Chapter 5. Client Choice among Standard Outpatient, Intensive Outpatient, Residential, and Inpatient Alcohol Treatment in State-Monitored Programs

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

Introduction

Treatment for those with substance use disorders has evolved over the years from a largely inpatient to a largely outpatient activity. In the 1970s and early 1980s, treatment providers believed that inpatient was the only acceptable treatment setting because individuals needed to be removed from their environments to overcome their disorders (Washton, 1997). However, several studies conducted in the mid-1980s concluded that outcomes were the same for both treatment settings, and, because outpatient treatment is less costly, it was more cost-effective (Annis, 1985–1986; Miller & Hester, 1986). Those findings, coupled with the growth of managed behavioral health care and the burden on the treatment system caused by the influx of cocaine-addicted clients in the mid- to late 1980s, led treatment to shift from predominantly inpatient to predominantly outpatient settings (Washton, 1997). By October 1, 1998, 89 percent of the almost 1 million individuals in treatment for substance use disorders were in some form of outpatient treatment (Office of Applied Studies [OAS], 2000). By that time, however, there also was a growing recognition that although many clients may not need inpatient treatment, some needed more structure than is provided in the standard outpatient (SOP) settings (Gottheil, 1997). This increased structure could be provided by intensive outpatient (IOP) treatment. By 1998, approximately 20 percent of clients in treatment nationwide were in IOP treatment (OAS, 2000).

In this chapter, we extend the research on treatment-setting choice by examining the association between characteristics of alcohol-abusing clients and admission to State-monitored inpatient hospital (IPH), short- or long-term residential (STR or LTR), IOP, or SOP rehabilitation treatment settings. Earlier studies of the choice between inpatient and outpatient treatment provided evidence that both alcohol disorder severity factors, such as frequency of use, and socioeconomic variables, such as homelessness and employment status, were associated with admission or referral to inpatient versus outpatient treatment (Goodman, Holder, Nishiura, & Hankin, 1992; Goodman, Nishiura, & Hankin, 1998; Gregoire, 2000), but Chapter 4 in this compendium indicates that those relationships varied across States. A study of care authorizations made by a managed behavioral health care organization among inpatient/acute, residential, partial hospitalization, intensive outpatient, and standard outpatient treatment found that age and severity variables were positively associated with referrals to a more intensive treatment (Marques et al., 2001). In this chapter, we study the choice of admission to one of five types of treatment settings among those in the State-monitored treatment sector of two States to examine whether it is appropriate to combine types of inpatient and outpatient treatment into two broad choices for analyses of this population. Then we explore whether the same general findings regarding the associations between client severity and socioeconomic variables and treatment-setting choice are revealed when the determinants of treatment-setting choice are examined in a five-choice model.

To do so, we used data from the 1996 Treatment Episode Data Set (TEDS) for adult males from Iowa and New Jersey with alcohol as the primary substance of abuse. Our findings suggest that it is preferable to examine treatment-setting choice as a five-setting choice rather than collapsing it into an inpatient versus outpatient choice. We find that the key distinctions among clients occur between SOP and all other clients. Those who enter SOP treatment have less severe alcohol disorders and are more likely to be employed than are those who enter any other type of setting.

Treatment Settings

As defined by the American Society of Addiction Medicine's Patient Placement Criteria (ASAM-PPCs), IOP (Level II) falls between traditional SOP (Level I) and residential/inpatient services (Level III) in terms of treatment intensity. For adults, IOP treatment generally involves a structured day or evening treatment program of 9 hours of skilled treatment services provided each week.¹ Such services may include individual and group counseling, family therapy, educational groups, occupational and recreational therapy, psychotherapy, or other therapies (ASAM, 1996). SOP treatment encompasses similar nonresidential services as IOP that are provided in regularly scheduled sessions. However, SOP treatment generally consists of fewer than 9 contact hours per week. Both SOP and IOP treatment are geared toward individuals who do not suffer from severe medical problems and who prefer a modality that fits well with their daily work schedule. However, IOP treatment may be more appropriate for individuals whose emotional conditions distract from recovery, so they need monitoring in excess of that provided under SOP treatment. These individuals may also be resistant to treatment and lack a supportive recovery environment, both of which signal the need for a more structured program.

Residential treatment serves clients who need a safe and stable living environment to develop sufficient recovery skills (ASAM, 1996). It provides organized services by designated treatment personnel who provide a planned regimen of care in a 24–hour setting. Residential treatment can be classified according to intensity (ASAM, 1996) or length of stay. In our data, short-term residential (STR) treatment is defined as 30 days or fewer with long-term residential (LTR) treatment lasting for more than 30 days. Inpatient hospital (IPH) treatment is usually more intensive than residential treatment but with shorter lengths of stay.

Data

As in Chapter 4, we used data from the 1996 Treatment Episode Data Set (TEDS), maintained by the Office of Applied Studies (OAS) of the Substance Abuse and Mental Health Services Administration (SAMHSA) (OAS, 1999). TEDS contains admissions data collected by treatment providers and sent to State agencies, which use them to monitor their substance abuse treatment systems. The data are submitted at regular intervals by the States in a common format to SAMHSA. The data include admissions information—such as the primary and secondary substances of abuse, frequency of use, age at first use, source of referral, number of prior treatment episodes, and planned type of service—as well as mental health and socioeconomic measures. Our analysis focused on adult males with alcohol as their primary substance of abuse. We did not include women because a variable that might be relevant to their treatment setting choice, whether or not they are pregnant, is not well reported.

Unfortunately, all States do not report all variables or collect data from all substance abuse treatment facilities in the TEDS universe (i.e., those receiving public substance abuse treatment funding). Therefore, we focused on two States, Iowa and New Jersey, that differentiated IOP admissions from SOP admissions, provided data covering 90 percent or more of their estimated substance abuse treatment clients in programs receiving public funds in 1996, and collected data from some private facilities on a voluntary basis. These States also collected variables hypothesized to affect alcohol treatment choice.

Empirical Framework

Our empirical specification was motivated by random utility theory, which is commonly used to model choice behavior when the alternatives are naturally discontinuous (Ben-Akiva & Lerman, 1985). Random utility theory assumes that the consumer is rational and that he or she can compare alternatives based on his or her tastes and preferences. The individual's characteristics, which are associated with his or her tastes and preferences, as well as other individual-specific variables that may affect choice, enter an objective function, the utility function, that the individual is assumed to maximize. The individual does that by selecting the alternative that provides him or her with the highest utility. Thus, the coefficients, which can be estimated using maximum likelihood methods, are based on revealed preference exhibited by a sample of individuals.

We assumed that the individual chooses from among five alternative treatment settings: IPH, STR, LTR, IOP, and SOP. Formally, the individual maximizes the following utility function, U_i:

U_i = U_i(IPH, STR, LTR, IOP, SOP),     D  (1)

where

IPH = inpatient hospital,
STR = short-term residential,
LTR = long-term residential,
IOP = intensive outpatient, and
SOP = standard outpatient.

The level of utility that individual i obtains from setting j is a function of the individual's characteristics (X_i):

U_ij = U_j(X_i),      j = (IPH, STR, LTR, IOP, SOP), i = 1,..., N,     D  (2)

where N is the number of individuals in the sample.

Random utility theory assumes that there is a deterministic portion of each of these utilities, which is known with certainty to the individual making the choice, and a random component, which is due to measurement error or some other process that clouds the analyst's ability to fully model the individual's utility. Assuming a linear functional form, the deterministic portion of the utility function for the five setting choices is

_ij = _j X_i,      j = (IPH, STR, LTR, IOP, SOP), i = 1,..., N,     D  (3)

where _ij is the deterministic component of the individual's utility from each choice, X_i is as defined above, and the _j's are choice-specific coefficients on individual characteristics. To obtain the full random utility model for each choice, we added on an error term, _ij, assumed to be distributed jointly according to the extreme value distribution, to yield (4):

U_ij = _ij + _ij.     D  (4)

The probability that individual i will choose treatment setting j is the probability that setting j will bring the individual the greatest utility:

Prob (U_ij > U_ik), for all k j.     D  (5)

Given the distributional assumption noted above, and removing an indeterminacy, we have

 ,     D  (6)

where J is the total number of choices (in our case, five treatment settings) (Greene, 2000). We estimated the model using standard multinomial logit (MNL) techniques. We chose this approach over an ordinal approach based on an underlying latent variable, such as the one used by Marques et al. (2001), because although these treatment settings may vary with regard to the intensity of services, we do not believe that intensity is the only factor clients use in deciding among settings. Furthermore, as has been noted elsewhere, it is not always true that inpatient treatment settings provide the most intensive treatment (Samarasinghe, 1996). For these reasons, we chose the multinomial model, which does not impose an ordering a priori.

