considered out-of-scope. Phase II data were initially collected during a site visit to 280 facilities. These 280 facilities included 234 of the 294 eligible facilities that agreed to be surveyed, as well as 46 shadow3 facilities for the 60 that chose not to participate. Thus, the overall response rate for cost study facilities was approximately 86 percent. The site visit consisted of an in-person interview with the facility director or administrator to collect data on the facility's characteristics, including expense and revenue information; compilation of a sampling frame and selection of a representative sample of client records; and collection of client-level data from the sample of client records at each facility (OAS, 2000a). In addition to the variables that were imputed on the final Phase II facility file, we used the same methods to impute additional missing variables that we needed for our analysis (Krenzke & Mohadjer, 2002). The variables and the number of observations we imputed are noted below.

Phase II administrator data were used as the basis for the ADSS cost study, the main source of data for this chapter. Data on facilities' expenses, revenues, and client volumes were entered into a data audit spreadsheet developed by Capital Consulting Corporation (CCC) to check for accuracy (OAS, 2003). CCC had developed this audit instrument after intensive study of some 400 substance abuse treatment facilities using their Cost Allocation Methodology (CCC, 1998; CCC, The Lewin Group, & Caliber Associates, 1998). In that study, CCC sent professional accountants to facilities to collect cost data. From this information, CCC developed data reliability and validation procedures to test the accuracy of other provider-supplied data. Those procedures are at the heart of the data audit spreadsheet used in the ADSS cost study. The facilities' original responses were entered into the audit instrument, and key financial ratios were examined. Facilities with anomalous results on their Phase II cost, client volume, or personnel variables were contacted and given the opportunity to change their responses. Some 96 percent of the facilities required a callback for at least one variable. Of those, most were able to either verify the originally reported data or provide new information, much of it documented with information from the facility's financial information system. Some facilities either could not be contacted or could not provide more information on their expenses. For those facilities, which amounted to about 26 percent of the outpatient nonmethadone facilities studied here, data were imputed. In most cases, almost 22 percent of all outpatient nonmethadone facilities, missing expense data were imputed using the facility's own volume information and the data audit instrument, or the facility's Phase I response to the expense question. In 4.5 percent of the facilities, expenses were imputed using information from other facilities (OAS, 2003). The cost study data file includes final Phase II facility weights, which were adjusted for facility nonresponse using a raking procedure (OAS, 2000a). In addition to using the ADSS data, we used some supplementary data, as described below.

The variables included in the analysis are described in the following paragraphs.

Costs: Our measure of costs, the dependent variable, was the natural logarithm of the total substance abuse treatment costs in the facility, as reported by the administrator in the Phase II administrator interview and verified or edited as described above.

Output: We estimated a single output model, where we considered a unit of output as a treatment episode for a given client within a given type of care within a given facility. However, due to the characteristics of the substance abuse treatment industry and its clients, and the measures available in the data, we estimated the models using two different measures to determine whether they exhibited different relationships with costs. The number of annual admissions was the first measure of output we considered. Many substance abuse treatment facilities have to keep track of admissions because they are required to report them to their State substance abuse treatment agencies (OAS, 2000c). However, the number of admissions may overstate the annual output of a treatment center because a large number of clients who are admitted do not finish treatment. In the ADSS Phase I sample, for example, administrators reported that, on average, more than 40 percent of the clients in outpatient nonmethadone facilities failed to complete their planned treatment. In our study, we entered the facility's completion rate as an explanatory variable to partially control for that issue. We also considered using discharges, an output measure commonly used in hospital studies. However, that measure also has its limitations. As mentioned earlier, a large number of clients leave programs without finishing their treatment plan. Their discharge often is not documented until some time period elapses, usually 30 days, during which they have received no treatment. Furthermore, some facilities do not formally discharge anyone, due to the relapsing nature of the substance use disorder. In these facilities, clients are not discharged and are allowed to return for more care if needed. Therefore, discharge numbers may underestimate treatment output. In this study, we estimated two models—one with admissions as the measure of output and one with discharges. Because the results were virtually identical, we report only the results of the admissions regressions here. Given that many facilities must report these routinely to State substance abuse treatment authorities, we believe they may be more accurately reported than are discharges.

Input Prices: The inputs for which we had prices were various categories of labor and office space. To measure wages, the input price of labor, we used the Metropolitan and Balance of State Area Occupational Employment and Wage Estimates for selected occupations, collected by the Bureau of Labor Statistics (2003b) as part of their Occupational Employment Statistics Survey (OESS). The three occupations we included were (a) Substance Abuse Counselors, (b) Senior Administrative (includes five job categories: administrative services manager, general managers and other top executives, all other managers, and financial managers), and (c) Administrative Clerical Workers (includes seven job categories: file clerks, general office clerks, payroll and timekeeping clerks, and secretaries [excluding legal and medical]). Although the latter two occupation categories are not specific to substance abuse treatment facilities, we believe the average wages across administrative workers in all industries should serve as a good proxy for administrative wages in substance abuse treatment facilities. Even though the ADSS cost study collected facility-specific wage information, we used the OESS estimates of the market wage rates to overcome two possible problems in estimation. The first was the problem of missing observations. Not all facilities employ all types of workers, so we would not have wage information for all possible types of employees for all facilities, which is required for estimating a cost function. The second potential problem was endogeneity. If we were to use a weighted average of each facility's wages as its wage variable, to overcome the missing values problem, we would no longer just be measuring the effect of exogenous factors, such as location in a high cost or highly unionized area, on the facility's cost function. Rather, we would be measuring both the prevailing wage rate and the facility's choice of inputs, which is endogenous. For example, a facility's higher costs may be attributed to its being in a higher wage area, when, in reality, its costs are higher because it has an inefficient staffing mix. To account for this possibility, economists often use local area wage rates (e.g., see Custer & Wilke, 1991; Salkever et al., 1986; Zuckerman et al., 1994), as was our plan here. Because the OESS does not collect data on all occupations for each location each year, for some variables we used data from later years, deflated using the average Consumer Price Index (CPI) representing changes in the prices of all goods and services purchased for consumption by urban households (Bureau of Labor Statistics, 2003a).

