Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2007 Dec 28.
Published in final edited form as: J Drug Issues. 2007;37(2):321–340. doi: 10.1177/002204260703700205

The Multidimensional Structure of Internal Barriers to Substance Abuse Treatment and Its Invariance Across Gender, Ethnicity, and Age

Jiangmin Xu 1, Jichuan Wang 2, Richard C Rapp 3, Robert G Carlson 4
PMCID: PMC2168036  NIHMSID: NIHMS29635  PMID: 18167519

Abstract

The goal of the present study was to identify the dimensions present in items representing internal barriers to substance abuse treatment and to test their invariance across gender, ethnic, and age groups. Twenty items from the Barriers to Treatment Inventory (BTI) were used to assess the structure and nature of the internal treatment barriers of 518 clients presenting to a central intake unit for a substance abuse assessment. Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) revealed that a five factor model provided the best fit to the data. Internal treatment barriers were best described by five dimensions: absence of problem, negative social support, fear of treatment, privacy concerns, and committed lifestyle. Extending the confirmatory factor analysis to test multi-group invariance, there were some differences in measurement and structural relations among the internal barrier dimensions across gender, ethnic, and age groups. However, the lack of invariance was small and practically insubstantial. The findings led to the conclusion that the theoretical constructs measured by the five internal barrier dimensions are equivalent across important characteristics in this population.

Introduction

Although drug abuse treatment is widely available in the United States, there is still a low utilization rate for users, and most substance users who need these services have never been in treatment (Brown & Needle, 1994; Sobell, Sobell, & Toneatto, 1992). A substantial body of literature has found numerous barriers associated with poor rates of treatment entry (Beckman & Amaro, 1986; Hser, Maglione, Polinsky, & Anglin, 1998; Siegal, Falck, Wang, & Carlson, 2002) and retention (Joe, Simpson, & Broome, 1998; Vaughn, Sarrazin, Saleh, Huber, & Hall, 2002).

Barriers to treatment are events or characteristics of the individual or system that restrain or serve as obstacles to the person receiving health care or drug treatment (Cunningham, Sobell, Sobell, Agrawal & Toneatto, 1993; Janz & Becker, 1984; Tsogia, Copello, & Orford, 2001). Previous barrier studies have tried to identify the constructs of barriers to treatment. Melnyk (1990) operationalized the treatment barrier variables and identified five categories of the barriers construct: relationship, site-related aspects, cost, fear, and inconvenience. Allen and Dixon (1994) identified four factors with factor loadings of 0.40 or more, including interaction with others, aspects of treatment programs, social support, and financial concerns. Asking participants to rate barriers to, or reasons for, seeking treatment, Tucker, Vuchinich, and Rippens (2004) produced a three-factor solution that reflected (a) privacy concerns, (b) participant beliefs that treatment was unnecessary or not beneficial, and (c) practical and economic obstacles to participation. However, these studies did not clearly and systematically classify treatment barriers as internal or external ones. Therefore, there is a need for greater conceptual clarity in the structure of barriers to treatment at both the theoretical and the empirical levels.

Barriers to treatment can be conceptualized along internal and external dimensions (Allen, 1994; Allen, 1995; Hser et al., 1998; Melnyk, 1990). Allen (1994) defined internal treatment barriers as “subjective phenomena—beliefs or perceptions arising from within the person” and external barriers as “health care system, structural characteristics of a program, and socio-cultural–environmental factors.” Subjective (internal) sources of barriers included failure to recognize having a problem, fear of others' reactions, fear of stigma, and fear of the unknown related to treatment (Allen, 1995). Developing the Barriers to Treatment Inventory (BTI) to systematically assess both internal and external barriers among a pretreatment sample of substance abusers, Rapp, et al. (2006) demonstrated the presence of four internal barriers—absence of problem, negative social support, fear of treatment, and privacy concerns—as well as three external barriers—time conflict, poor treatment availability, and admission difficulty.

Internal barriers have been identified as particularly important in interfering with treatment linkage (Allen, 1995). Not recognizing a problem and having little motivation to enter treatment are two of the most frequently cited reasons for not entering treatment (Beckman & Amaro, 1986; Cunningham et al., 1993;Hingson, Mangione, Meyers, & Scotch, 1982;Tucker, 1995). McCoy, Metsch, Chitwood, and Miles (2001) reported that “not wanting health treatment” and “deciding to treat oneself” acted as two of five most frequently reported barriers for not seeking treatment. Rounsaville and Kelber (1985) reported that 18% of opiate addicts considered treatment, but did not need it badly enough at that point to seek help. Carroll and Rounsaville (1991) found that 50% of cocaine abusers who failed to seek drug treatment felt no need for treatment, and 56% believed their cocaine use was under control.

Internal barriers such as attitude regarding treatment have been found to be more important than external ones such as lack of financial resources or facilities for childcare (Grant, 1997). Fear of treatment and privacy concerns appear as factors that inhibit treatment entry (Jessup, Humphreys, Brindis, & Lee, 2003; Melnyk, 1990; Sheehan, Oppenheimer, & Taylor, 1986; Tucker et al., 2004). Injection drug users identified “fear of treatment, bad previous treatment experiences, or aversion to a specific type, most methadone maintenance” as their main barriers to treatment (Appel, Ellison, Jansky, & Oldak, 2004). The exact reasons for fear of treatment and privacy concerns are not always specified, although the inability to share problems with others and stigma may play a role (Cunningham et al., 1993; Grant, 1997).