We included several individual-specific characteristics in the model, based on prior research on treatment-setting choice (Goodman et al., 1992, 1998; Gregiore, 2000; also see Chapter 4). The variables are similar to those used in Chapter 4, but they have been collapsed as necessary to allow estimation of the MNL model.²

Substance Use Disorder Severity

Previous studies of treatment-setting choice revealed that disorder severity affects the intensity of care received, as would be predicted by models of health services utilization and demand. This suggests that individuals with more severe alcohol disorders may receive greater utility from choosing a more intensive treatment setting. We used several variables to measure the individual's alcohol use disorder severity.

Frequency of Use: We hypothesized that the greater the frequency of alcohol use, the more severe the individual's disorder, and the greater the likelihood that the individual will demand a more intensive treatment setting. We included two dichotomous variables that measure frequency of use in the month prior to admission: used 3 or more times in the week before admission; and used at least once in the month prior, but less than 3 times in the week prior to admission. No use in the past month was the reference category.

Intoxication Before Age 15: Research suggests that individuals who first use alcohol before the age of 15 are more likely to become alcohol dependent (Grant & Dawson, 1997), suggesting that they may demand more intense treatment. We included a dichotomous variable indicating first alcohol intoxication before age 15.

Secondary Substance: Having a secondary substance of abuse can indicate a more severe alcohol use disorder. We included a dichotomous variable that equaled 1 if the client used any secondary substance, 0 otherwise.

Prior Treatment: Prior treatment is used in many treatment studies as another indicator of disorder severity (e.g., Etheridge, Craddock, Hubbard, & Rounds-Bryant, 1999; French & Zarkin, 1992; Hubbard et al., 1989; McLellan et al., 1999; Simpson, Joe, Rowan-Szal, & Greener, 1995). We entered this as a dichotomous variable that equaled 1 if the individual had at least one prior treatment, 0 otherwise.

Co-Occurring Mental Disorders, Homelessness: Based on the ASAM-PPCs and the results of prior research on treatment choice (e.g., Goodman et al., 1992, 1998; also see Chapter 4), we expected that those with a co-occurring mental disorder and those who are homeless may have a higher likelihood of entering more intensive treatment setting. To capture this, we entered two indicator variables, one for co-occurring mental disorder and the other for homelessness.

Socioeconomic and Demographic Characteristics

Several authors have conceptualized addictive behavior as consistent with choice theory and suggested that socioeconomic variables should be included in models examining choice behavior among those with substance use disorders (Becker & Murphy, 1988; Brito & Strain, 1996; Green & Kagel, 1996; Montoya, Atkinson, & Trevino, 2000). We included several variables to measure these characteristics.

Employment Status: Employed individuals have higher time cost associated with inpatient treatment, other things equal, and therefore may gain greater utility from some form of outpatient treatment. Furthermore, the fact that they are employed suggests that they may have a less severe disorder in ways we cannot otherwise measure. Our research on the inpatient/outpatient choice found that employed individuals are less likely to enter inpatient treatment settings (see Chapter 4). To examine whether a similar relationship is uncovered when we consider a more finely divided set of treatment-setting options, we included a dichotomous variable that equaled 1 if the client was employed at admission, either full- or part-time, and 0 for those who were unemployed or not members of the labor force.

Expected Payer: Economic models suggest that individuals who pay out of pocket for their own treatment may demand less costly treatment than those who do not. However, we only observed the expected payer for this particular admission and did not know, for example, whether the client's insurance covered both inpatient and outpatient treatment. In some cases, a payer may limit the individual's choice set if, for example, an indigent client chooses an inpatient treatment program because no publicly funded outpatient slots are available. Therefore, the direction of the expected payer effects is unclear a priori, and the variables should be thought of as control variables. The estimated coefficients may not accurately reflect any direct causal effect of pay source on treatment choice. We included the following three categories: (a) self-pay, (b) private health insurance, and (c) Medicare, Medicaid, and other insurance. Other government funding and no charge formed the reference category. The reference category consisted of clients whose treatment was funded by State agency funds, including those received through the Federal Government's Substance Abuse Prevention and Treatment (SAPT) block grant program.

Education Level: Based on Muurinen (1982), better-educated individuals may be less likely to demand care or may demand less intensive care because, all else equal, the rate of depreciation of the health stock should be lower for better-educated individuals. At the same time, better-educated individuals may have higher income, which may mean they can afford a more intensive treatment choice. We included a dichotomous variable to measure education, which equaled 1 if the client had at least a high school education, 0 otherwise.

Age: Human capital theory suggests that age has a positive effect on an individual's choice of treatment intensity because the rate of depreciation of health stock is a positive function of age (Muurinen, 1982). On the other hand, Suranovic, Goldfarb, and Leonard (1999) suggested that those who are older are more motivated to quit using substances, perhaps making it less likely that they would need intensive treatment. Therefore, we entered age without an a priori prediction on the sign of its impact.

Race/Ethnicity: Research has shown that people of different races and ethnicities have different cultural attitudes about seeking treatment and the types of treatment sought (Arroyo, Westerberg, & Tonigan, 1998; Lundgren, Amodeo, Ferguson, & Davis, 2001; Sheikh & Furnham, 2000). We included three indicator variables to capture the client's race/ethnicity: Hispanic, non-Hispanic black, and other race. Non-Hispanic white was the reference category.

Marital Status: Although imperfect, marital status may proxy for the ASAM-PPCs' emotional and behavioral disorders criteria. Currently married clients may have less severe emotional and behavioral disorders than those who are single, divorced, separated, or widowed. Marital status also may proxy a more stable living environment, more family responsibilities, or more family support, all of which may make it less likely that a married individual would enter the inpatient or residential settings. The variable equaled 1 if the individual was currently married, 0 otherwise.

Referral Source: An individual's referral source may influence his choice of treatment setting and the utility he derives from that choice. The referral may help the individual choose a treatment setting by providing him with information about the best treatment choice for his needs; or the referral source may limit the individual's choice set, causing him to choose from among the remaining options.

Clients who self-refer into alcohol treatment may do so more because of life problems associated with their use of alcohol than because of their alcohol use (Majella Jordan & Oei, 1989). These individuals may be more likely to enter a program based on convenience than on their clinical needs. Likewise, clients referred by third parties, such as their schools, employers, or physicians, also may choose a program based more on convenience than on clinical factors because they may be entering treatment to appease their referrer. On the other hand, clients referred by an alcohol or drug treatment provider, who may have knowledge of clinical placement criteria, may be more likely than others to be referred to the setting most consistent with their needs.

Finally, the criminal justice system has became a frequent source of referral into alcohol treatment. The likelihood that clients referred in this way will choose treatment consistent with their needs depends, in part, on the placement criteria used by the referring criminal justice system (CJS). Clients with a criminal justice referral may face a limited choice set. For example, they may face a simple choice between the treatment setting dictated by the criminal justice agency or incarceration. However, in our data a CJS referral also may occur for clients with other CJS involvement, such as being on parole, on work or home furlough, or for a civil commitment. Therefore, having a CJS referral does not necessarily mean that treatment referred was required in lieu of prison. The client may still face a choice among multiple treatment settings. We included three dichotomous variables that indicated whether the client was referred by an alcohol or drug abuse treatment provider; another medical care provider; his employer or school; or was self-referred. CJS referral was the reference category.