We also included a variable measuring the cost of office space in the facility's area, again turning to an external source, this time because the variable was not collected in ADSS. However, it too is potentially endogenous. Unfortunately, to our knowledge, no data exist on commercial real estate rental rates throughout the Nation. So, instead, we used the four-bedroom Section 8 Fair Market Rent as defined by the U.S. Department of Housing and Urban Development's Office of Policy Development and Research (1995) as a proxy for commercial rents. This is the variable that is used in SAMHSA's Substance Abuse Prevention and Treatment (SAPT) block grant formula to measure the rental component of the cost of doing business in each State (U.S. Department of Health and Human Services [DHHS], 1996).

Case Mix Adjustment for Substance Abuse Treatment Facilities

Adjusting for case mix in analyses of substance abuse treatment facilities has become more common in the past few years as governments and other payers demand accountability from service providers. One way to determine whether a provider is accountable is to compare that provider's performance with that of other providers. For the comparison to be fair and meaningful, however, it must account for differences across providers in the mix of clients they treat. Several recent studies of substance abuse treatment providers have used the same basic method of case mix adjustment to study outcomes of substance abuse treatment (Koenig, Fields, Dall, Ameen, & Harwood, 2000; Moos, Moos, & Andrassy, 1999; Phibbs, Swindle, & Recine, 1997; Phillips et al., 1995), access to substance abuse treatment (Deck, McFarland, Titus, Laws, & Gabriel, 2000), and participation in self-help groups among those in treatment (Ouimette et al., 2001). Although the results from these studies are informative and useful in outcome studies, they unfortunately provide little guidance on how to measure case mix complexity as it relates to the cost of treatment at a given facility.

The study conducted by Moos et al. (1999) is illustrative. These researchers analyzed outcomes of four treatment approaches (therapeutic community, psychosocial rehabilitation, 12–step, and undifferentiated) in 88 residential programs. The outcomes they investigated were posttreatment abstinence, substance use disorders, distress, mental disorder symptoms, arrests, and employment. They presented results from separate regressions run with each outcome as the dependent variable and the client as the unit of analysis. Each regression included variables describing clients and the programs in which they had been treated. Client characteristics included age, marital status, prior mental disorder episode, mental disorder diagnosis, and the value of the outcome variable at intake. With the exception of the outcome variable measured at intake, no variable was significant in all regressions. Only one, prior mental disorder episode, had the same effect on all the outcomes for which it was significant.

This lack of consistency in results holds across the other studies cited as well. No variable had a similar effect on all outcomes in all of the studies in which it was included. In state-of-the-art studies of substance abuse treatment, case mix adjustment methods vary by the purpose of the analysis and the outcome and population being studied. To our knowledge, no one has attempted to adjust for the costliness of a provider's client population when comparing substance abuse treatment providers on the basis of costs.

This is in contrast to cost studies of the hospital industry in which case mix adjustment, based on data routinely collected by hospitals from patients' discharge abstracts, has been used for more than 20 years. Information on the clients' medical diagnoses, procedures, length of stay, age, and gender is abstracted from each patient's record, then standardized and combined with information on the amount the patient was charged for the stay to create an electronic discharge abstract record. Electronic databases containing these records then are forwarded to some 40 States across the country that require hospitals to submit these data, many of which make it available for researchers. Using such data, researchers create a case mix index that describes the costliness of treating each hospital's patients. The case mix index usually takes the following form:

CMIh = hospital h's case mix index;

weightj = a resource use weight, such as the total charge or the length of stay, assigned to homogeneous patient group j;

grouphj = proportion of hospital h's patients in group j; and

groupj = average proportion of patients in group j.

A hospital with a CMI greater than 1 has a case mix that is more costly to treat than the average hospital's; a hospital with values less than 1 has a case mix that is less costly to treat than the average hospital's.

When examining hospital costs, researchers often use groupings and weights created by the U.S. Government as part of Medicare's hospital inpatient prospective payment system. The groups, called diagnosis related groups (DRGs), are groupings of patients that, according to analysis of discharge abstract data, cost a similar amount to treat. The weights are the average costs incurred in treating cases in that DRG, relative to the average costs incurred in treating all DRGs. Each hospital discharge can be classified into a DRG based on the principal diagnosis, up to eight additional diagnoses, and up to six procedures performed during the stay, as well as age, gender, and discharge status of the patient. There are more than 500 DRGs, and their weights are reviewed each year and updated as necessary (Centers for Medicare & Medicaid Services [CMS], 2001).

Unfortunately, the data required to create a grouping system that could be used to construct weights for substance abuse treatment client groups cannot easily be obtained. Although States routinely require substance abuse treatment providers to submit data on individual clients, these are usually admissions data, which do not include any information on the costs of the client's treatment or the client's disposition at discharge. Sometimes these data elements are collected in special studies, as was done in ADSS. However, an examination of the ADSS Phase II client discharge abstract data reveals that the data are missing from a large number of clients. Total charge was missing or invalid for some 35 percent of the outpatient nonmethadone clients in ADSS, it was zero for another 12 percent, and it was on a sliding scale or otherwise reduced for another 17 percent. Without complete charge or cost information for each client, it is difficult to create groups of clients based on resource use. Although length of stay could conceivably be used, as it was in the early work on DRGs (Fetter, Shin, Freeman, Averill, & Thompson, 1980), the problems noted above about how discharges are documented and handled make using length of stay to measure resource use questionable.