Substance abusers may receive social pressure and influence from significant others (e.g. partners, friends, families, and natural social networks) (Marlowe et al., 1996; Wilsnack, 1991). This social pressure and influence may either promote or inhibit treatment linkage. In a sample of alcoholic women, almost a quarter (23% of alcoholic women but only 2% of alcoholic men) reported that their friends and family members opposed their seeking treatment during the month prior to entry (Beckman & Amaro, 1986). Studying fears about treatment in a sample of 50 clients of a London drug dependency clinic, Sheehan et al. (1986) reported that the fears of disappointing those trying to help and being unable to keep away from drug-using friends were considered to be important barriers to seeking treatment.

In addition to examining the structure of internal barriers, it is important to determine whether there is reasonable support for the invariance of the internal barrier structure across different groups. Demographic variables are often used to predict treatment entry; however, the research findings and conclusions have been inconsistent. For example, some approaches have found that gender (Kaskutas, Weisner, & Caetano, 1997; Weisner, 1993), race/ethnicity (Farabee, Leukefeld, & Hays, 1998; Lundgren, Amodeo, Ferguson, & Davis, 2001), and age (Hajema, Knobbed, & Drop, 1999; Kaskutas et al., 1997; Pfeiffer, Feuerlein, & Brenk-Schulte, 1991) are associated with treatment entry, whereas others have not found them to be important (Hasin, 1994; Hser et al., 1998; Tucker, 1995). McCoy et al. (2001) reported that men were 1.3 times more likely not to want treatment and less likely to self-treat compared to women. However, Grant (1997) reported that there were few differences between gender groups in perceived barriers to treatment. He also found that non-African Americans were more likely than African Americans to enter treatment programs because they were afraid of what their boss, friends, family, or others would think (8.4% vs. 0.3%).

The consequences of using nonequivalent measures are potentially problematic. Statistically, if a measure is valid and accurate for one group but less for another group, it may result in misclassification in the second group, distort prevalence estimations, and yield inaccurate information. Clinically, if an instrument is not consistent for different groups (e.g., men and women), then clinicians who used the instrument may make incorrect assumptions about the perceptions that different groups have of different internal barriers to treatment. Therefore, a better understanding of potential differences in how barriers are perceived will help focus efforts to intervene with substance abusers seeking treatment.

The primary objective of this research was to determine whether different internal barriers reflect a single underlying factor structure, or are they better viewed as a multidimensional factor structure. The second objective was to examine whether the internal factor structure of the Barriers to Treatment Inventory (BTI) is similar across independent samples of the same population of pretreatment substance abusers. A confirmatory factor analysis model was used to reassess the structure of internal barriers found during original development of the BTI. Specifically, factorial invariance of internal barriers was examined across gender, ethnic, and age groups.

Methods

Sampling

Data were collected as part of the ongoing longitudinal study, Reducing Barriers to Drug Abuse Treatment Services project (RBP), funded by the National Institute on Drug Abuse (NIDA). All participants were referred for substance abuse assessment by a centralized intake unit (CIU), the county's crisis, assessment, and referral service for individuals with mental health and substance abuse problems. A total of 523 participants were recruited between April 2004 and September 2005 following assessment and referral in order to assess their views of barriers to treatment. However, five participants, including two American Indian, one Asian, and two Hispanic participants, were excluded from this sample because they were insufficiently represented for statistical comparison with Whites and African Americans.

Substance abusers who meet the following criteria were referred to the Reducing Barriers Project: (a) over 18 years of age, (b) diagnosed as having a substance abuse and/or dependence disorder using criteria from the Diagnostic and Statistical Manual IV-R (American Psychiatric Association, 2000) (participants who were diagnosed only with alcohol abuse or dependence were not eligible), (c) not diagnosed with schizophrenia or any other psychotic disorder, and (d) referred to either residential or outpatient substance abuse services. Targeted sampling was used to match the CIU population and Reducing Barriers Project sample on characteristics of race, gender, age, and court referral status.

Eligible participants were referred to RBP research staff by CIU assessment therapists. RBP research assistants provided a summary of the project, and if an individual was interested, an informed consent approved by a university's institutional review board was read to them. The confidential nature of the study was stressed as was the fact that refusal to participate did not affect CIU services for which an individual was otherwise eligible. Individuals who wished to enter the study then participate in a baseline interview lasting about 1 ½ hours. Most interviews took place immediately following a clinical assessment, although some potential participants were scheduled to return at a later time. Follow-up interviews were conducted at three and six months following baseline. Participants were paid a $30 stipend for their time spent answering questions on each interview.

Measures

Only items from the Barriers to Treatment Inventory (BTI) questionnaire that represented internal barriers to treatment were examined in this study. The BTI was developed to assess the barriers to treatment among pretreatment substance abusers (Rapp et al., 2006). The participants were asked to indicate on a five-point scale how much the barriers influenced their ability to access treatment services. The scales ranged from (1) strongly disagree, (2) disagree, (3) neutral, (4) agree, to (5) strongly agree.

Items from the original four internal barrier factors were selected for inclusion in this analysis. They included (a) absence of problem, consisting of six items (e.g., “I don't think treatment will make my life better”); (b) negative social support, consisting of five items gauging the influences of others (e.g., “people will think badly of me if I go to treatment”); (c) fear of treatment, consisting of four items (e.g., “I'm afraid of what might happen in treatment”); and (d) privacy concerns, consisting of three items (e.g., “I hate being asked personal questions”).

Statistical Analyses

The factorial structure of the internal treatment barriers were verified and evaluated by exploratory and confirmatory factor analyses using SPSS 14.0 (SPSS Inc., 2005) and the structural equation-modeling software AMOS 5.0 (Analysis of Moment Structures) (Arbuckle, 2003). Extraction of factors was based on the minimum Eigenvalue and the amount of variance that was explained. Internal consistency of the items for each subscale of barriers was assessed by Cronbach's Alpha, which is also a measure of reliability of each construct.