Season of Admission: The time of year the client is admitted also may affect the likelihood that an individual chooses inpatient treatment. For example, those who are seasonally employed, such as teachers and college professors, may be more likely to choose inpatient treatment in the summer. We included indicator variables for summer, fall, and winter, with spring as the reference category.

Specification Tests

Because most of the literature on treatment-setting choice examined a simple dichotomous choice between inpatient and outpatient treatment, we first tested whether combining the treatment settings into those two broad categories was supported by the data. To do so, we ran a series of likelihood ratio chi-square tests to determine whether the coefficient vectors for each possible pair of treatment-setting choices were significantly different. The results appear below.

We then tested whether the multinomial logit (MNL) model was an appropriate way to estimate our model. MNL is a popular way to estimate polychotomous choice models, such as the one we estimate here, because of its relative ease of estimation and interpretation. However, use of the MNL also imposes the rather restrictive "irrelevance of independent alternatives" (IIA) assumption. Behaviorally, IIA implies that the ratio of the utility levels between two choices, say IPH and SOP, remain constant regardless of how many other choices there are. We implemented two tests that are frequently used to test this assumption—the Hausman and Small-Hsiao tests. The results appear below.

Results

Descriptive Statistics

Table 5.1 displays the mean values and proportions by State of the variables included in our analysis. It shows that there are several statistically significant differences in the variable means and proportions between the two States. To determine statistical significance, we conducted simple t tests of the means for the continuous variables and chi-square tests for differences in proportions for the dichotomous and categorical variables. Unless otherwise indicated, differences discussed in the text are significant at or better than the 5 percent level. As Table 5.1 shows, we found that statistically significant differences existed between the Iowa and New Jersey clients for all variables except secondary substance use and using drugs prior to 15 years of age.

In both States, the majority of clients entered SOP, although the proportion was somewhat higher in Iowa (71 percent) than New Jersey (59 percent). Clients next most frequently entered IOP, 16 percent in Iowa compared with almost 20 percent in New Jersey. STR was the next most frequent setting, accounting for 8.7 percent of the admissions in Iowa and almost 10 percent in New Jersey. LTR was the fourth most frequent treatment setting for Iowa clients, while it was the fifth most frequent for New Jersey clients. IPH, on the other hand, was the fifth most frequent for Iowa clients, and the fourth most frequent for New Jersey clients.

Turning to the explanatory variables, Iowa and New Jersey clients differed on measures of disorder severity, although not always in the same direction. For example, in New Jersey almost 48 percent of the clients had used alcohol more than 3 times in the week prior to admission compared with only 26 percent of the clients in Iowa. Almost 39 percent of the clients in Iowa had no drug use in the week prior to admission compared with fewer than 22 percent of the clients in New Jersey. However, in Iowa almost 59 percent of the clients had at least one prior treatment episode compared with fewer than 50 percent in New Jersey. Clients in both States were similarly likely to have a secondary drug of abuse (approximately 41 percent) and to have become intoxicated prior to age 15 (approximately 34 percent). Fewer than 9 percent of clients in New Jersey indicated a mental disorder compared with almost 16 percent of clients in Iowa. Homelessness also varied between the two States, with fewer than 2 percent of clients in Iowa being homeless at treatment admission compared with almost 5 percent of clients in New Jersey.

Differences between the two States also existed among the socioeconomic and demographic variables. Almost 66 percent of clients in Iowa were employed at treatment admission compared with only 56 percent of clients in New Jersey. About 30 percent of clients in New Jersey paid for their own treatment (i.e., self-paid) compared with fewer than 7 percent in Iowa. Other government/no pay was the overwhelming payment form for clients in Iowa, accounting for 70 percent, compared with only about 39 percent of clients in New Jersey. More consistent across States was the proportion of clients whose expected payer was either private health insurance (16 percent in Iowa and 18.4 percent in New Jersey) and or another payer (7.2 percent in Iowa and 10.5 percent in New Jersey).

Table 5.1 Variable Means for Iowa and New Jersey

Variable	Iowa	New Jersey
Number of Observations	17,495	10,151
Treatment Setting (dependent variable)***
Inpatient hospital (IPH)	0.015	0.065
	(0.123)	(0.247)
Short-term residential (STR)	0.087	0.099
	(0.282)	(0.299)
Long-term residential (LTR)	0.025	0.049
	(0.156)	(0.216)
Intensive outpatient (IOP)	0.160	0.199
	(0.366)	(0.399)
Standard outpatient (SOP)	0.713	0.588
	(0.452)	(0.492)
Alcohol Use Prior to Admission***
Used at least 3 times in the week before admission	0.261	0.475
	(0.439)	(0.499)
Used at least 1 time in the month before admission	0.353	0.308
	(0.478)	(0.462)
No use in the past month	0.386	0.218
	(0.487)	(0.413)
Had a secondary substance of abuse	0.407	0.415
	(0.491)	(0.493)
Intoxicated prior to 15 years of age	0.343	0.338
	(0.475)	(0.473)
Had at least one prior treatment episode***	0.585	0.498
	(0.493)	(0.500)
Existing mental disorder***	0.159	0.087
	(0.365)	(0.282)
Homeless at treatment admission***	0.014	0.046
	(0.117)	(0.210)
Employed at treatment admission***	0.655	0.560
	(0.475)	(0.496)
Expected Form of Payment***
Self-pay	0.066	0.321
	(0.248)	(0.467)
Private health insurance	0.160	0.184
	(0.367)	(0.388)
Other pay (e.g., Medicaid, Medicare, worker's compensation)	0.072	0.105
	(0.259)	(0.307)
Other government pay, no charge	0.702	0.389
	(0.457)	(0.488)
High school graduate***	0.796	0.733
	(0.403)	(0.442)
Age of Respondent***	34.292	36.131
	(10.674)	(10.425)
Race/Ethnicity***
Non-Hispanic white	0.905	0.6824
	(0.294)	(0.466)
Non-Hispanic black	0.043	0.199
	(0.204)	(0.399)
Hispanic	0.034	0.105
	(0.181)	(0.306)
Other race	0.018	0.014
	(0.132)	(0.119)
Married***	0.338	0.250
	(0.473)	(0.433)
Referral Source to Treatment***
Self-referred	0.177	0.194
	(0.382)	(0.395)
Alcohol/drug treatment provider	0.073	0.099
	(0.260)	(0.299)
Other (employer, school, physician)	0.114	0.175
	(0.318)	(0.380)
Criminal justice referral	0.636	0.532
	(0.481)	(0.499)
Season of Admission***
Spring	0.323	0.258
	(0.468)	(0.438)
Summer	0.189	0.258
	(0.392)	(0.438)
Fall	0.188	0.231
	(0.390)	(0.421)
Winter	0.300	0.253
	(0.458)	(0.435)
*** Statistically significant at the 0.01 level. ** Statistically significant at the 0.05 level. Note: Standard deviations are in parentheses; t tests and chi-square tests were used where appropriate. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

About 91 percent of clients in Iowa were non-Hispanic white compared with only 68 percent of clients in New Jersey. Fewer than 5 percent of clients in Iowa were non-Hispanic black compared with almost 20 percent of clients in New Jersey. A minority of clients in both States were currently married, about 34 percent of clients in Iowa compared with 25 percent in New Jersey, but most clients in both States had at least a high school education (about 80 percent in Iowa and 73 percent in New Jersey).

The criminal justice system (CJS) was the most common route of treatment referral for clients in both States although a significantly greater proportion of clients in Iowa were CJS-referred compared with New Jersey (64 percent in Iowa vs. 53 percent in New Jersey). The second most common route for both States was self-referral, which was the route to treatment for almost 18 percent of clients in Iowa and more than 19 percent of those in New Jersey. A significantly greater proportion of clients in New Jersey were referred by the next most common referral category, referral by the client's employer, school, or health care provider other than an alcohol or drug treatment provider, compared with Iowa (almost 18 vs. 11 percent), while the least common route in both States was an alcohol or drug treatment provider (7.3 percent in Iowa and 9.9 percent in New Jersey).

Finally, season of admission appeared to differ somewhat across the States. Although New Jersey had a fairly even distribution of clients admitted across the four seasons, about 25 percent in each season, more than 60 percent of clients in Iowa were admitted in winter and spring compared with fewer than 40 percent in fall and summer.