For these reasons, we did not control for case mix among substance abuse treatment facilities using a single case mix index. Instead, we used a method that was employed in hospital studies before case mix indices were widely adopted. We included several summary measures of characteristics that we believe may make clients more costly to treat.

To determine which summary measures to use, we reviewed the substance abuse treatment literature relating treatment costs to client characteristics. Unfortunately, there appears to be a dearth of literature on this subject as well, and most of what is available is based on analyses of claims data, usually for specific time periods rather than by treatment episode at a given facility (Ettner, Frank, McGuire, Newhouse, & Notman, 1998; Goodman, Holder, Nishiura, & Hankin, 1992; Goodman, Nishiura, & Hankin, 1998; Goodman, Nishiura, Hankin, Holder, & Tilford, 1996; Holder & Blose, 1991; Huskamp, 1999; Salomé, French, Scott, Foss, & Dennis, 2003; Westermeyer, Eames, & Nugent, 1998).

Thus, we chose the following summary measures of client severity based on data availability and characteristics that we expected to affect the cost of treatment. Some clients have more complex substance use disorders and therefore require more resources to treat. These may include those who, at admission, used both drugs and alcohol, injected drugs (imputed for nine facilities), or who received supplemental security income (SSI) or social security disability insurance (SSDI) (imputed for two facilities). We included in our models measures representing the percentage of each facility's admissions with each of these characteristics. This information comes from the discharge abstracts, when available, or from the Phase I ADSS administrator survey (OAS, 2000a).

Other clients may be more costly to treat for other reasons. Those who are referred by the criminal justice system, for example, may be less compliant and ready for treatment and therefore require more resources to treat compared with those referred in some other way, so we included the percentage of clients who were referred by the criminal justice system in the model (imputed for one facility). Minority clients may be more costly to treat if they require more services due to their social and economic disadvantages (D'Aunno & Vaughn, 1995). To capture this, we entered four variables from the Phase I administrator survey: the administrator's estimate of the percentage of each facility's clients who were Hispanic, non-Hispanic black, other race, or unknown race on the point-prevalence date (October 1, 1996), with non-Hispanic white as the reference category (imputed for one facility). Ten facilities reported that the race of all of their clients was "unknown," an allowable response to the survey. It is unclear if these facilities truly did not know the race of all their clients, or chose not to respond to the question. If they truly did not know or ask, it could be that race/ethnicity plays no part in their treatment decisions and therefore resources required to treat these clients. Because this was unclear, we ran the model both including and excluding these 10 facilities and found that the results did not differ significantly. We report the results including all of the facilities and include "unknown" race as a category.

We ran versions of our model with other client mix variables we thought might affect facility costs, but they were not significant, and removing them from the model did not affect any of the coefficients on the remaining variables. The variables we considered but excluded were the percentage of clients who were homeless, unemployed, or had co-occurring substance use and mental disorders at admission, and the percentage who completed treatment.

Facility Characteristics

We included several facility characteristics in the model. We entered dummy variables to measure the extent of urbanization of the facility's location. These dummy variables indicated whether the facility was in a small- or medium-sized metropolitan area or a nonmetropolitan area, with large metropolitan area comprising the reference cell. Metropolitan area classifications were based on the Beale urbanicity codes, which were assigned based on the facility's zip code (Butler & Beale, 1994). Nonmetropolitan facilities are located in nonmetropolitan counties. Small and medium-sized metropolitan area facilities are located in metropolitan areas with fewer than 1 million people. Large metropolitan area facilities are located in metropolitan areas with more than 1 million people. We expected facilities in large urban areas to have higher costs than other facilities for input prices we did not measure (e.g., security). In addition, we entered a dummy variable indicating whether or not the facility was owned by a private for-profit entity. Government-owned and nonprofit facilities comprised the reference group. According to economic theory, in a competitive market, the profit motive forces firms to minimize the costs of doing business. Private nonprofit and government-owned facilities face no profit incentive, so they may be less likely to be cost minimizers.4 Previous research suggested that such differences may exist in the substance abuse treatment industry (Wheeler, Fadel, & D'Aunno, 1992). However, it also should be noted that, if our client characteristic variables did not adequately account for case mix, we also might have expected a profit/nonprofit differential—private for-profit facilities may admit less severely ill clients compared with private nonprofit and government-owned facilities, which may be more severely ill clients' last resort.

Another difference among facilities is the number of services and the number of special programs for specific populations they offer. A substantial number of those with substance use disorders also have other problems, such as homelessness, unemployment, or physical or mental illnesses that must be addressed for treatment to be successful (D'Aunno & Vaughn, 1995). Because, again, we could find no direction in the literature as to which services were most likely to affect costs, we included a variable that measured the number of the following services offered by the facility, based on the administrator's response to the Phase I ADSS survey: comprehensive assessment/diagnosis, transportation, individual therapy, relapse prevention, family counseling, employment counseling, academic education, HIV/AIDS education/counseling/support, combined substance abuse treatment and mental health services, tuberculosis (TB) screening, prenatal care, smoking cessation, acupuncture, aftercare, outcome follow-up, urine screens, alcohol and other drug tests, medical detox, mental health services, and medical treatment.

Finally, the diversity of a facility's clients may affect its cost due to the importance of offering culturally competent care. For example, facilities that serve clients who speak many different languages may need to hire a greater variety of counselors or counselors with special skills. To measure this, we computed a race/ethnicity index similar to the Herfindahl index commonly used in the industrial organization literature to measure market structure. The race/ethnicity index for facility t, Rt, is


where Stj is the share of the ith racial/ethnic group in facility t. We included the five groups identified earlier: non-Hispanic white, non-Hispanic black, Hispanic, other, and unknown. Rt is bounded by 0 and 1 and is inversely related to diversity. The greater Rt is, the less diverse is the facility; facilities with only one racial/ethnic group have Rt = 1. Rt declines as the number of racial/ethnic groups at the facility increases. It increases with rising inequality among any given number of racial/ethnic groups. We hypothesized that Rt is negatively related to costs.