The structure invariance of internal barriers was tested by a series of confirmatory factor analyses (CFAs). Prior to extending CFA to test multi-group invariance across gender, ethnic, and age groups, the baseline model was tested separately for each group with no invariance constraints and then assessed to see if the model fit the data well. If models fit the data well, we conducted subsequent tests for multi-group invariance. Otherwise some further specifications would be needed.

As suggested by Byrne (1998), the multi-group invariance was then evaluated by a series of logical order confirmatory factor analyses (CFAs) to determine if and how the factor structure underlying internal barriers, and the subscales used to measure these constructs, were equivalent across groups for this population. Testing of factorial invariance allowed us to determine whether or not there was equivalent operation between the barrier items comprising a measuring instrument and the factorial structure relations among the facets of treatment barriers. Although the pattern of factor loadings and structural relations among factors were both of interest, the equivalence of the measurement models were always tested first and then followed by the equivalence of the structural models. In this study, sets of parameters were tested in a logical order: (a) factor loading, (b) factor variance, and (c) factor covariance.

In testing for factorial invariance, a series of factor models was examined by a series of multiple-group analyses. A baseline Chi-square value was derived by computing model fit for the pooled sample of all groups. Then a series of increasingly restrictive hypotheses for sets of parameters were tested and equality constraints were made to test the invariance of parameters across groups. In other words, various model parameters were constrained to be equal across groups, yielding a Chi-square for the constrained model. A Chi-square difference test was then applied to see if the difference was significant. If a new set of parameters was found to be noninvariant across groups, we no longer constrained them to be equal and the next step was to locate the nonequivalent parameters in the model. If a new set of parameters was found to be group invariant, equality constraints related to the parameters were cumulatively held in place and we continued to test equality across groups.

Several goodness-of-fit indices were examined in evaluating CFAs. The primary criterion for evaluating the fit of each model was the comparative fit index (CFI) (Bentler, 1990). This index is determined by comparing the fit of the model and the fit of the independent model. The measure should be 0 to 1, and values above 0.90 represent a good fit. The Tucker-Lewis coefficient (TLI) (Bentler & Bonett, 1980), root mean squared error of approximation (RMSEA) (Steiger, 1980), and probability of close fit (PCLOSE) are three other indicators that measure the fit between the model and the observed data. RMSEA is one of the most sensitive indexes for models with misspecified factor loadings (Hu & Bentler, 1999). RMSEA values of less than 0.05 indicate a close fit and less than 0.08 a reasonable fit (Browne & Cudeck, 1993; Marsh, Balla, & Hau, 1996). Employing this definition, PCLOSE gives a test of close fit. PCLOSE is a p value for testing the null hypothesis that the population RMSEA is not greater than 0.05 (Arbuckle, 1997, pp.403).

Results

Client Characteristics and Descriptive Statistics

Among the 518 participants, 62.4% were male, and 47.3% were African American. The mean age of participants was 33.61 years (S.D=10.64, range: 18-64). Twenty-four percent (24.3%) of the sample were between ages 18 and 24; 34.7% were between ages 25 and 34; and 40.9% were aged 35 or older. The participants had completed a mean of 11.19 years of education (S.D. 1.91, range 6-16); 68.9% had graduated from high school or completed GED; and 31.5 % completed a vocational or technical training program. About 22.4% of clients were employed either full or part time. About 70% of the sample (males 40.1% and females 59.9%) had been previously treated for alcohol or drug abuse. The drug problems identified by clients were most often cocaine (38.4%), followed by heroin (25.1%), marijuana (14.9%), and alcohol (11.2%).

Factor Loadings, Inter-Item Reliability, and Intercorrelations

The goal of the confirmatory factor analysis was to determine the multidimensional factor structure of the internal barriers presented in the barriers to treatment inventory. Four of the original factors were again present: absence of problem, negative social support, fear of treatment, and privacy concerns. An additional two-item factor was also present: committed lifestyle, consisting of two items measuring whether the substance abusers wanted to change their life or not (e.g., “Using drugs is a way of life for me”).

Table 1 provides factor loadings and Cronbach's alpha for each of the internal barrier constructs. Within this CFA model, factor loadings were significant with standardized loadings ranging from 0.48 to 0.90. Reliability analyses of factors produced a standardized alpha for each factor as follows: absence of problem, 0.85; negative social support, 0.75; fear about treatment, 0.72; privacy concerns, 0.79, and committed lifestyle, 0.56.

Table 1.

Factor Loadings for CFA Model (N=518)

Internal Treatment Barrier items F1 (AP) F2 (NSS) F3 (FT) F4 (PC) F5 (CL) Cronbach's Alpha
Factor Loadings
Absence of Problem (AP) .85
AP1. I do not think I have a problem with drugs .71
AP2. No one has told me I have a problem with drugs .59
AP3. My drug use is not causing any problems .75
AP4. I do not think treatment will make my life better .67
AP5. I can handle my drug use on my own .71
AP6. I do not think I need treatment .83
Negative Social Support (NSS) .75
NSS1. I will lose my friends if I go to treatment .48
NSS2. Friends tell me not to go to treatment .61
NSS3. People will think badly of me if I go to treatment .73
NSS4. Someone in family doesn't want me to go to treatment .65
NSS5. My family will be embarrassed or ashamed if I go to treatment .63
Fear of Treatment (FT) .72
FT1. Treatment will add another stress to my life .63
FT2. I am afraid of what might happen in treatment .67
FT3. I am afraid of the people I might see in treatment .62
FT4. I am too embarrassed or ashamed to go to treatment .57
Privacy Concerns (PC) .79
PC1. I don't like to talk in groups .61
PC2. I hate being asked personal questions .78
PC3. I don't like to talk about my personal life with other people .90
Committed Lifestyle (CL) .56
CL1. I cannot live without drugs .79
CL2. Using drugs is a way of life for me .50
Model fit: χ2=300.00, df=160, p<.001, CFI=.96, TLI=.95, RMSEA=.04, and PCLOSE=.98.
Open in a new tab

Note: χ2=Chi-square; df=degree of freedom; CFI = Comparative Fit Index; TLI=Tucker-Lewis index; RMSEA = Root Mean Square Error of Approximation; PCLOSE=Probability of Close Fit (Pr[RMSEA<.05]). Significance tests based on robust maximum likelihood estimates. All Coefficients were standardized coefficients. All factor loadings were significant at p<.001.