Specification Tests

Pooling

The results of the likelihood ratio tests appear in Table 5.2. These tests reveal that the null hypothesis of no difference in the coefficients for each pair can be rejected at better than the 1 percent level. Thus, we ran our MNL model with all five treatment settings as distinct outcomes.

IIA Tests

Unfortunately, the two tests that are commonly used to test for IIA, the Small-Hsiao and Hausman tests, frequently arrive at different conclusions. Table 5.3 shows that was the case when we tested our data. Therefore, the results of the IIA tests were inconclusive. We explored using a multinomial probit model, which also can be used for multinomial choice estimation and does not suffer from IIA, but which can be difficult to estimate with more than three choices unless other restrictive assumptions are imposed. We attempted to run the model for both States using assumptions (standard deviation = 1; correlation = 0.5) that are far from those implied by the MNL. Although we could not get the New Jersey model to converge, we did successfully estimate the Iowa model. Because the results from that model were similar qualitatively to the MNL Iowa model, and because we did not have unequivocal information that the IIA was violated by our data, we believe the MNL results are reasonable and report them here.

Table 5.2 Likelihood Ratio Test for Pooling

Categories Tested		Chi-Square	Degrees of Freedom	p > Chi-Square
IPH	STR	361.700	23	0.000
IPH	LTR	560.325	23	0.000
IPH	IOP	471.707	23	0.000
IPH	SOP	890.473	23	0.000
STR	LTR	425.733	23	0.000
STR	IOP	1,096.722	23	0.000
STR	SOP	2,514.803	23	0.000
LTR	IOP	1,074.206	23	0.000
LTR	SOP	1,308.670	23	0.000
IOP	SOP	1,009.398	23	0.000
IOP = intensive outpatient. IPH = inpatient hospital. LTR = long-term residential. SOP = standard outpatient. STR = short-term residential. Null hypothesis: All coefficients expect intercepts associated with given pair of outcomes are 0 (i.e., categories can be collapsed). Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

Table 5.3 Hausman and Small-Hsiao Tests of Irrelevance of Independent Alternatives (IIA)
Assumption: Iowa and New Jersey

Category Omitted	Hausman Results		Small-Hsiao Results
	Chi-Square Value	p Value	Chi-Square Value	p Value
Iowa
Inpatient hospital (IPH)	-5.136	-----	78.490	0.000
Short-term residential (STR)	25.112	1.000	76.239	0.000
Long-term residential (LTR)	-32.088	-----	75.345	0.000
Intensive outpatient (IOP)	-26.465	-----	79.938	0.000
New Jersey
Inpatient hospital (IPH)	-55.732	-----	74.294	0.000
Short-term residential (STR)	-51.492	-----	76.996	0.000
Long-term residential (LTR)	8.772	1.000	67.746	0.000
Intensive outpatient (IOP)	-72.283	-----	103.954	0.000
Null hypothesis: Odds (Outcome-J vs. Outcome-K) are independent of other alternatives. Note: If chi-square < 0, the estimated model does not meet asymptotic assumptions of the test. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

Multivariate Results

Tables 5.4 through 5.7 present the results from our MNL estimation for Iowa and New Jersey. Estimated coefficients from MNL models can be difficult to interpret because the sign of the coefficient may not be equivalent to the change in the probability due to a change in the independent variable. Therefore, in Tables 5.4 and 5.5, we present the estimated marginal effects (me) of each explanatory variable (x_i) on each alternative j. The marginal effects were estimated at the means of the independent variables, and their standard errors were computed using the Delta Method. To further aid in interpretation, we used those marginal effects to compute the percentage change in the predicted probability of each outcome as a function of each independent variable. Those results, which we discuss in the text, appear in Tables 5.6 and 5.7. We present the results separately for the two States.

Iowa

Severity Variables. Table 5.6 reveals that several of the variables measuring the alcohol use disorder severity had significant effects on treatment-setting admission among clients in Iowa. As might be expected based on previous research, clients with greater severity prior to treatment entry (as measured by frequency of alcohol use) were more likely to enter what are usually thought of as more intensive treatment settings. The main distinction, however, appears to be between SOP and all other settings. Admission to SOP treatment was generally associated with lower levels of severity. For example, having used alcohol 3 or more times in the week before entering treatment decreased the probability of entering SOP compared with not having used in the month before treatment. The marginal effect of -0.343 (see Table 5.4) translates into approximately a 44 percent decline in the probability of entering SOP.³ The marginal effect of -0.102 of use in the past month translates into a 13.2 percent lower probability of entering SOP compared with not having used in the past month. Intoxication prior to age 15 reduced the probability of SOP admission by about 3 percent compared with having begun using later. Those with at least one prior treatment had a 4.8 percent lower probability of entering SOP than did those with no prior treatment. Being homeless at admission led to a 36 percent lower probability of entering the SOP setting.

In contrast, most severity variables generally were positively associated with entry into settings more likely to provide more intensive treatment. For example, those who drank more than 3 times in the week before treatment had a 55.2 percent higher probability of entering IOP treatment than those who had not drunk in the month before treatment, and those who drank at least once in the month before treatment had a 28.6 percent higher probability of entering IOP treatment. Likewise, those who became intoxicated prior to age 15 had an 8.2 percent higher probability of entering IOP treatment than those who started drinking later.

Table 5.4 Marginal Effects for the Choice of Substance Abuse Treatment Setting: Iowa

Variable	Inpatient Hospital (IPH)	Short-Term Residential (STR)	Long-Term Residential (LTR)	Intensive Outpatient (IOP)	Standard Outpatient (SOP)
Number of Observations	270	1,518	437	2,791	12,479
Alcohol Use Prior to Admission
Used at least 3 times in the week before admission	0.023***	0.224***	0.007***	0.089***	-0.343***
	(0.004)	(0.011)	(0.002)	(0.009)	(0.012)
Used at least 1 time in the month before admission	0.006***	0.052***	-0.003**	0.046***	-0.102***
	(0.002)	(0.006)	(0.001)	(0.008)	(0.009)
Intoxicated prior to 15 years of age	-4.86E-4	0.009**	0.002	0.013**	-0.023***
	(0.001)	(0.003)	(0.001)	(0.007)	(0.007)
Had a secondary substance of abuse	-1.38E-4	-0.007**	0.001	-1.48E-4	0.006
	(0.001)	(0.003)	(0.001)	(0.006)	(0.007)
Had at least one prior treatment episode	-0.001	0.015***	0.011***	0.012	-0.037***
	(0.001)	(0.003)	(0.001)	(0.006)	(0.007)
Existing mental disorder	0.004***	0.004	0.003**	-0.011	2.43E-4
	(0.001)	(0.004)	(0.001)	(0.008)	(0.009)
Homeless at admission	0.012**	0.158***	0.043***	0.066	-0.279***
	(0.005)	(0.029)	(0.010)	(0.038)	(0.047)
Employed at admission	-0.003***	-0.060***	-0.015***	-0.028***	0.105***
	(0.001)	(0.004)	(0.002)	(0.007)	(0.008)
Expected Form of Payment
Self-pay	0.013***	0.009	-0.004***	0.164***	-0.183***
	(0.003)	(0.006)	(0.001)	(0.016)	(0.016)
Private health insurance	0.010***	-0.013***	-0.007***	0.180***	-0.169***
	(0.002)	(0.004)	(0.001)	(0.011)	(0.012)
Other pay (e.g., Medicaid, Medicare, worker's         compensation)	0.012***	-0.012***	-0.007***	0.014	-0.007
	(0.003)	(0.004)	(0.001)	(0.013)	(0.014)
High school graduate	-0.002***	-0.003	-0.002	-0.001	0.009
	(0.001)	(0.004)	(0.001)	(0.008)	(0.009)
Age of respondent	1.10E-4	-2.43E-4	3.17E-5	1.43E-6	1.01E-4
	(3.00E-5)	(1.50E-4)	(5.00E-5)	(3.00E-4)	(3.40E-4)
Race/Ethnicity
Non-Hispanic black	-0.002	0.032***	-3.12E-4	0.068***	-0.098***
	(0.001)	(0.009)	(0.002)	(0.017)	(0.019)
Hispanic	0.004	-4.80E-4	-0.001	0.012	-0.015
	(0.003)	(0.009)	(0.003)	(0.017)	(0.019)
Other race	0.004	0.068***	0.003	-0.005	-0.069**
Married	-0.003***	-0.016***	-0.006***	-0.016***	0.041***
	(0.001)	(0.003)	(0.001)	(0.006)	(0.007)
Referral Source to Treatment
Self-referred	0.002**	0.006	0.011***	-0.021***	0.002
	(0.001)	(0.004)	(0.003)	(0.008)	(0.009)
Alcohol/drug treatment provider	0.002	0.089***	0.067***	-0.071***	-0.087***
	(0.002)	(0.010)	(0.009)	(0.010)	(0.016)
Other (employer, school, physician)	0.003***	0.005	0.006**	-0.024***	0.009
	(0.001)	(0.005)	(0.002)	(0.009)	(0.011)
Season of Admission
Summer	-0.001	0.023***	-0.002	0.023**	-0.043***
	(0.001)	(0.005)	(0.001)	(0.009)	(0.010)
Fall	-0.002***	0.015***	-0.002	0.009	-0.021**
	(0.001)	(0.005)	(0.001)	(0.009)	(0.010)
Winter	0.002***	0.007	0.002	0.010	-0.022**
	(0.001)	(0.004)	(0.001)	(0.008)	(0.009)
*** Statistically significant at the 0.01 level. ** Statistically significant at the 0.05 level. LR chi-square: 5028.39. Prob > chi-square: 0.0000. Pseudo R-square: 0.1592. Note: Standard errors are in parentheses. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