Several other facility characteristics also may affect costs. One such characteristic is whether or not the facility is part of a larger organization. Although facilities were encouraged to report total costs, including dollar values of items supplied by a parent company, some facilities may have had trouble doing so. Furthermore, being part of a larger entity may allow the facility to obtain inputs at lower prices, or take part in other efficiencies, which may lower its costs. Therefore, we hypothesized that facilities that were part of a larger organization may have had lower costs. Because many substance abuse treatment facilities are nonprofit, they may received in-kind donations of goods or services, such as space, furniture, or volunteer help. Facilities that received in-kind donations should have had lower expenses, so we included a dummy variable from the ADSS Phase II survey that indicated that the facility received such donations (imputed for one facility).

We suspected other facility characteristics might have affected cost, such as the number of special programs offered by the facility (intensive outpatient treatment and special programs for women, pregnant women, adolescents, DWI/DUI clients, AIDS/HIV positive, and co-occurring disorder clients), the age of the facility, and a variable indicating that the facility was a multimodality facility. However, we excluded these variables from the final model because the estimated parameters were not significantly different from zero, and excluding these variables did not affect the other results.


We estimated the cost function parameters using design-weighted least squares regression techniques. Because the ADSS data were collected according to a complex survey design, including stratification, unequal probabilities of inclusion and clustering, as well as nonresponse adjustment, we used methods appropriate for the design to compute the point estimates, standard errors, and statistical tests. ADSS was designed to be used with a jackknife replicate variance estimation method. However, comparing the jackknife variance estimates with those computed using the Taylor linearization method revealed great differences between these estimates (with the jackknife standard errors between 22 and 242 percent greater than the Taylor linearization ones). Bell and McCaffrey (2002) and McCaffrey, Bell, and Botts (2001) demonstrated that, under certain conditions, the jackknife method overestimates the variance, although the Taylor linearization method underestimates it. These authors offered methods for estimating an unbiased variance using data from unstratified designs, both weighted and unweighted. However, the extension of these methods to stratified designs is not straightforward. Further, the assumptions made regarding the population model may be hard to justify in our case. Therefore, we developed, using simulations, a new method to estimate standard errors from stratified data.

We started with the observation that the Taylor linearization and the jackknife methods normally yield closer estimates when the number of clusters in the sample is large and the number of estimated parameters is not very large. However, when this is not the case, the biases of these methods may be significant. This appears to be the case with our use of the ADSS cost study data because the estimates obtained by the jackknife method were much larger than those obtained by the Taylor series method, and our model contains a relatively large number of parameters. We estimated the standard error using the Taylor linearization method, then adjusted the result upward by a factor. This method was motivated by the known result that the full maximum likelihood estimate of the variance in a linear regression model with independently and identically distributed (IID) data is negatively biased on the order of p/n, where p is the number of covariates and n is the sample size. This bias arises from overfitting, as was also suggested by Bell and McCaffrey (2002) and validated by our simulations. We showed that, in the case of complex data, the bias of the variance estimated using Taylor linearization is still approximately p/n where now n, the effective sample size, is a function of the number of strata, clusters, and units within a cluster, and the intracluster correlations. We used a simulation approach to determine this effective sample size. Our simulations used a sample design similar to that of the ADSS sample, with a similar mix of continuous and dichotomous covariates. These simulations showed that the Taylor linearization bias was approximately -p/46. Thus, we used an effective sample size of 46 for correction of the bias of the Taylor linearization variance estimation. Accordingly, we multiplied the Taylor series standard error estimates by the square root of [n/(n-p)], or 1.356.


Table 7.1 displays the weighted mean estimates of the variables included in our analysis for our sample of outpatient nonmethadone substance abuse treatment facilities. After incorporating the final facility weights, the total population size represented was 9,166 facilities (N = 222, unweighted sample).

The mean of the log of total costs was approximately 12.13 in 1997, which translates into $185,350. The mean of the log of annual admissions was approximately 4.94, or 140 admissions, yielding a cost per admission of $1,324. Of the three different mean hourly wages included as input prices, the senior administrative wage was the highest ($24.29), followed by the substance abuse counselor wage ($12.55), and the clerical administrative wage ($9.97). The average of our proxy for office space costs was approximately $916 per month.

The average facility treated a client population that was somewhat racially diverse. Most clients in the facilities were non-Hispanic white, with the mean across facilities of 63 percent of clients. The second largest racial/ethnic group in most facilities was non-Hispanic black, with the mean across facilities of 22 percent of clients. Hispanics accounted for, on average, about 9 percent of a facility's client population. Clients of other races comprised about 2 percent of total clients, and those with an unknown race or ethnicity, approximately 4 percent. The derived race/ethnicity index had a weighted mean estimate of 0.7 across facilities, indicating at least some diversity in the average facility.

Almost half of the clients in the facilities were referred by the criminal justice system, with the mean across facilities of 42 percent. A majority of clients in the facilities received treatment for both alcohol and drug use disorders, with the mean across facilities of 54 percent. The average proportion of those who inject drugs within a facility was low (10 percent), and few clients reported receiving SSI or SSDI (mean across facilities of 8 percent of clients).

Most of the facilities in the sample were private nonprofit (59 percent). An estimated 27 percent were private for-profit, and 14 percent were government-owned. On average, each facility offered approximately 10 services. Finally, most facilities were located in a large-sized metropolitan area (55 percent), with 35 percent located in a small to medium-sized metropolitan area and 11 percent in a nonmetropolitan area.