Table 2 shows the intercorrelations among the five internal barrier factors. All intercorrelations were statistically significant, ranging from −0.12 to 0.66. Relatively lower correlations were found between absence of problem and other factors (rs ranging from −0.12 to 0.33).

Table 2.

Standardized Correlation Coefficients Among Internal Barrier Factors

Internal Barriers 1 2 3 4 5
F1. Absence of Problem (AP) 1.00 .24*** .33*** .16** −.12*
F2. Negative Social Support (NSS) 1.00 .66*** .29** .35***
F3. Fear of Treatment (FT) 1.00 .45*** .38***
F4. Privacy Concerns (PC) 1.00 .20***
F5. Committed Lifestyle (CL) 1.00
Open in a new tab
*

p<.05

**

p<.01

***

p<.001

Baseline Model Fits for Each Gender, Ethnic, and Age Group

The CFAs were conducted in which parameters were estimated separately for each gender, ethnic, and age group. The model fit of the baseline models was considered adequate with CFIs from 0.90 to 0.95, TLIs from 0.88 to 0.94, and RMSEAs from 0.04 to 0.07. All values of PCLOSE were statistically nonsignificant except for the 18 to 24 age group. These results provided strong support for the internal barrier structure and indicated that the models with five correlated factors fit the data significantly for these groups.

Factorial Invariance Tests Across Gender, Ethnic, and Age Groups

The multigroup invariance tests were then conducted to compare parameters across groups. Tables 4 through 6 provide a summary of Chi-square values and difference values related to the application of invariance testing procedures for gender, ethnic, and age groups. Model 1 reflects the extent to which the underlying structure fits the data across groups when no cross-group equality constraints were imposed. Model 1, as a hypothesized and baseline model, provides the basis for determining which model is equivalent across groups. The key statistic of interest are the Chi-square value of 540.98 (320 df) for gender groups, 499.91 (320 df) for ethnic groups, and 722.86 (480 df) for age groups.

Table 4.

Goodness-of-Fit Statistics for Tests of Invariance Across Gender Groups

Model Description χ2 df Δχ2 Δdf p value
Model 1: Hypothesized model
(Baseline model with no invariance constraints)		 540.98 320
Model 2: Factor loadings, variance and
Covariance constrained equal		 587.66 350 46.68 30 p<.05
Model 3: All Factor loadings constrained equal 567.06 335 26.08 15 p<.05
Model 4: Model 1 with Factor Loadings
on AP constrained		 549.04 325 8.06 5 NS
Model 5: Model 4 with Factor Loadings
on NSS constrained equal		 562.73 329 21.75 9 p<.01
Model 6: Model 4 with factor loading of
item NSS2 on NSS constrained		 555.21 326 14.23 6 p<.05
Model 7: Model 4 with factor loading of
NSS3 on NSS constrained		 549.78 326 8.8 6 NS
Model 8: Model 4 with factor loadings of
item NSS3 and NSS4 on NSS constrained		 552.72 327 11.74 7 NS
Model 9: Model 4 with factor loadings of
item NSS3 and NSS4 and NSS5
on NSS constrained		 558.06 328 17.08 8 p<.05
Model 10: Model 8 with factor loadings on
FT constrained equal		 553.29 330 12.31 10 NS
Model 11: Model 8 with factor loading on
FT and PC constrained equal		 556.74 332 15.76 12 NS
Model 12: Model 8 with factor loading on
FT, PC, and CL constrained equal		 557.11 333 16.13 13 NS
Model 13: Model 12 with all variances
constrained equal		 559.63 338 18.65 18 NS
Model 14: Model 13 with all covariance
constrained equal		 575.79 348 34.81 28 NS
Open in a new tab

Note: χ2=chi-square; df=degree of freedom; Comparative model is baseline model (model 1); Δχ2, difference in chi-square values between models; Δdf, difference in degrees of freedom between models

*

p<.05

**

p<.01

***

p<.001

Table 6.

Goodness-of-Fit Statistics for Tests of Invariance Across Age Groups

Model Description χ2 df Δχ2 Δdf p value
Model 1: Hypothesized model
(Baseline model with no invariance constraints)		 722.86 480
Model 2: Factor loadings, variance
and covariance constrained equal		 806.79 540 83.93 60 p<.01
Model 3: Factor loadings constrained equal 746.76 510 23.90 30 NS
Model 4: Model 3 with all variance
constrained equal		 780.89 520 58.03 40 p<.05
Model 5: Model 3 with variance of
AP constrained equal		 770.26 512 47.40 32 p<.05
Model 6: Model 3 with variance of
NSS constrained equal		 749.55 512 26.29 32 NS
Model 7: Model 3 with variances of
NSS and FT constrained equal		 753.06 514 30.20 34 NS
Model 8: Model 3 with variance of
NSS, FT, and PC constrained equal		 755.87 516 33.01 36 NS
Model 9: Model 3 with variance of
NSS, FT, PC, and CL constrained equal		 757.94 518 35.08 38 NS
Model 10: Model 9 with all covariance
constrained equal		 780.81 538 57.95 58 NS
Open in a new tab

Note: χ2=chi-square; df=degree of freedom; Comparative model is baseline model (model1); Δχ2, difference in chi-square values between models; Δdf, difference in degrees of freedom between models

*

p<.05

**

p<.01

***

p<.001

Model 2 is the model in which all equality constraints were specified. Comparing the constrained models (Model 2) with the original unconstrained models (Model 1) yields Chi-square difference values that suggest whether or not the model is equivalent across groups or not. In this study, Chi-square difference values of 46.68 with 30 degrees of freedom for gender groups, 75.08 with 30 degrees of freedom for ethnic groups, and 83.93 with 60 degrees of freedom for age groups, were statistically significant, which suggests that the model was not equivalent across groups.