Table 5.5 Marginal Effects for the Choice of Substance Abuse Treatment Setting: New Jersey

Variable	Inpatient Hospital (IPH)	Short-Term Residential (STR)	Long-Term Residential (LTR)	Intensive Outpatient (IOP)	Standard Outpatient (SOP)
Number of Observations	663	1,006	498	2,018	5,966
Alcohol Use Prior to Admission
Used at least 3 times in the week before admission	0.061***	0.047***	0.004***	0.170***	-0.281***
	(0.009)	(0.007)	(0.001)	(0.014)	(0.015)
Used at least 1 time in the month before admission	0.026***	-0.043***	-0.003	0.082***	-0.062***
	(0.008)	(0.007)	(0.001)	(0.017)	(0.018)
Intoxicated prior to 15 years of age	2.35E-4	0.002	0.001	0.024**	-0.028**
	(0.002)	(0.005)	(0.001)	(0.010)	(0.012)
Had a secondary substance of abuse	6.42E-5	0.031***	0.002**	0.075***	-0.108***
	(0.002)	(0.005)	(0.001)	(0.011)	(0.012)
Had at least one prior treatment episode	0.009***	-0.004	0.005***	0.034***	-0.044***
	(0.002)	(0.005)	(0.001)	(0.010)	(0.011)
Existing mental disorder	0.084***	0.025***	-0.004***	-0.078***	-0.028
	(0.011)	(0.009)	(0.001)	(0.015)	(0.022)
Homeless at admission	0.016**	2.04E-4	0.083***	-0.132***	0.032
	(0.007)	(0.011)	(0.018)	(0.022)	(0.035)
Employed at admission	-0.016***	-0.051***	-0.011***	-0.079***	0.157***
	(0.003)	(0.006)	(0.002)	(0.011)	(0.013)
Expected Form of Payment
Self-pay	0.003	-0.093***	-0.006***	-0.088***	0.184***
	(0.003)	(0.006)	(0.001)	(0.012)	(0.013)
Private health insurance	0.034***	-0.017***	-0.007***	0.142***	-0.153***
	(0.006)	(0.005)	(0.001)	(0.017)	(0.019)
Other pay (e.g., Medicaid, Medicare, worker's         compensation)	0.038***	-0.033***	-0.007***	0.079***	-0.078***
	(0.007)	(0.004)	(0.001)	(0.018)	(0.020)
High school graduate	0.002	-0.014***	-0.001	0.018	-0.006
	(0.002)	(0.005)	(0.001)	(0.011)	(0.013)
Age of respondent	4.66E-5	-0.001***	-5.25E-6	-2.91E-5	0.001
	(8.00E-5)	(2.40E-4)	(4.00E-5)	(0.001)	(0.001)
Race/Ethnicity
Non-Hispanic black	-0.007***	-0.018***	0.001	0.029**	-0.005
	(0.002)	(0.005)	(0.001)	(0.013)	(0.014)
Hispanic	-0.010***	0.020**	1.98E-4	-0.051***	0.041**
	(0.002)	(0.009)	(0.001)	(0.016)	(0.018)
Other race	-0.001	0.014	0.005	-0.104***	0.086**
	(0.007)	(0.024)	(0.006)	(0.033)	(0.042)
Married	-0.003	0.006	-0.004***	-1.30E-4	0.001
	(0.002)	(0.006)	(0.001)	(0.012)	(0.013)
Referral Source to Treatment
Self-referred	0.045***	0.058***	0.009***	0.056***	-0.167***
	(0.007)	(0.010)	(0.002)	(0.014)	(0.017)
Alcohol/drug treatment provider	-0.003	0.163***	0.047***	0.039**	-0.245***
	(0.003)	(0.020)	(0.009)	(0.019)	(0.024)
Other (employer, school, physician)	0.019***	0.063***	-0.001	0.080***	-0.161***
	(0.005)	(0.010)	(0.001)	(0.015)	(0.017)
Season of Admission
Summer	-0.003	0.016**	0.001	-0.009	-0.005
	(0.002)	(0.007)	(0.001)	(0.013)	(0.015)
Fall	-0.009***	0.022***	0.001	-0.005	-0.009
	(0.002)	(0.008)	(0.001)	(0.014)	(0.016)
Winter	-0.006***	0.020***	0.001	0.019	-0.035**
	(0.002)	(0.007)	(0.001)	(0.013)	(0.015)
*** Statistically significant at the 0.01 level. ** Statistically significant at the 0.05 level. LR chi-square: 7,326.61. Prob > chi-square: 0.0000. Pseudo R-square: 0.3036. Note: Standard errors are in parentheses. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

Table 5.6 Marginal Effects as a Proportion of the Predicted Probability of Each Outcome: Iowa

Variable	Inpatient Hospital (IPH)	Short-Term Residential (STR)	Long-Term Residential (LTR)	Intensive Outpatient (IOP)	Standard Outpatient (SOP)
Number of Observations	0.004	0.053	0.009	0.162	0.772
Alcohol Use Prior to Admission
Used at least 3 times in the week before admission	5.628***	4.234***	0.760***	0.552***	-0.444***
Used at least 1 time in the month before admission	1.456***	0.989***	-0.285**	0.286***	-0.132***
Intoxicated prior to 15 years of age	-0.122	0.161**	0.198	0.082**	-0.030***
Had a secondary substance of abuse	-0.035	-0.133**	0.096	-0.001	0.008
Had at least one prior treatment episode	-0.152	0.283***	1.277***	0.071	-0.048***
Existing mental disorder	1.011***	0.072	0.314**	-0.067	0.000
Homeless at admission	3.104**	2.977***	4.729***	0.406	-0.361***
Employed at admission	-0.644***	-1.123***	-1.672***	-0.172***	0.136***
Expected Form of Payment
Self-pay	3.267***	0.172	-0.406***	1.013***	-0.237***
Private health insurance	2.456***	-0.249***	-0.779***	1.110***	-0.219***
Other pay (e.g., Medicaid, Medicare, worker's         compensation)	2.922***	-0.230***	-0.732***	0.088	-0.009
High school graduate	-0.589***	-0.065	-0.205	-0.007	0.011
Age of respondent	0.027	-0.005	0.004	0.000	0.000
Race/Ethnicity
Non-Hispanic black	-0.405	0.606***	-0.035	0.419***	-0.127***
Hispanic	0.901	-0.009	-0.063	0.075	-0.019
Other race	0.897	1.275***	0.280	-0.030	-0.089**
Married	-0.627***	-0.309***	-0.644***	-0.102***	0.053***
Referral Source to Treatment
Self-referred	0.608**	0.105	1.220***	-0.128***	0.002
Alcohol/drug treatment provider	0.411	1.678***	7.478***	-0.438***	-0.113***
Other (employer, school, physician)	0.819***	0.100	0.665**	-0.147***	0.012
Season of Admission
Summer	-0.271	0.434***	-0.182	0.141**	-0.056***
Fall	-0.551***	0.290***	-0.177	0.057	-0.027**
Winter	0.624***	0.135	0.256	0.062	-0.028**
*** Statistically significant at the 0.01 level. ** Statistically significant at the 0.05 level. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