Cost Function Results

As mentioned earlier, estimating standard errors using these data was not straightforward. In Table 7.2, we present the coefficient estimates, along with three estimates of the standard errors. As Table 7.2 shows, the standard errors estimated using the Taylor series method are much smaller than those estimated using the Jackknife method. Because we believe there is bias in both estimates—Taylor standard errors are biased downward and Jackknife estimates are biased upward—we adjusted the Taylor standard errors to create an adjusted standard error that falls between the two. We present this adjusted standard error as well and its p value. These are the results we discuss below.

Table 7.1 Means and Standard Errors of Analysis Variables, Outpatient Nonmethadone Substance Abuse Treatment Facilities, ADSS Phase II Administrator and Cost Studies, 1997 (N = 222; Weighted N = 9,166)
Variable Mean Standard Error
     Log of Costs ($185,350 total costs) 12.13 0.09
     Log of Admissions ($1,324 per admission, 140 admissions) 4.94 0.12
     Log of Substance Abuse Counselor Wage ($12.55 per hour) 2.53 0.03
     Log of Clerical Administrative Wage ($9.97 per hour) 2.30 0.02
     Log of Senior Administrative Wage ($24.29 per hour) 3.19 0.03
     Log of Office Space Cost ($916 per month) 6.82 0.04
     % of Clients Who Are White 0.63 0.04
     % of Clients Who Are Black 0.22 0.03
     % of Clients Who Are Hispanic 0.09 0.02
     % of Clients Who Are Another Race 0.02 0.01
     % of Clients Whose Race Is Unknown 0.04 0.02
     Herfindahl Index 0.70 0.03
     % of Clients Referred by Criminal Justice System 0.42 0.05
     % of Clients with Drug and Alcohol Use Disorders 0.54 0.05
     % of Clients Who Injected Drugs 0.10 0.02
     % of Clients Who Received SSI or SSDI 0.08 0.02
     Private For-Profit Facility 0.27 0.07
     Private Nonprofit Facility 0.59 0.07
     Government Facility 0.14 0.04
     Sum of Special Services 10.38 0.33
     Part of a Larger Organization 0.57 0.07
     Facility Receives In-Kind Goods or Services 0.45 0.08
     Facility Located in a Large Metro Area 0.55 0.08
     Facility Located in a Medium or Small Metro Area 0.35 0.08
     Facility Located in a Nonmetro Area 0.11 0.06
SSI = supplemental security income.
SSDI = social security disability income.
Source: SAMHSA, Office of Applied Studies, Alcohol and Drug Services Study, Phase II administrator data, 1997 to 1999, and ADSS cost study, 1997.

Our results suggest that there are substantial economies of scale in the outpatient nonmethadone substance abuse treatment industry. We found that a 10 percent increase in the total number of admissions was associated with only a 6.7 percent increase in total costs (p < 0.0001). Because the estimate also was statistically different from 1 at the 0.01 percent level, the results suggest that the outpatient nonmethadone substance abuse treatment industry experiences economies of scale. Larger facilities were less costly on a per admission basis than smaller ones.

With one exception, the facility characteristics were insignificant. The only client mix variable that was significant was the percentage of clients with SSI or SSDI. Facilities with a larger percentage of clients on SSI or SSDI had higher costs (p = 0.0418).

Table 7.2 Cost Function Results, Outpatient Nonmethadone Substance Abuse Treatment Facilities, ADSS Phase II Administrator and Cost Studies, 1997
Variable Coefficient Taylor Series Standard Error Jackknife Replicate Standard Error Adjusted Standard Error Adjusted T Statistic Adjusted p Value
Intercept 6.71 1.41 2.25 1.91 3.51 0.0009
Log of Admissions 0.67 0.04 0.06 0.05 12.36 <0.0001
Log of Substance Abuse Counselor Wage -0.20 0.42 0.67 0.57 -0.36 0.7201
Log of Clerical Administrative Wage -1.09 1.06 1.84 1.44 -0.76 0.4502
Log of Senior Administrative Wage 0.92 0.58 0.91 0.79 1.17 0.2466
Log of Office Space Cost 0.23 0.41 0.61 0.56 0.41 0.6833
% of Clients Who Are Black 0.17 0.19 0.26 0.26 0.66 0.5118
% of Clients Who Are Hispanic 0.00 0.24 0.39 0.33 -0.01 0.9921
% of Clients Who Are Another Race -1.66 0.84 2.88 1.14 -1.46 0.1495
% of Clients Whose Race Is Unknown 0.23 0.29 0.54 0.39 0.57 0.5708
Herfindahl Index 0.50 0.29 0.50 0.39 1.27 0.2090
% of Clients Referred by Criminal Justice System -0.12 0.20 0.31 0.27 -0.44 0.6614
% of Clients with Drug and Alcohol Use Disorders -0.12 0.15 0.24 0.20 -0.60 0.5508
% of Clients Who Injected Drugs 0.28 0.32 0.39 0.43 0.64 0.5246
% of Clients Who Received SSI or SSDI 0.65 0.23 0.35 0.31 2.08 0.0418
Private For-Profit Facility -0.15 0.14 0.20 0.19 -0.80 0.4269
Sum of Special Services 0.04 0.02 0.03 0.03 1.52 0.1338
Part of a Larger Organization -0.09 0.10 0.15 0.14 -0.67 0.5054
Facility Receives In-Kind Goods or Services 0.23 0.11 0.15 0.15 1.53 0.1313
Facility Located in a Medium or Small Metro Area -0.21 0.12 0.18 0.16 -1.26 0.2126
Facility Located in a Nonmetro Area -0.40 0.21 0.43 0.28 -1.40 0.1666
SSI = supplemental security income.
SSDI = social security disability income.
Source: SAMHSA, Office of Applied Studies, Alcohol and Drug Services Study, Phase II administrator data, 1997 to 1999, and ADSS cost study, 1997.