Three steps were taken to determine which parameters in the model were group noninvariant. First, invariance related to the measurement model was tested. In other words, the pattern of factor loadings for each observed measure was tested for its equivalence across the groups. Because the factor loading reflects the relation between a specific indicator and the underlying latent construct, it represents the validity of the indicator. For example, as shown in Table 4, the Chi-square difference between Model 1 (Baseline Model) and Model 3 (all factor loadings constrained equally) is statistically significant. This indicates that it is likely each factor loading for each measured variable is identical in gender groups.

To locate the nonequivalent factor loadings, we tested for invariance relative to each factor separately. If the equivalences of factor loadings related to a factor were not statistically significant, those constraints would be held in place while we tested for the invariance of the factor loadings for next factor; if the equivalences of factor loadings related to a factor were statistically significant, they indicated some discrepancies in the measurement of the factor between groups and those constraints would be released while we tested for the invariance of the factor loadings for the next factor. A series of tests would pinpoint these nonequivalent factor loadings, and this process was continued until all parameters of interest were tested.

For gender groups, the test results for invariance related to absence of problem (AP), fear about treatment (FT), privacy concerns (PC), and committed lifestyle (CL) were not statistically significant, thereby indicating its equality across gender. However, the test for invariance related to negative social support (NSS) was statistically significant. Therefore, a series of tests revealed the factor loadings associated with item NSS2 and item NSS5 to be groups invariant. As shown in Table 5 and Table 6, the differences in Chi squares between Model 1 (baseline) and Model 3 with all factor loadings constrained equally are not statistically significant, indicating that all factor loadings of internal treatment barriers are identical for the ethnic and age groups.

Table 5.

Goodness-of-Fit Statistics for Tests of Invariance Across Ethnic Groups

Model Description χ2 df Δχ2 Δdf p
Model 1: Hypothesized model
(Baseline model with no invariance constraints)		 499.91 320
Model 2: Factor loadings, variance
and covariance constrained equal		 574.99 350 75.08 30 p<.001
Model 3: Factor loadings constrained equal 515.13 335 15.22 15 NS
Model 4: Model 3 with all variance
constrained equal		 544.66 340 44.75 20 p<.01
Model 5: Model 3 with variance of
AP constrained equal		 536.85 336 36.94 16 p<.01
Model 6: Model 3 with variance of
NSS constrained equal		 517.44 336 17.53 16 NS
Model 7: Model 3 with variances of
NSS and FT constrained equal		 523.66 337 23.75 17 NS
Model 8: Model 3 with variance of
NSS, FT, and PC constrained equal		 523.75 338 23.84 18 NS
Model 9: Model 3 with variance of
NSS, FT, PC, and CL constrained equal		 524.63 339 24.72 19 NS
Model 10: Model 9 with all covariance
constrained equal		 545.73 349 45.82 29 NS
Open in a new tab

Note: χ2=chi-square; df=degree of freedom; Comparative model is baseline model (model1); Δχ2, difference in chi-square values between models; Δdf, difference in degrees of freedom between models

*

p<.05

**

p<.01

***

p<.001

The next step was to test invariance of variances. Increasingly restrictive models were tested while continuing to hold all constrained parameters found to be cumulatively invariant across groups. As shown in Table 4, the difference in Chi-square value between Model 1 and Model 13 with all variance constrained equal was not statistically significant (Δχ2(18))=18.65. This indicated that variances of all five factors were invariant across gender groups. As noted in Tables 5 and 6, because the differences in Chi-square values between Model 1 and Model 4 with all variance constrained equal were statistically significant (Δχ2(20))=44.75 and (Δχ2(40))=58.03, the hypotheses of invariant factor variance were rejected. Therefore, tests would determine which variances were contributing to this inequality across ethnic and age groups. Turning to Tables 5 and 6, significant ethnic and age differences were found with respect to the variances of absence of problem (AP) (Model 5).

The final step was to test equality of factor covariances related to the structure relations among the five factors of internal treatment barriers. Based on the procedures of keeping equality constraints for those invariant items, covariances were tested for all groups. The results showed that covariance was invariant across gender, ethnic, and age groups.

Discussion

The results of this study indicate that different internal barriers are better viewed as a multidimensional factor structure and that the internal factor structure of the barriers to treatment inventory is similar across independent samples of the same population of pretreatment substance abusers.

The most consistent finding for this sample is the significant contribution of the multidimensional factor structure. The CFA approach reveals that the factor structure of internal barriers is most accurately viewed as a five-factor model, comprising absence of problem, negative social support, fear of treatment, privacy concerns, and committed lifestyle. The good fit of baseline model test for each gender, ethnic, and age group clearly demonstrates the multidimensionality of internal barrier structure. These findings are very similar to those of our earlier analyses (Rapp et al., 2006), differing only by the addition of committed lifestyle. Committed lifestyle is significantly correlated with the other four barrier factors, all in the expected direction. The new factor suggests that total immersion in getting and using substances is itself a barrier, perhaps independent of other barriers. Prior literature has suggested that the most frequently cited self-reported reason for seeking drug abuse treatment was that substance abusers were tired of the drug-using lifestyle (79%) (Hser et al., 1998). For most substance abusers, changing drug using lifestyle was the most difficult aspect because drug use had become ingrained in their daily social life (Parker, Bakx, & Newcombe, 1988; Power, Jones, Kearns, Ward, & Perera, 1995).