Table 5.7 Marginal Effects as a Proportion of the Predicted Probability of Each Outcome: New Jersey

Variable	Inpatient Hospital (IPH)	Short-Term Residential (STR)	Long-Term Residential (LTR)	Intensive Outpatient (IOP)	Standard Outpatient (SOP)
Number of Observations	0.016	0.061	0.006	0.219	0.699
Alcohol Use Prior to Admission
Used at least 3 times in the week before admission	3.820***	0.770***	0.724***	0.777***	-0.402***
Used at least 1 time in the month before admission	1.617***	-0.706***	-0.472	0.375***	-0.089***
Intoxicated prior to 15 years of age	0.015	0.040	0.166	0.111**	-0.040**
Had a secondary substance of abuse	0.004	0.505***	0.346**	0.343***	-0.154***
Had at least one prior treatment episode	0.592***	-0.068	0.973***	0.154***	-0.064***
Existing mental disorder	5.307***	0.418***	-0.811***	-0.354***	-0.039
Homeless at admission	0.998**	0.003	15.062***	-0.601***	0.046
Employed at admission	-1.028***	-0.835***	-1.977***	-0.361***	0.225***
Expected Form of Payment
Self-pay	0.162	-1.535***	-1.035***	-0.400***	0.263***
Private health insurance	2.167***	-0.279***	-1.288***	0.650***	-0.218***
Other pay (e.g., Medicaid, Medicare, worker's         compensation)	2.374***	-0.539***	-1.225***	0.363***	-0.111***
High school graduate	0.157	-0.230***	-0.106	0.081	-0.008
Age of respondent	0.003	-0.014***	-0.001	0.000	0.001
Race/Ethnicity
Non-Hispanic black	-0.437***	-0.290***	0.106	0.132**	-0.007
Hispanic	-0.623***	0.324**	0.036	-0.234***	0.059**
Other race	-0.062	0.238	0.950	-0.476***	0.122**
Married	-0.185	0.097	-0.702***	-0.001	0.001
Referral Source to Treatment
Self-referred	2.836***	0.958***	1.551***	0.255***	-0.240***
Alcohol/drug treatment provider	-0.210	2.681***	8.454***	0.180**	-0.351***
Other (employer, school, physician)	1.216***	1.045***	-0.208	0.365***	-0.231***
Season of Admission
Summer	-0.211	0.271**	0.258	-0.043	-0.007
Fall	-0.547***	0.367***	0.123	-0.023	-0.013
Winter	-0.367***	0.328***	0.266	0.087	-0.049**
*** Statistically significant at the 0.01 level. ** Statistically significant at the 0.05 level. Source: SAMHSA, Office of Applied Studies, Treatment Episode Data Set, 1996.

The proportionate effect of a change in severity measures on the probability of entry into LTR treatment was not always as great as it was for IOP. Although using alcohol 3 or more times in the week before admission increased the probability of entry into LTR by 76 percent compared with 55 percent for IOP, use in the past month actually decreased the probability of entering LTR treatment by about 29 percent. Intoxication prior to age 15 did not significantly affect the probability of entry into LTR, although it had for IOP. Other measures of severity, however, did have a greater proportional effect on the probability of entering into LTR than they did on admission to IOP. A prior treatment episode, for example, increased the probability of entering LTR by 128 percent, being homeless increased it by 473 percent, and having an existing mental disorder increased it by 31 percent. None of these three variables affected the probability of entry into IOP.

Severity measures had a more mixed effect on the probability of admission to STR treatment. On the one hand, the frequency of use variables had large impacts on the probability of admission. Use 3 or more times in the week before admission increased the probability of admission to STR by 423 percent, which is an enormous impact compared with its effect on LTR. Use at least one time in the past month increased the probability of admission to STR about 99 percent compared with a decrease of 28.5 percent for LTR. Unlike LTR, having been intoxicated prior to age 15 increased the probability of STR admission by 16.1 percent. Having had prior treatment also increased the probability of admission to STR, although by a smaller 28.3 percent compared with 128 percent for LTR. Those who were homeless had an increased probability of admission to STR of about 298 percent compared with those who were not homeless, which was somewhat smaller than the effect on the probability of admission to LTR. On the other hand, those with a secondary substance of abuse had a 13.3 percent lower probability of entering STR than did those without a secondary substance of abuse. Secondary substance use was not significantly associated with the probability of entering LTR. A client having a mental disorder had no higher or lower probability of entering STR than did someone without a mental disorder, in contrast to LTR where having a mental disorder did increase the probability of admission.

As for the IPH setting, frequency of use again affected the probability of admission, and the effects were greater than they were in STR. Use at least 3 times in the week prior to treatment increased the probability of IPH by about 563 percent compared with no use, which was somewhat larger than its effect on the probability of entry into STR. Use at least once in the past month increased the probability of IPH by about 146 percent compared with no use. Both of these magnitudes were larger than those for STR and LTR. However, although both intoxication before age 15 and having had a prior treatment episode increased the probability of STR admission, neither affected the probability of IPH admission. Homeless clients had a 310 percent higher probability of IPH treatment compared with those who were not homeless, which was slightly larger than the effect of homelessness on the probability than it was for STR, but smaller than that for LTR. Those having a mental disorder at admission, on the other hand, had a 101 percent greater probability of IPH treatment than did those without one, which was larger than its impact on the probability of admission to STR or LTR.

Socioeconomic and Demographic Variables. A distinction between SOP and all other treatment settings appears again with respect to employment. Clients who were employed at admission had a significantly higher probability of entering SOP treatment than did unemployed clients, and a significantly lower probability of entering any of the more intensive settings. Employment had the largest proportional negative impact on entry to LTR and the smallest on entry to IOP.

There also were differences among the effects of expected source of payment on the probability of admission to each setting, but they were not as might be expected. As mentioned earlier, the estimated coefficients may not accurately measure the effect of expected payer on the probability of entry because the two may be codetermined. These variables should be thought of as control variables. We found that self-payment was associated with a 23.7 percent lower probability of entering SOP, the least costly option, than was the reference group, charity care or other government assistance (excluding Medicaid or Medicare). Self-payment also was associated with a more than 100 percent higher probability of entering IOP, but a 40.6 percent lower probability of entering LTR. We found similar results for private health insurance and other sources of payments (e.g., Medicaid, Medicare, worker's compensation). Compared with the reference cell, private insurance coverage was associated with a lower probability of entering SOP and a higher probability of entering IOP or IPH. However, in contrast to self-payment, private insurance coverage was associated with a lower probability of entering either residential setting compared with the reference cell. Other payment source was likewise positively associated with admission to IPH and a lower probability of admission to either type of residential setting. However, its association with entry into either type of outpatient setting was no different statistically than the reference cell.

We found some evidence that the SOP setting in Iowa was further distinguished from the other settings with regard to the race/ethnicity variables. We found that non-Hispanic black clients in Iowa had a 41.9 percent higher probability of entering IOP and a 60.6 percent higher probability of entering STR than non-Hispanic white clients, but non-Hispanic black clients had a 12.7 percent lower probability of entering SOP compared with non-Hispanic white clients. A similar pattern emerged for those of other race when compared with non-Hispanic whites. Those of other race had an 8.9 percent lower probability of entering SOP, but a 127.5 percent higher probability of entering STR than did non-Hispanic whites. However, the probability that Hispanics would enter any of the treatment settings was not different from that of whites.

