Addictive Behaviors 39 (2014) 15001503

Contents lists available at ScienceDirect

Addictive Behaviors

Short Communication

A Rasch analysis of the Harm Reduction Self-Efcacy Questionnaire

in Portugal
Paulo Lopes a,, Gerardo Prieto b, Ana R. Delgado b
a
Universidade Lusfona de Humanidades e Tecnologias, COPELABS, Centro de Estudos em Psicologia Cognitiva e da Aprendizagem, Ares do Pinhal, Portugal
b
Universidad de Salamanca, Spain

H I G H L I G H T S

We analyze the metric quality of the Harm Reduction Self-Efcacy Questionnaire.

We use the Rating Scale Model, an extension of the Rasch Model for polytomous items.
The 11-category response system did not work adequately in any of the situations.
After collapsing the categories into 4, the HRSEQ scores show psychometric adequacy.

a r t i c l e i n f o a b s t r a c t

Available online 27 May 2014 Introduction: A high degree of self-efcacy is required from drug users if they are to adopt long-term abstinence-
maintaining or harm-reducing behaviors. Our objective was to analyze the psychometric quality of the Harm
Keywords: Reduction Self-Efcacy Questionnaire (HRSEQ) with the Rating Scale Model (RSM), an extension of the Rasch
Drug dependence Model for polytomous items.
Harm reduction
Methods: Participants were 419 drug dependents from both harm reduction low threshold methadone program
Rasch Rating Scale Model
and therapeutic community programs in Portugal (84.5% male).
Self-efcacy
Results: Indicate that the original 11-category response system did not work adequately in any of the HRSEQ
situations (Withdrawal, Negative emotions and Social pressure). There was no gender-related or group-related
DIF and thus the ndings of neither group nor gender-related difference in harm reduction self-efcacy cannot
be plausibly attributed to that technical problem.
Conclusions: After recoding the rating scales by collapsing the eleven categories into four, the HRSEQ scores show
good model t and psychometric adequacy.
2014 Elsevier Ltd. All rights reserved.

1. Introduction Self-efcacy is one of the personal factors inuencing recovery from

Since the end of the twentieth century, opiates have been considered
as one of the major risk factors of disease in the world (Ezzati et al.,
2002; Lopez, Krueger, & Walters, 2010). In the development of pro-
grams to prevent drug use, setting realistic goals and selecting outcome
measures related to program activities are the keys to a meaningful
evaluation (Midford, 2009). These programs are linked to a change in
attitude, accepting the drug users's own lifestyle (Fuchs & Degkwitz,
1995). Even though abstinence is the optimal objective, it is not the
only acceptable one. Paying attention to quality of life and preventing
undesirable consequences are important goals too. But a high degree
of self-efcacy is required of drug users if they are to adopt long-term
abstinence-maintaining or harm-reducing behaviors.
Corresponding author at: COPELABS, University Lusfona, Campo Grande, 376, 1749-
024 Lisbon, Portugal.
E-mail address: paulo.jorge@ulusofona.pt (P. Lopes).
drug dependence. The term is understood as the drug users' level of
condence in their abilities to adopt behaviors that are useful in main-
taining drug abstinence (Bandura, 1995, 1999). In substance abusers,
the role of self-efcacy is related to: a) deciding to change their relation-
ship with drugs; b) reducing substance use during therapy; and c) pre-
serving therapy achievements (Levin, Ilgen, & Moos, 2007).
Adequate measurement scales are needed in order to assess self-
efcacy in recovering drug addicts. Having good psychometric instru-
ments will allow us to assess the role of the construct regarding efcacy
(Lopes, Prieto, Delgado, Gamito, & Trigo, 2010, 2011). Thus, the objec-
tive of this research is to test, by means of the Rating Scale Model
(RSM), the psychometric quality of the Harm Reduction Self-Efcacy
Questionnaire (HRSEQ; Phillips & Rosenberg, 2008), a self-report instru-
ment designed to assess the condence of (injecting) drug dependents
in employing certain harm reducing strategies.
The RSM is an extension of the Rasch Model (RM) for polytomous
items. These models (Andrich, 1978; Wright & Mok, 2004) are especially