Second, our findings indicate that the factor structure of internal barrier measurement is reasonably invariant across different groups and that the scales to assess barriers to treatment are the same for various groups. The five-factor internal barrier model using multi-group CFAs provides not only a good fit to the data, but also strict information on the equality of estimates across groups described by gender, ethnicity, and age. There are some minor differences in the measurement and structural relations among the internal barrier facets. For example, some factor loadings (NSS2 and NSS5) associated with negative social support are group variant (see Table 4), indicating that certain items in this scale are not invariant across gender groups. We also found ethnic and age noninvariance in absence of problem variances. However, the lack of invariance in this sample is small and practically insubstantial, which indicates that the multidimensional structure of internal barriers does apply across gender, ethnic, and age groups.

It is important to emphasize group differences in the factor structure of internal barriers to treatment. When factor loadings differ substantially across groups, the meanings of the latent constructs will differ substantially, even though the overall factor is retained. In other words, if underlying meanings of measurement or structural components of a model differ depending on gender, ethnicity, or age, then interpretational confounding could occur and assessing internal treatment barriers would be difficult. However, group differences in this sample are small enough for us to justify the conclusion that the items comprising the internal barriers of the BTI and the factorial structure relations of the measuring instrument among the facets of internal barriers operate equivalently across independent samples of the same population and that the theoretical construct measured by multiple items is also equivalent across gender, ethnic, and age groups in this population.

Because this study has some limitations, the results must be interpreted cautiously. One such limitation is substance abusers' self-reports of their treatment barriers. Fortunately, reviews of the literature in this area have suggested that self-report data from such samples tend to be reasonably reliable (Adair, Craddock, Miller, & Turner, 1995; Needle et al., 1995; Siegal et al., 2002). In addition, identifying barriers to treatment is somewhat less sensitive than reporting information about drug use or criminal activity.

Furthermore, although the BTI has strong theoretical and empirical applications as a measure of treatment barriers, it will benefit from further refinement. The important directions for future research are to consider new items that might be used to replace items we identified as problematic. Additional items relating to immersion in the drug using lifestyle might provide conceptual clarity to that factor.

A further limitation concerns the use of the CIU-based sample. This sample includes only the clients who present to the CIU for a substance abuse assessment and does not include those with their own private insurance and those who are not actively seeking help. As such, it is likely to result in underestimates of the prevalence of treatment barriers among a more general population of drug abusers.

The present study achieves a conceptual advance in our understanding of internal barriers to treatment. This study makes a significant contribution to the literature examining the factorial structure of treatment barriers by providing support for a well-defined five factor solution for internal barriers and by showing that the items comprising a measuring instrument and the factorial structure relations underlying internal barriers are equivalent across independent samples of the same population. These results provide strong support for the multidimensionality of internal treatment barriers and the invariant factor structure across different gender, ethnicity, and age groups. A clearer understanding of internal barriers to treatment will provide researchers and clinicians with a greater understanding of the relationship among treatment barriers, treatment entry, and treatment retention.

Table 3.

Fit Indices of Baseline Models for Each Gender and Ethnic and Age Group

Model χ2 df CFI TLI RMSEA PCLOSE
Male 293.95*** 160 .94 .92 .05 .46
Female 246.94*** 160 .93 .92 .05 .33
African American 240.33*** 160 .95 .94 .05 .74
White 259.58*** 160 .93 .92 .04 .62
Age group1 (18-24) 247.81*** 160 .90 .88 .07  .05*
Age group2 (25-34) 232.80*** 160 .93 .92 .05 .47
Age group3 (35+) 242.04*** 160 .94 .93 .04 .52
Open in a new tab

Note: χ2=Chi-square; df=degree of freedom; CFI = Comparative Fit Index; TLI=Tucker-Lewis index; RMSEA = Root Mean Square Error of Approximation; PCLOSE=Probability of Close Fit (Pr[RMSEA<.05])

Significance tests based on robust maximum likelihood estimates

*

p<.05

**

p<.01

***

p<.001

Acknowledgment

The authors acknowledge research support from Research Grant R01 DA15690 from the National Institute on Drug Abuse of the National Institutes of Health.

Contributor Information

Jiangmin Xu, Research analyst and associate project director in the Center for Interventions, Treatment, and Addiction Research at Wright State University's Boonshoft School of Medicine.

Jichuan Wang, Professor and research director of the Center for Interventions, Treatment, and Addictions Research at Wright State University's Boonshoft School of Medicine.

Richard C. Rapp, Assistant professor and director for Case Management Studies in the Center for Interventions, Treatment, and Addiction Research at Wright State University's Boonshoft School of Medicine. He is also the principal investigator on the Reducing Barriers to Drug Abuse Treatment Services project.

Robert G. Carlson, Director of the Center for Interventions, Treatment, and Addictions Research and professor in the Department of Community Health, Boonshoft School of Medicine. He is a co-principal investigator on the Reducing Barriers to Drug Abuse Treatment Services project.