A distinction between SOP and all other treatment settings existed in Iowa for marital status as well. Married clients had a 5.3 percent greater probability of entering SOP than did those not currently married and a lower probability of entering any other treatment setting. The proportional effect on these probabilities was greatest for LTR and smallest for IOP.

Referral Source and Season of Admission. Referral source also was associated with treatment-setting choice in Iowa, although this time the pattern was somewhat different. Rather than having the demarcation at SOP as with most of the other variables, it appeared between IOP treatment and the inpatient treatments. Compared with those referred by the CJS, those who were referred by any other source had a lower probability of entering the IOP setting. However, those with any other referral source had a higher probability of entering one of the inpatient settings, especially LTR.

Finally, we found that season of admission also affected choice in Iowa with clients entering treatment in summer, fall, or winter months being less likely to enter SOP than clients entering treatment during the spring.

New Jersey

Severity Variables. As in Iowa, the main distinction with respect to most severity variables appears to have been between clients admitted to SOP compared with those admitted to all other treatment settings (Table 5.7). Again, those with higher severity levels had lower probabilities of entering SOP. Those who drank alcohol 3 or more times in the week before admission were 40.2 percent less likely to enter SOP than were those who did not drink in the month prior to treatment. Those who drank at least once in the past month were 8.9 percent less likely to enter SOP. Secondary substance use, intoxication prior to age 15, and having at least one prior treatment episode also decreased the probability of entering SOP by 15.4, 4.0, and 6.4 percent, respectively, compared with their reference categories. However, those who were homeless at admission or had a mental disorder were no more or less likely to enter an SOP setting in New Jersey than were those who did not.

On the other hand, those who drank alcohol 3 or more times in the week before treatment had a 77.7 percent higher probability of entering IOP treatment than those who had not drunk at all. Those who drank at least once in the month before treatment had a 37.5 percent higher probability of entering IOP than those who did not drink in that time period. Those with secondary drug use, who first became intoxicated before age 15, or who had at least one prior treatment episode also had higher probability of admission to IOP treatment than those who did not. Those who were homeless or who had an existing mental disorder were less likely to enter IOP treatment than those who were not homeless or did not have a mental disorder.

Those who were homeless did, however, have a significantly higher probability of entering LTR than those who were not. Those who drank alcohol at least 3 times in the month before admission, had a secondary substance of abuse, or had at least one prior treatment episode also had a greater probability of admission to LTR. However, those who drank at least once in the month before treatment or had become intoxicated with alcohol before age 15 had the same probability of entering LTR as those who did not, and those who had an existing mental disorder had a significantly lower probability of entering LTR than those who did not.

In contrast, clients with an existing mental disorder did have a 41.8 percent higher probability of admission to STR than those who did not have a mental disorder. Additionally, those who drank at least 3 times in the week prior to admission had a 77 percent higher probability of entering STR than those who did not drink in the month prior to admission, and those with secondary substance use had a 50.5 percent higher probability of doing so. However, those who drank at least once in the past month had a 71 percent lower probability of entering STR than those who did not, and neither intoxication prior to age 15 nor homelessness had any effect on the probability.

Regarding entry to IPH, almost all of the severity variables that were significant were positively associated with the probability of admission. Those who drank alcohol at least 3 times in the week prior to treatment were 382 percent more likely to enter IPH than were those who had not drunk during that time, while those who drank at least once in the month prior to admission had an almost 162 percent higher probability. Those with at least one prior treatment episode had a 59.2 percent higher probability of admission to IPH, those who were homeless a 99.8 percent higher probability, and those with an existing mental disorder almost a 530.7 percent higher probability. The exceptions were secondary drug use and intoxication before age 15, which did not significantly affect the probability of entry into the IPH setting.

Socioeconomic and Demographic Variables. As was the case in Iowa, employment at admission was significantly negatively associated with entry into all settings except SOP. Those who were employed had a 36.1 percent lower probability of entering IOP, a 197.7 percent lower probability of entering LTR, an 83.5 percent lower probability of entering STR, and a 102.8 percent lower probability of entering IPH than those who were not employed. In contrast, those who were employed at admission had a 22.5 percent higher probability of entering SOP treatment compared with those who were not employed.

The results for expected payer were again not straightforward, but again care should be used in their interpretation. In contrast to Iowa, self-payment was positively associated with entry to SOP and negatively associated with the presumably more expensive IOP, LTR, and STR than was the reference category, government funding or charity care. However, self-payment was not associated with entry into IPH. The associations revealed for the other payment sources, on the other hand, were more similar to those in New Jersey. Private insurance or other pay source (e.g., Medicaid, Medicare) was associated with a lower probability of entering SOP and higher probabilities of entering the presumably more expensive IOP and IPH than the reference category, but a lower probability of entering either type of residential care.

Race and ethnicity variables were significantly associated with entry into the various modes of treatment. Although Hispanics and those in the other race category had a higher probability of entering SOP and a lower probability of entering IOP than did non-Hispanic whites (the reference category), neither group was any more or less likely to enter LTR. Hispanics had a higher probability of entering STR, but a lower probability of entering IPH than did non-Hispanic whites. Non-Hispanic blacks were more likely to enter IOP, less likely to enter both STR and IPH, but equally likely as non-Hispanic whites to enter the other two treatment settings.

In contrast to Iowa, marital status did not strongly affect the probability of entry into most settings. The only significant effect of marital status was on the probability of entering LTR. Clients who were married had a 7.02 percent lower probability of entering LTR than those who were unmarried.

Referral Source and Season of Admission. Referral source was strongly associated with the probability of entering different treatment settings in New Jersey, although the pattern was somewhat different than it was in Iowa. In New Jersey, those referred by any of the other sources (e.g., self-referral, alcohol/drug treatment provider) had a significantly lower probability of entering SOP than did those who were referred by the CJS. Clients referred by these other sources also had significantly higher probabilities of entering both IOP and STR than did the CJS clients. In addition, those who self-referred had significantly higher probabilities of entering both LTR and IPH than those referred by the CJS, while those referred by an employer, school, or physician had significantly higher probabilities of being admitted to the STR and IPH settings.

As for season of admission, clients who were admitted to treatment in the fall and winter had a significantly higher probability of entering STR, and a lower probability of entering IPH, than did those who were admitted to treatment in the spring. Those admitted in the winter had a 4.9 percent lower probability of entering SOP treatment than those entering in the spring.

Discussion

As in our earlier work on choice of alcohol treatment setting, we found that, in general, those with more severe alcohol use disorders were admitted to treatment settings generally considered more intensive, but that socioeconomic and demographic variables also affected treatment-setting admission. However, when extending the analysis to a polychotomous choice among several different treatment settings, rather than a dichotomous choice between inpatient and outpatient, the major distinction appeared between SOP and all other treatment settings, including IOP. In both States, clients with more severe alcohol use disorders, as measured by such variables as frequency of use, intoxication prior to age 15, and having had a prior treatment episode, generally had a reduced probability of being admitted to SOP, but an increased probability of being admitted to any of the more intensive settings. Furthermore, the results do not suggest an inherent ordering among the more intensive settings. The proportionate effects of the severity variables were not always monotonically related to the continuum of intensity as it is often described, for example, by the ASAM criteria. These findings, along with the results of the chi-square test, suggest that treatment settings can be better studied as a polychotomous, rather than dichotomous, choice, and that a multinomial approach is more appropriate than an ordinal approach.

Our analysis revealed other key similarities between the Iowa and New Jersey client samples regarding treatment-setting choice. For example, in both States, clients who were employed were significantly more likely to enter SOP and less likely to enter any of the other treatment settings than were unemployed clients. Also similar in both States was the pattern associated with private health insurance coverage. Private health insurance was associated with a higher probability of entering IOP or IPH, but lower probability of entering SOP or either form of residential care than government assistance or charity care. Furthermore, clients referred by an alcohol or drug treatment provider in either State were more likely to enter STR and LTR and less likely to enter SOP treatment compared with clients referred through the CJS.