Our results reveal that economies of scale exist throughout the output range in outpatient nonmethadone substance abuse treatment facilities. This suggests that larger facilities may be able to provide care at a lower price than smaller facilities. If other studies conclude that larger facilities provide care that is at least as good as smaller facilities, and client access can be maintained, then State governments and other payers may wish to consider promoting the formation of larger programs through their payment and licensing policies. For example, they could consider setting payments at a rate at which smaller facilities cannot survive.5 Other things being equal, such policies may free resources that can be used to treat more individuals with these disorders.

With one exception, we find that the mix of clients, at least to the extent we can measure it, does not appear to explain variations in the cost of running a substance abuse treatment facility. That one exception was clients who were on SSDI or SSI. These clients appear to be more costly to treat. No other facility characteristics were significant.

The presence of economies of scale calls into question the practice of using costs estimated from a small number of nonrandomly selected facilities in benefit-cost and cost-effectiveness analyses of the substance abuse treatment industry. If the facilities included in those studies have costs that are sufficiently higher than the average facility because they are much smaller than the average facility, a promising addition to treatment may fail to be implemented because the additional benefits do not appear to justify the costs. However, the costs may be justified at a more efficient facility. The opposite may hold for results from studies that are conducted at larger facilities. Treatments that are cost-effective in larger facilities may not be in smaller ones.

Some limitations must be noted. First, further research is needed to overcome at least one limitation of this study. Although we attempted to control for case mix using the data and methods currently available, these methods are not as advanced as those used to examine other health care facilities, and it is likely that they are not entirely sufficient. Better controls for case mix that are based on detailed analyses of data on the costs associated with treating clients with different kinds of disorders and treatment needs have to be developed to promote a fairer and more accurate comparison of treatment facility performance. Second, there are some limitations to the data used in this study. Especially noteworthy is that a relatively high percentage of the expense data could not be used as originally submitted and had to be revised or imputed using the methods described earlier. Also as described earlier, we used proxies for some of the input prices because exact measures were unavailable. Improved data collection would lead to more accurate results.


Anderson, D. W., Bowland, B. J., Cartwright, W. S., & Bassin, G. (1998). Service-level costing of drug abuse treatment. Journal of Substance Abuse Treatment, 15, 201–211.

Anderson, G. F., & Lave, J. R. (1986). Financing graduate medical education using multiple regression to set payment rates. Inquiry, 23, 191–199.

Bell, R. M., & McCaffrey, D. F. (2002). Bias reduction in standard errors for linear regression with multi-stage samples. Survey Methodology, 28(2), 169–181.

Bilodeau, D., Crémieux, P.-Y., & Ouellette, P. (2000). Hospital cost function in a non-market health care system. Review of Economics & Statistics, 82, 489–498.

Bureau of Labor Statistics, U.S. Department of Labor. (2003a). Home page of the Bureau of Labor Statistics. Retrieved November 5, 2003, from

Bureau of Labor Statistics, U.S. Department of Labor. (2003b). Home page of Occupational Employment Statistics (OES) program. Retrieved November 5, 2003, from

Butler, M. A., & Beale, C. L. (1994). Rural-urban continuum codes for metro and non-metro counties, 1993 (Staff Report No. AGES 9425). Washington, DC: U.S. Department of Agriculture, Economic Research Service. [Codes available at]

Capital Consulting Corporation. (1998, June 30). Uniform system of accounting and cost reporting for substance abuse treatment providers (Contract No. SAMHSA 270–97–7001). Fairfax, VA: Author. [Available as a PDF at]

Capital Consulting Corporation, The Lewin Group, & Caliber Associates. (1998). Measuring the cost of substance abuse treatment services: An overview. Rockville, MD: Substance Abuse and Mental Health Services Administration, Center for Substance Abuse Treatment.

Carey, K. (1987). A panel data design for estimation of hospital cost functions. Review of Economics & Statistics, 79, 443–453.

Cartwright, W. S. (2000). Cost-benefit analysis of drug treatment services: Review of the literature. Journal of Mental Health Policy and Economics, 3(1), 11–26.

Centers for Medicare & Medicaid Services [formerly HCFA, Health Care Financing Administration]. (2001, May 4). Changes to the hospital inpatient prospective payment systems and fiscal year 2002 rates: Proposed rule (HCFA-1158–P). Federal Register, 66(87), 22646–22891. [Available as a PDF at]

Cowing, T. G., & Holtman, A. G. (1983). Multiproduct short-run hospital cost functions: Empirical evidence and policy implications from cross-section data. Southern Economic Journal, 49, 637–653.

Custer, W. S., & Willke, R. J. (1991). Teaching hospital costs: The effects of medical staff characteristics. Health Services Research, 25, 831–857.

D'Aunno, T., & Vaughn, T. E. (1995). An organizational analysis of service patterns in outpatient drug abuse treatment units. Journal of Substance Abuse, 7, 27–42.

Deck, D. D., McFarland, B. H., Titus, J. M., Laws, K. E., & Gabriel, R. M. (2000). Access to substance abuse treatment services under the Oregon Health Plan. Journal of the American Medical Association, 284, 2093–2099.

Dor, A., Duffy, S., & Wong, H. (1997). Expense preference behavior and contract-management: Evidence from U. S. hospitals. Southern Economic Journal, 64, 542–554.

Duffy, S. Q., Ruseski, J. E., & Cavanaugh, S. (2000). Graduate medical education costs in nonacademic health center teaching hospitals: Evidence from Maryland. Medical Care Research and Review, 57(1), 3–23. Discussion in 57(1), 24–28.

Escarce, J. J., & Pauly, M. V. (1998). Physician opportunity costs in physician practice cost functions. Journal of Health Economics, 17, 129–151.

Ettner, S. L., Frank, R. G., McGuire, T. G., Newhouse, J. P., & Notman, E. H. (1998). Risk adjustment of mental health and substance abuse payments. Inquiry, 35, 223–239.

Evans, R. G. (1971). 'Behavioral' cost functions for hospitals. Canadian Journal of Economics, 4, 198–215.