References

  1. Adair EBG, Craddock SG, Miller HG, Turner CF. Assessing consistency of responses to questions on cocaine use. Addiction. 1995;90:1497–1502. doi: 10.1046/j.1360-0443.1995.901114978.x. [DOI] [PubMed] [Google Scholar]
  2. Allen K. Development of an instrument to identify barriers to treatment for addicted women, from their perspective. International Journal of Addictions. 1994;29(4):429–444. doi: 10.3109/10826089409047391. [DOI] [PubMed] [Google Scholar]
  3. Allen K. Barriers to treatment for addicted African-American women. Journal of the American Medical Association. 1995;87:751–756. [PMC free article] [PubMed] [Google Scholar]
  4. Allen K, Dixon M. Psychometric assessment of the Allen Barriers to Treatment Instrument. International Journal of Addictions. 1994;29(5):545–563. doi: 10.3109/10826089409047399. [DOI] [PubMed] [Google Scholar]
  5. American Psychiatric Association . Diagnostic and statistical manual of psychiatric disorders. 4th ed. American Psychiatric Association; Washington, D. C.: 2000. [Google Scholar]
  6. Appel PW, Ellison AA, Jansky HK, Oldak R. Barriers to enrollment in drug abuse treatment and suggestions for reducing them: Opinions of drug injecting street outreach clients and other system stakeholders. The American Journal of Drug and Alcohol Abuse. 2004;30(1):129–153. doi: 10.1081/ada-120029870. [DOI] [PubMed] [Google Scholar]
  7. Arbuckle JL. AMOS 4.0 user's guide. SmallWaters; Chicago: 1997. [Google Scholar]
  8. Arbuckle JL. AMOS. SmallWaters; Chicago: 2003. [Version 5.0] [Google Scholar]
  9. Beckman L, Amaro H. Personal and social difficulties faced by women and men on entering alcoholism treatment. Journal of Studies on Alcohol. 1986;47:135–145. doi: 10.15288/jsa.1986.47.135. [DOI] [PubMed] [Google Scholar]
  10. Bentler PM. Comparative fit indexes in structural models. Psychological Bulletin. 1990;107:238–246. doi: 10.1037/0033-2909.107.2.238. [DOI] [PubMed] [Google Scholar]
  11. Bentler PM, Bonett DG. Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin. 1980;88(3):588–606. [Google Scholar]
  12. Brown BS, Needle RH. Modifying the process of treatment to meet the threat of AIDS. International Journal of Addictions. 1994;29(13):1739–1752. doi: 10.3109/10826089409047959. [DOI] [PubMed] [Google Scholar]
  13. Browne MW, Cudeck R. Alternative ways of assessing model fit. In: Bollen KA, Long JS, editors. Testing structural equation models. Sage; Newbury Park, CA: 1993. pp. 136–162. [Google Scholar]
  14. Byrne BM. Structural equation modeling with LISREL, PRELIS, and SIMPLIS: Basic concepts, applications, and programming. Lawrence Erlbaum Associates, Inc., Publishers; New Jersey: 1998. [Google Scholar]
  15. Carroll KM, Rounsaville BJ. Contrast of treatment-seeking and untreated cocaine abusers. Archives of General Psychiatry. 1991;49:464–471. doi: 10.1001/archpsyc.1992.01820060044007. [DOI] [PubMed] [Google Scholar]
  16. Cunningham JA, Sobell LC, Sobell MB, Agrawal S, Toneatte T. Barriers to treatment: Why alcohol and drug abusers delay or never seek treatment. Addictive Behaviors. 1993;18(3):347–353. doi: 10.1016/0306-4603(93)90036-9. [DOI] [PubMed] [Google Scholar]
  17. Farabee D, Leukefeld CG, Hays L. Accessing drug abuse treatment: Perceptions of out of-patient injectors. Journal of Drug Issues. 1998;28:381–394. [Google Scholar]
  18. Grant BF. Barriers to alcoholism treatment: Reasons for not seeking treatment in a general population sample. Journal of Studies on Alcohol. 1997;58:365–371. doi: 10.15288/jsa.1997.58.365. [DOI] [PubMed] [Google Scholar]
  19. Hajema K, Knobbed RA, Drop MJ. Social resources and alcohol-related losses as predictors of help-seeking among male problem drinkers. Journal of Studies on Alcohol. 1999;60:120–129. doi: 10.15288/jsa.1999.60.120. [DOI] [PubMed] [Google Scholar]
  20. Hasin DS. Treatment/self-help for alcohol-related problems: Relationship to social pressure and alcohol dependence. Journal of Studies on Alcohol. 1994;55:660–666. doi: 10.15288/jsa.1994.55.660. [DOI] [PubMed] [Google Scholar]
  21. Hingson R, Mangione T, Meyers A, Scotch N. Seeking help for drinking problems: A study in the Boston metropolitan area. Journal of Studies on Alcohol. 1982;43:273–288. doi: 10.15288/jsa.1982.43.273. [DOI] [PubMed] [Google Scholar]
  22. Hser YI, Maglione M, Polinsky ML, Anglin D. Predicting drug treatment entry among treatment-seeking individuals. Journal of Substance Abuse Treatment. 1998;15:213–220. doi: 10.1016/s0740-5472(97)00190-6. [DOI] [PubMed] [Google Scholar]
  23. Hu L, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling. 1999;6(1):1–55. 1999. [Google Scholar]
  24. Janz NK, Becker MH. The health belief model: A decade later. Health Education Quarterly. 1984;11:1–47. doi: 10.1177/109019818401100101. [DOI] [PubMed] [Google Scholar]
  25. Jessup MA, Humphreys JC, Brindis CD, Lee KA. Extrinsic barriers to substance abuse treatment among pregnant drug dependent women. Journal of Drug Issues. 2003;33(2):285–304. [Google Scholar]
  26. Joe GW, Simpson D, Broome KM. Effects of readiness for drug abuse treatment on client retention and assessment of process. Addiction. 1998;93(8):1177–1190. doi: 10.1080/09652149835008. [DOI] [PubMed] [Google Scholar]