http://dx.doi.org/10.1016/j.addbeh.2014.05.014
0306-4603/ 2014 Elsevier Ltd. All rights reserved.
 P. Lopes et al. / Addictive Behaviors 39 (2014) 15001503 1501

recommended for testing psychological assessment instruments (Prieto,
Delgado, Perea, & Ladera, 2010), because of properties such as conjoint
measurement of persons and items, which allows variable-referred
norms to be used to interpret scores (as well as the traditional group
norms). Variable-referred norms are really useful in drug treatment con-
texts, as their scores reveal what specic behaviors a patient can adopt.
Choosing a RM is justied by its optimal metric properties (Wilson,
2005). The RSM is empirically useful in determining the quality of re-
sponse categories in Likert-type scales (Bond & Fox, 2007).
2. Methods
2.1. Participants
Sixty-four questionnaires were rst eliminated for lacking answers
to more than 20% of the items. The sample was thus composed of 419
drug dependents from both Harm Reduction Low Threshold Methadone
Program (MP) and therapeutic community programs (TC). Most partic-
ipants were male (84.5%; Mean age: 34.74, SD = 7.03). Female mean
age was 37.74 (DT = 7.03). No gender differences were found in the
characteristics of the sample associated with drug use and therefore
data from both male and female participants have been jointly analyzed.
2.2. Materials
The HRSEQ is composed of 15 polytomous items that respondents
must answer while assuming they are in one of three hypothetical situ-
ations: Withdrawal (W), Negative Emotions (NE), and Social Pressure
(SP) to use drugs in unsafe ways. Participants are asked to evaluate
their condence on a 11-category Likert scale (from 0 = Not at all con-
dent to 10 = Very condent, in each high-risk situation). Scores for
each situation are obtained by adding the 15 item scores. The original
validation study, carried out in the context of the Classical Test model,
revealed high internal consistency of the scores in the three situations
(Cronbach range = .82 to .91). High correlations between situations
were also observed (.67 to .79).
2.3. Procedure
In order to adapt the HRSEQ, the original English version was rst
translated into Portuguese and then revised by three experts in the
drug dependence eld. Participation in the study was voluntary, and an-
onymity and condentiality were guaranteed, as well as the possibility
of leaving the study at any moment. Data were collected in two different
programs: City of Lisbon MP and TC. Data analysis was carried out with
Winsteps 3.69.1.9 (Linacre, 2009).
3. Data analysis
explain a substantial percentage of the variance (at least 20% according
to Reckase, 1979) and the rst component of the residuals amounts to
less than 10% of the variance.
Person and item measure reliability can be estimated with the stan-
dard error and with group statistics called Person Separation Reliability
and Item Separation Reliability whose range is 01 (it should be over
.70). Group-related and gender-related Differential Item Functioning
(DIF) has also been tested in order to corroborate the metric equiva-
lence of the HRSEQ scores for males and females and for MP and CT
groups. An HRSEQ item would suffer from gender-related DIF if males
and females with the same level of self-efcacy had different probabili-
ties of answering that item with the same category. In that case, the lo-
cation parameter would be different for the two genders. Thus DIF is
tested by calculating the difference between male and female item loca-
tion estimators (Dm Df) controlling for the difference in the latent var-
iable. Statistically signicant differences (Bonferroni corrected) below
.50 are considered negligible from a substantive point of view. If any
of the t-tests in the list has p b .05/(number of t-tests in the list), then
the hypothesis of no DIF is rejected. Eliminating items with DIF is rec-
ommended because of the lack of generalized validity.
4. Results
First, the metric quality of the HRSEQ standard version 11-category
system was tested. Table 1 shows step values (Fk) between the 11 suc-
cessive categories in the three HRSEQ situations (W, NE and SP). The
lack of monotonic ordering of these steps indicates category
malfunctioning: some of the categories are never modal along the
continuum.
To get a metrically acceptable category system, a four-category system
was constructed by collapsing the original categories as follows: 0 = Not
at all condent (0 + 1), 1 = Low condence (2 + 3 + 4), 2 = Somewhat
condent (5 + 6 + 7) and 3 = Very condent (8 + 9 + 10).
It can be observed in Table 1 that the new category system works
well. Thus, subsequent HRSEQ score analyses were carried out with
measures calculated with the four-category response system.
Results from the principal component analysis of residuals for the
scores indicate that the assumption of essential unidimensionality can
be empirically corroborated given that there is no dimension in the re-
siduals explaining more than 10% of the variance (7.9, 7.6 and 7.6 are
the percentages for W, NE and SP situations, respectively) and the per-
centage of variance explained by the Rasch dimension is over 20% (44.3,
47.1 and 49.2 for W, NE and SP situations, respectively).
Table 2 shows the HRSEQ item location, standard error and outt
values. Standard error values are low, indicating that parameter estima-
tion was precise. Model t is adequate: no item outt is over 2 (severe
mist) and average outt values are close to 1.