Fetter, R. B., Shin, Y., Freeman, J. L., Averill, R. F., & Thompson, J. D. (1980). Case mix definition by diagnosis-related groups. Medical Care, 18(2 Supplement), iii, 1–53.

Gaskin, D. J., & Hadley, J. (1997). The impact of HMO penetration on the rate of hospital cost inflation, 1985–1993. Inquiry, 34, 205–216.

Goodman, A. C., Holder, H. D., Nishiura, E., & Hankin, J. R. (1992). An analysis of short-term alcoholism treatment cost functions. Medical Care, 30, 795–810.

Goodman, A. C., Nishiura, E., & Hankin, J. R. (1998). Short term drug abuse treatment costs and utilization: A multi-employer analysis. Medical Care, 36, 1214–1227.

Goodman, A. C., Nishiura, E., Hankin, J. R., Holder, H. D., & Tilford, J. M. (1996). Long-term alcoholism treatment costs. Medical Care Research and Review, 53, 441–464.

Grannemann, T. W., Brown, R. S., & Pauly, M. V. (1986). Estimating hospital costs: A multiple-output analysis. Journal of Health Economics, 5, 107–127.

Hadley, J. (1983). Teaching and hospital costs. Journal of Health Economics, 2, 75–79.

Hadley, J., & Swartz, K. (1989). The impacts on hospital costs between 1980 and 1984 of hospital rate regulation, competition, and changes in health insurance coverage. Inquiry, 26, 35–47.

Harwood, H. J., Hubbard, R. L., Collins, J. J., & Rachal, J. V. (1995). A cost benefit analyses [sic] of drug abuse treatment. Research in Law and Policy Studies, 3, 191–214.

Harwood, H. J., Kallinis, S., & Liu, C. (2001, July). Do larger residential service delivery units have lower costs? (National Evaluation Data Services [NEDS] technical report). Rockville, MD: Substance Abuse and Mental Health Services Administration, Center for Substance Abuse Treatment. [Available as a PDF at]

Holder, H. D., & Blose, J. O. (1991). Typical patterns and cost of alcoholism treatment across a variety of populations and providers. Alcoholism, Clinical and Experimental Research, 15, 190–195.

Hornbrook, M. C., & Monheit, A. C. (1985). The contribution of case-mix severity to the hospital cost-output relation. Inquiry, 22, 259–271.

Huskamp, H. A. (1999). Episodes of mental health and substance abuse treatment under a managed behavioral health care carve-out. Inquiry, 36, 147–161.

Koenig, L., Fields, E. L., Dall, T. M., Ameen, A. Z., & Harwood, H. J. (2000, March). Using case-mix adjustment methods to measure the effectiveness of substance abuse treatment: Three examples using client employment outcomes (National Evaluation Data Services [NEDS] technical report). Rockville, MD: Substance Abuse and Mental Health Services Administration, Center for Substance Abuse Treatment. [Available as a PDF at]

Krenzke, T., & Mohajder, L. (2002, August). Alcohol and Drug Services Study: Cost study imputation final report (Contract No. 4–44U-7652–000). Rockville, MD: Substance Abuse and Mental Health Services Administration, Office of Applied Studies.

Lave, J. R., & Lave, L. B. (1970). Hospital cost functions. American Economic Review, 60, 379–395.

Li, T., & Rosenman, R. (2001). Estimating hospital costs with a generalized Leontief function. Health Economics, 10, 523–538.

Mark, T. L., Coffey, R. M., King, E., Harwood, H., McKusick, D., Genuardi, J., & Das, M. (2000). National estimates of expenditures for mental health and substance abuse treatment, 1997: Volume II technical report appendices (Contract No. 270–96–0007). Rockville, MD: Substance Abuse and Mental Health Services Administration, Center for Substance Abuse Treatment and Center for Mental Health Services.

McCaffrey, D., Bell, R. M., & Botts, C. H. (2001, August 5–9). Generalizations of biased reduced linearization. Paper presented at the Annual Meeting of the American Statistical Association, Atlanta, GA.

McGeary, K. A., French, M. T., Sacks, S., McKendrick, K., & De Leon, G. (2000). Service use and cost by mentally ill chemical abusers: Differences by retention in a therapeutic community. Journal of Substance Abuse, 11, 265–279.

McKay, N. L. (1988). An econometric analysis of costs and scale economies in the nursing home industry. Journal of Human Resources, 23(1), 57–75.

Moos, R. H., Moos, B. S., & Andrassy, J. M. (1999). Outcomes of four treatment approaches in community residential programs for patients with substance use disorders. Psychiatric Services, 50, 1577–1583.

New Jersey Substance Abuse Prevention and Treatment Advisory Task Force. (2001). Improving substance abuse treatment in New Jersey. Trenton, NJ: New Jersey Department of Health and Senior Services. [Available as a PDF download at]

Office of Applied Studies. (2000a). User manual for the ADSS Phase II data files. Rockville, MD: Substance Abuse and Mental Health Services Administration.

Office of Applied Studies. (2000b). Data file user's manual for the ADSS Phase I facility interview file. Rockville, MD: Substance Abuse and Mental Health Services Administration.

Office of Applied Studies. (2000c). Treatment Episode Data Set (TEDS): 1993–1998: National admissions to substance abuse treatment services (DHHS Publication No. SMA 00–3465, Drug and Alcohol Services Information System Series S-11). Rockville, MD: Substance Abuse and Mental Health Services Administration. [Available at]

Office of Applied Studies. (2001). Uniform Facility Data Set (UFDS): 1999 data on substance abuse treatment facilities (DHHS Publication No. SMA 01–3516, Drug and Alcohol Services Information System Series S-13). Rockville, MD: Substance Abuse and Mental Health Services Administration. [Available at]

Office of Applied Studies. (2003, July). Alcohol and Drug Services Study (ADSS) cost study: Costs of substance abuse treatment in the specialty sector (DHHS Publication No. SMA 03–3762, Analytic Series A-20). Rockville, MD: Substance Abuse and Mental Health Services Administration. [Available as a PDF at]

Ouimette, P., Humphreys, K., Moos, R. H., Finney, J. W., Cronkite, R., & Federman, B. (2001). Self-help group participation among substance use disorder patients with posttraumatic stress disorder. Journal of Substance Abuse Treatment, 20, 25–32.