  27. Kaskutas LA, Weisner C, Caetano R. Predictors of help-seeking among a longitudinal sample of the general population, 1984-1992. Journal of Studies on Alcohol. 1997;58:155–161. doi: 10.15288/jsa.1997.58.155. [DOI] [PubMed] [Google Scholar]
  28. Lundgren LM, Amodeo M, Ferguson F, Davis K. Racial and ethnic differences in drug treatment entry of injection drug users in Massachusetts. Journal of Substance Abuse Treatment. 2001;21:145–153. doi: 10.1016/s0740-5472(01)00197-0. [DOI] [PubMed] [Google Scholar]
  29. Marlowe DB, Kirby KC, Bonieskie LM, Glass DJ, Dodds LD, Husband SD, Platt JJ, Festinger DS. Assessment of coercive and noncoercive pressures to enter drug abuse treatment. Drug and Alcohol Dependence. 1996;42:77–84. doi: 10.1016/0376-8716(96)01261-6. [DOI] [PubMed] [Google Scholar]
  30. Marsh HW, Balla JR, Hau KT. An evaluation of incremental fit indices: A clarification of mathematical and empirical process. In: Marcoulides GA, Schumacker RE, editors. Advanced structural equation modeling techniques. Erlbaum; Hillsdale, N. J.: 1996. pp. 315–353. [Google Scholar]
  31. McCoy CB, Metsch LR, Chitwood DD, Miles C. Drug use and barriers to use of health care services. Substance Use & Misuse. 2001;36(67):789–806. doi: 10.1081/ja-100104091. [DOI] [PubMed] [Google Scholar]
  32. Melnyk K. Barriers to care: Operationalizing the variable. Nursing Research. 1990;39(2):108–112. [PubMed] [Google Scholar]
  33. Needle R, Fisher DG, Weatherby N, Chitwood D, Brown B, Cesari H, Booth R, Williams ML, Watters J, Andersen M, Braunstein M. Reliability of self-reported HIV risk behaviors of drug users. Psychology of Addictive Behaviors. 1995;9:242–250. [Google Scholar]
  34. Parker H,J, Bakx K, Newcombe R. Living with heroin. Open University Press; Philadelphia: 1988. [Google Scholar]
  35. Pfeiffer W, Feuerlein W, Brenk-Schulte E. The motivation of alcohol dependents to undergo treatment. Drug and Alcohol Dependence. 1991;29:87–95. doi: 10.1016/0376-8716(91)90025-t. [DOI] [PubMed] [Google Scholar]
  36. Power R, Jones S, Kearns G, Ward J, Perera J. Drug user networks, coping strategies and HIV prevention in the community. Journal of Drug Issues. 1995;25(3):565–581. [Google Scholar]
  37. Rapp RC, Xu J, Carr CA, Lane DT, Wang J, Carlson RG. Factor analysis of treatment barriers identified by substance abusers following assessment at a centralized intake unit. Journal of Substance Abuse Treatment. 2006;30(3):227–235. doi: 10.1016/j.jsat.2006.01.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Rounsaville BJ, Kelber HD. Untreated opiates addicts: How do they differ from those seeking treatment? Archives of General Psychiatry. 1985;42:1072–1077. doi: 10.1001/archpsyc.1985.01790340050008. [DOI] [PubMed] [Google Scholar]
  39. Sheehan M, Oppenheimer E, Taylor E. Why drug users sought help from one London drug clinic. British Journal of Addiction. 1986;81(6):765–775. doi: 10.1111/j.1360-0443.1986.tb00404.x. [DOI] [PubMed] [Google Scholar]
  40. Siegal HA, Falck RS, Wang J, Carlson RG. Predictors of drug abuse treatment entry among crack-cocaine smokers. Drug & Alcohol Dependence. 2002;68(2):159–166. doi: 10.1016/s0376-8716(02)00192-8. [DOI] [PubMed] [Google Scholar]
  41. Sobell LC, Sobell MB, Toneatto T. Recovery from alcohol problems without treatment. In: Heather N, Miller WR, Greeley J, editors. Self-control and the addictive behaviors. Maxwell MacMillan; New York: 1992. pp. 198–242. [Google Scholar]
  42. SPSS Inc. SPSS for Windows. SPSS Inc.; New York: 2005. [Version 14.0] [Google Scholar]
  43. Steiger JH. Tests for comparing elements of a correlation matrix. Psychological Bulletin. 1980;87:245–251. [Google Scholar]
  44. Tsogia D, Copello A, Orford J. Entering treatment for substance misuse: A review of the literature. Journal of Mental Health (UK) 2001;10(5):481–499. [Google Scholar]
  45. Tucker JA. Predictors of help-seeking and the temporal relationship of help to recovery among treated and untreated recovered problem drinkers. Addiction. 1995;90:805–809. doi: 10.1046/j.1360-0443.1995.9068057.x. [DOI] [PubMed] [Google Scholar]
  46. Tucker JA, Vuchinich RE, Rippens PD. A factor analytic study of influences on patterns of help-seeking among treated and untreated alcohol dependent persons. Journal of Substance Abuse Treatment. 2004;26:237–242. doi: 10.1016/S0740-5472(03)00209-5. [DOI] [PubMed] [Google Scholar]
  47. Vaughn T, Sarrazin MV, Saleh SS, Huber DL, Hall JA. Participation and retention in drug abuse treatment services research. Journal of Substance Abuse Treatment. 2002;23:387–397. doi: 10.1016/s0740-5472(02)00299-4. [DOI] [PubMed] [Google Scholar]
  48. Weisner C. Toward an alcohol treatment entry model: A comparison of problem drinkers in the general population and in treatment. Alcoholism: Clinical and Experimental Research. 1993;17:746–752. doi: 10.1111/j.1530-0277.1993.tb00833.x. [DOI] [PubMed] [Google Scholar]
  49. Wilsnack S. Barriers to treatment for alcoholic women. Addiction & Recovery. 1991;7&8:10–12. [Google Scholar]

RESOURCES