Empirical verication of the category system requires that the steps

Table 1
be monotonically ordered, which reveals whether (1) the category
HRSEQ 11- and 4-category step calibrations by situation.
structure is meaningful, i.e., the higher categories correspond to higher
person level in the attribute, and (2) all categories are functioning, i.e., Category Withdrawal Negative emotions Social pressure
they are all modal at some point in the latent continuum. Collapsing ad- 0
jacent categories is the empirical solution when steps are disordered 1 .09 .05 .06
(Andrich, 2013; Linacre, 2002). A detailed description of the model 2 .22 .38 .33
3 .13 .29 .27
can be found in Wright and Masters (1982). 4 .05 .09 .03
To prot from the advantages of RSM, the data must t to the model 5 .74 .71 .66
predictions. First, items and persons that do not t, if any, must be detect- 6 1.01 .89 .82
ed. This can be done calculating the outt mean square statistic that aver- 7 .10 .22 .15
8 .11 .02 .01
ages over the response residuals and whose expected value is 1. Mist is
9 .54 .56 .63
detected when outt values are over 1.5, but it is only considered as severe 10 .59 .46 .24
(and measure-disturbing) when outt values are over 2 (Linacre, 2009). 0
A specic test of the data t to the unidimensionality assumption 1 .63 .84 .86
was carried out, a principal component analysis of residuals. It is not 2 .01 .03 .05
3 .64 .88 .92
risky to assume essential unidimensionality when the Rasch measures
1502 P. Lopes et al. / Addictive Behaviors 39 (2014) 15001503

Table 2
Item t (outt), location (D) and standard error (SE) by situation.

Item Outt (W) D (W) SE (W) Outt (NE) D (NE) SE (NE) Outt (SP) D (SP) SE (SP)

1 1.67 .43 .06 1.86 .43 .06 1.66 .39 .06

2 1.26 .09 .06 1.32 .21 .06 1.17 .16 .07
3 .99 .25 .06 1.07 .25 .06 1.11 .37 .06
4 1.09 .20 .06 .96 .38 .06 1.05 .34 .07
5 .92 .53 .06 .91 .45 .07 .85 .44 .07
6 1.11 .25 .06 .85 .18 .06 .89 .31 .07
7 .77 .21 .06 .82 .26 .06 .79 .20 .07
8 .82 .32 .06 .81 .28 .06 .89 .21 .07
9 1.21 .03 .06 1.12 .02 .06 1.06 .02 .07
10 1.21 .08 .06 1.16 .15 .06 1.24 .08 .07
11 .82 .64 .06 .80 .64 .07 .78 .74 .07
12 .78 .00 .06 .87 .12 .06 .81 .21 .07
13 1.00 .23 .06 1.02 .19 .06 1.04 .20 .07
14 .78 .19 .06 .78 .32 .06 .82 .37 .07
15 .88 .41 .06 1.00 .31 .06 1.21 .30 .07
Mean 1.02 .00 .06 1.02 .00 .06 1.03 .00 .07
Max. 1.67 .43 .06 1.86 .43 .07 1.66 .39 .07
Min. .77 .64 .06 .78 .64 .06 .78 .74 .06