Phibbs, C. S., Swindle, R. W., & Recine, B. (1997). Does case mix matter for substance abuse treatment? A comparison of observed and case mix-adjusted readmission rates for inpatient substance abuse treatment in the Department of Veterans Affairs. Health Services Research, 31, 755–771.

Phillips, C. D., Hubbard, R. L., Dunteman, G., Fountain, D. L., Czechowicz, D., & Cooper, J. R. (1995). Measuring program performance in methadone treatment using in-treatment outcomes: An illustration. Journal of Mental Health Administration, 22, 214–225.

Rogowski, J. A., & Newhouse, J. P. (1992). Estimating the indirect costs of teaching. Journal of Health Economics, 11, 153–171.

Rosko, M. D., & Broyles, R. W. (1988). The economics of health care: A reference handbook. New York: Greenwood Press.

Salkever, D. S., Steinwachs, D. M., & Rupp, A. (1986). Hospital cost and efficiency under per service and per case payment in Maryland: A tale of the carrot and the stick. Inquiry, 23, 56–66.

Salomé, H. J., French, M. T., Scott, C., Foss, M., & Dennis, M. (2003). Investigating variation in the costs and benefits of addiction treatment: Econometric analysis of the Chicago Target Cities Project. Evaluation and Program Planning, 26, 325–338.

Silberberg, E. (1978). The structure of economics: A mathematical analysis. New York: McGraw-Hill.

Sloan, F. A., Feldman, R. D., & Steinwald, A. B. (1983). Effects of teaching on hospital costs. Journal of Health Economics, 2, 1–28.

Sloan, F. A., & Steinwald, B. (1980). Effects of regulation on hospital costs and input use. Journal of Law & Economics, 23(1), 81–109.

Thorpe, K. E. (1988). The use of regression analysis to determine hospital payment: The case of Medicare's indirect teaching adjustment. Inquiry, 25, 219–231.

Troyer, J. L. (2002). Cross-subsidization in nursing homes: Explaining rate differentials among payer types. Southern Economic Journal, 68, 750–773.

U.S. Department of Health and Human Services, Substance Abuse and Mental Health Services Administration. (1996, June 17). Block grant allocation processes. Federal Register, 61(117), 30625–30632. [Available at]

U.S. Department of Housing and Urban Development, Office of Policy Development and Research. (1995, October). Fair market rents for the Section 8 Housing Assistance Payments Program (revised). Retrieved November 5, 2003, from

Vitaliano, D. F. (1987). On the estimation of hospital cost functions. Journal of Health Economics, 6, 305–318.

Welch, W. P. (1987). Do all teaching hospitals deserve an add-on payment under the prospective payment system? Inquiry, 24, 221–232.

Westermeyer, J., Eames, S. L., & Nugent, S. (1998). Comorbid dysthymia and substance disorder: Treatment history and cost. American Journal of Psychiatry, 155, 1556–1560.

Wheeler, J. R., Fadel, H., & D'Aunno, T. A. (1992). Ownership and performance of outpatient substance abuse treatment centers. American Journal of Public Health, 82, 711–718.

Zuckerman, S., Hadley, J., & Iezzoni, L. (1994). Measuring hospital efficiency with frontier cost functions. Journal of Health Economics, 13, 255–280. Comments in 13, 317–334. Discussion in 13, 335–340.

End Notes

1 Examples include the following: Anderson and Lave (1986); Bilodeau, Crémieux, and Ouellette (2000); Carey (1997); Cowing and Holtman (1983); Custer and Wilke (1991); Dor, Duffy, and Wong (1997); Duffy, Ruseski, and Cavanaugh (2000); Escarce and Pauly (1998); Evans (1971); Gaskin and Hadley (1997); Grannemann, Brown, and Pauly (1986); Hadley (1983); Hadley and Swartz (1989); Hornbrook and Monheit (1985); Lave and Lave (1970); Li and Rosenman (2001); McKay (1988); Rogowski and Newhouse (1992); Salkever, Steinwachs, and Rupp (1986); Sloan, Feldman, and Steinwald (1983); Sloan and Steinwald (1980); Thorpe (1988); Troyer (2002); Vitaliano (1987); Welch (1987); and Zuckerman, Hadley, and Iezzoni (1994).

2 Economies of scope are reductions in per unit costs that can be achieved when more than one product is produced. Input substitutability refers to the ease with which one input can be substituted for another in the production process.

3 A shadow unit is a unit that was not part of the original sample, but was identified as a possible replacement if the originally sampled unit declined to participate in the study.

4 We also ran the model including a dummy variable for private, nonprofit ownership to test the hypothesis that government-owned facilities, because they must function under larger bureaucracies, may have higher costs. The results did not indicate a difference between private nonprofit and government-owned facilities, so we collapsed the two categories to conserve degrees of freedom.

5 At least one State, New Jersey, has implemented what it calls an "economies of scale" adjustment. According to a recent report, New Jersey pays 10 percent less per slot to facilities that exceed 40 residential, 150 methadone, or 75 outpatient slots (New Jersey Substance Abuse Prevention and Treatment Advisory Task Force, 2001). However, this type of adjustment does not promote larger facilities. If anything, it promotes the establishment of smaller facilities, which New Jersey may have intended for other reasons.

Go to the Table of Contents

This page was last updated on June 03, 2008.