Our analysis also revealed some interesting differences between the two States. For example, in New Jersey clients who self-referred, were referred by an alcohol or drug treatment provider, or were referred by another source were more likely to enter IOP treatment compared with clients referred by the CJS. However, in Iowa we found the opposite to be true in that self-referred clients, clients referred by an alcohol or drug treatment provider, and clients referred by another source were less likely to enter IOP. Another interesting difference was the pattern revealed for self-payment. In Iowa, self-payment was positively associated with entering IOP and negatively associated with entering SOP compared with the reference cell. In New Jersey, we again found the opposite, with self-payment being positively associated with entering SOP and negatively associated with IOP compared with the reference cell. Further research using data that is sufficient to identify the possibly joint determination of payment source and treatment setting is needed to fully understand the effects of expected source of payment on the probability of entry into the various settings.

Although the results differed to some extent across the two States considered here, in both States, there is evidence that client placement is associated with the severity of the client's disorder, suggesting that clients are gaining admission to the facilities that may best meet their needs. Furthermore, use of IOP suggests that they are attempting to use their limited resources in the most cost-effective way. This finding suggests that States have been able to adapt to changes in the treatment practices in ways that both improve services and contain costs.

References

American Society of Addiction Medicine. (1996). Patient placement criteria for the treatment of substance-related disorders (2^nd ed.). Chevy Chase, MD: Author.

Annis, H. M. (1985–1986). Is inpatient rehabilitation of the alcoholic cost effective? Con position. Advances in Alcohol & Substance Abuse, 5(1–2), 175–190.

Arroyo, J. A., Westerberg, V. S., & Tonigan, J. S. (1998). Comparison of treatment utilization and outcome for Hispanics and non-Hispanic whites. Journal of Studies on Alcohol, 59, 286–291.

Becker, G. S., & Murphy, K. M. (1988). A theory of rational addiction. Journal of Political Economy, 96, 675–700.

Ben-Akiva, M., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand. Cambridge, MA: MIT Press.

Brito, D. L., & Strain, C. K. (1996). A model of the consumption of alcohol. In L. Green & J. H. Kagel (Eds.), Advances in behavioral economics: Volume 3. Substance use and abuse (pp. 163–182). Norwood, NJ: Ablex Publishing Corporation.

Etheridge, R. M., Craddock, S. G., Hubbard, R. L., & Rounds-Bryant, J. L. (1999). The relationship of counseling and self-help participation to patient outcomes in DATOS. Drug and Alcohol Dependence, 57, 99–112.

French, M. T., & Zarkin, G. A. (1992). The effects of drug abuse treatment on legal and illegal earnings. Contemporary Policy Issues, 10, 98–110.

Goodman, A. C., Holder, H. D., Nishiura, E., & Hankin, J. R. (1992). A discrete choice model of alcoholism treatment location. Medical Care, 30, 1097–1110.

Goodman, A. C., Nishiura, E., & Hankin, J. R. (1998). A discrete choice model of drug abuse treatment location. Health Services Research, 33(1), 125–145.

Gottheil, E. (1997). Intensive outpatient treatment: Methods and outcomes. Journal of Addictive Diseases, 16(2), xv-xix.

Grant, B. F., & Dawson, D. A. (1997). Age at onset of alcohol use and its association with DSM-IV alcohol abuse and dependence: Results from the National Longitudinal Alcohol Epidemiologic Survey. Journal of Substance Abuse, 9, 103–110.

Green, L., & Kagel, J. H. (Eds.). (1996). Advances in behavioral economics: Volume 3. Substance use and abuse. Norwood, NJ: Ablex Publishing Corporation.

Greene, W. H. (2000). Econometric analysis (4^th ed.). Upper Saddle River, NJ: Prentice Hall.

Gregoire, T. K. (2000). Factors associated with level of care assignment in substance abuse treatment. Journal of Substance Abuse Treatment, 18, 241–248.

Hubbard, R. L., Marsden, M. E., Rachal, J. V., Harwood, H. J., Cavanaugh, E. R., & Ginzburg, H. M. (1989). Drug abuse treatment: A national study of effectiveness. Chapel Hill, NC: University of North Carolina Press.

Lundgren, L. M., Amodeo, M., Ferguson, F., & Davis, K. (2001). Racial and ethnic differences in drug treatment entry of injection drug users in Massachusetts. Journal of Substance Abuse Treatment, 21, 145–153.

Majella Jordan, C., & Oei, T. P. (1989). Help-seeking behaviour in problem drinkers: A review. British Journal of Addiction, 84, 979–988.

Marques, P. R., Zaloshjna, E., Miller, T. R., Waehrer, G. M., Danseco, E. R., Levy, D. T., Romano, E. O., & Teti, L. O. (2001). Financial risk, plan types, and authorizations for managed alcohol treatment services. American Journal of Managed Care, 7, 1151–1161.

McLellan, A. T., Hagan, T. A., Levine, M., Meyers, K., Gould, F., Bencivengo, M., Durell, J., & Jaffe, J. (1999). Does clinical case management improve outpatient addiction treatment. Drug and Alcohol Dependence, 55(1–2), 91–103.

Miller, W. R., & Hester, R. K. (1986). Inpatient alcoholism treatment: Who benefits? American Psychologist, 41, 794–805.

Montoya, I. D., Atkinson, J. S., & Trevino, R. A. (2000). Economics as a factor in models of behavioral motivation and change. Substance Use & Misuse, 35, 329–346.

Muurinen, J. M. (1982). Demand for health: A generalised Grossman model. Journal of Health Economics, 1(1), 5–28.

Office of Applied Studies. (1999). Treatment Episode Data Set (TEDS): 1992–1997: National admissions to substance abuse treatment services (DHHS Publication No. SMA 99–3324, Drug and Alcohol Services Information System Series S-7). Rockville, MD: Substance Abuse and Mental Health Services Administration. [Available at http://www.oas.samhsa.gov/dasis.htm#teds2]

Office of Applied Studies. (2000). Uniform Facility Data Set (UFDS): 1998 data on substance abuse treatment facilities (DHHS Publication No. SMA 00–3463, Drug and Alcohol Services Information System Series S-10). Rockville, MD: Substance Abuse and Mental Health Services Administration. [Available at http://www.oas.samhsa.gov/dasis.htm#nssats2]

Samarasinghe, D. (1996). The meaning of treatment and setting. Addiction, 91, 1810–1811.

Sheikh, S., & Furnham, A. (2000). A cross-cultural study of mental health beliefs and attitudes towards seeking professional help. Social Psychiatry and Psychiatric Epidemiology, 35, 326–334.

Simpson, D. D., Joe, G. W., Rowan-Szal, G., & Greener, J. (1995). Client engagement and change during drug abuse treatment. Journal of Substance Abuse, 7, 117–134.

Suranovic, S. M., Goldfarb, R. S., & Leonard, T. C. (1999). An economic theory of cigarette addiction. Journal of Health Economics, 18(1), 1–29.

Washton, A. M. (1997). Evolution of intensive outpatient treatment (IOP) as a "legitimate" treatment modality. Journal of Addictive Diseases, 16(2), xxi-xxvii.

---

End Notes

¹ In the TEDS data, States are to identify as IOP treatment the client receives that lasts 2 or more hours per day on 3 or more days per week (OAS, 1999).

² Specifically, small cell sizes led to unreasonably large point estimates and standard errors in the MNL model. To increase cell sizes, we combined categories for a number of independent variables.

³ The percentage change in the probability is equal to the marginal effect divided by the predicted probability. In this case, that is -0.343 / 0.772 = -0.444.

SAMHSA, an agency in the Department of Health and Human Services, is the Federal Government's lead agency for improving the quality and availability of substance abuse prevention, addiction treatment, and mental health services in the United States.

Site Map | Contact Us | Accessibility Privacy Policy |  Freedom of Information Act |  Disclaimer  |  Department of Health and Human Services |  SAMHSA |  White House |  USA.gov

* Adobe™ PDF and MS Office™ formatted files require software viewer programs to properly read them. Click here to download these FREE programs now