Note that the item location parameters are not very variable. How-
ever, low dispersion does not imply low person score reliability because
items are polytomous. Item ordering is similar in the three situations
and therefore the correlations between location parameters are very
high (from .95 to .98).
Gender-related DIF analysis showed that the difference between lo-
cation parameters for males and females was low: from .22 to .24
(W), from .25 to .20 (NE), and from .29 to .27 (SP). Given the
small size of these differences, we can conclude that there is no
gender-related DIF. Neither was a group-related DIF found: the differ-
ence between location parameters for MP and TC was low: from .16
to .30 (W), from .41 to .39 (NE), and from .45 to .32 (SP).
As to person score statistics, the percentage of persons with outt
over 2 is small, below 10%, and average outt values are close to unity
in the three situations. Most scores t the RSM. It must be highlighted
that score reliability is high: Person Separation Reliability values are
context: close to the 6.2% (CT) and to the 0.8% (MP) program population
in 2008. There was no gender-related or group-related DIF and there-
fore the ndings of no group or gender-related difference in harm re-
duction self-efcacy cannot plausibly be attributed to that technical
problem.
The main limitation of our study is the small size of the female sam-
ple. This is due to the fact that there actually are fewer drug dependent
women than men. Another problem, always present in this kind of re-
search, is social desirability, which could have contaminated the an-
swers and threatened the validity of some of our inferences: e.g., that
there are no differences between situations. Considering all the above,
our principal conclusion is practical: future studies carried out with
the HRSEQ should employ the four-category response system.
Role of funding sources
None.

over .85 and Cronbach alphas over .90.

Contributors
Finally, the difference between males and females was not signicant
The three authors designed the study. The rst author carried out this investigation for
either in W (t = .07, p = .94), NE (t = .95, p = .34) or SP (t = 1.74, his doctoral thesis, supervised by the other two authors. They all contributed to and have
p = .08). Neither are the differences signicant between MP and CT: W approved the nal manuscript.
(t = .66, p = .94), NE (t = .66, p = .51) and SP (t = 1.21, p = .23).
Conict of interest
All authors declare that they have no conicts of interest.
5. Discussion

The psychometrical quality of the HRSEQ measures has been tested
by means of the RSM. The original HRSEQ 11-category response system
is psychometrically decient in the three hypothetical situations (W, NE
and SP): a monotonic increase in the steps was not found. Some catego-
ries were never modal along the variable. To improve the HRSEQ re-
sponse category functioning, we have proposed using just four
categories. Data reanalysis showed that the four-category system
works properly and so we recommend using this system in future
HRSEQ applications. No items with severe mist were found and essen-
tial unidimensionality was empirically corroborated.
As to item locations, they were close to the variable center (average
difculty). The easiest HRSEQ items were those related to using new
materials (needle, spoon, lter or cotton). They do not serve to signal
high self-efcacy. In contrast, items related to dose reduction or to the
use of rubber tourniquet were the most difcult ones, requiring more
self-efcacy from drug users. Person score was estimated with high pre-
cision in the three situations, suggesting that just 15 polytomous items
with four-category response system are enough to reliably measure
self-efcacy in harm reduction contexts.
Concerning the assessed sample, and taking into account how dif-
cult it is to reach the drug user population, we wish to highlight the
fact that this sample is sufciently representative of the Portuguese
References
Andrich, D. (1978). A rating scale formulation for ordered response categories.
Psychometrika, 43, 561573.
Andrich, D. (2013). An expanded derivation of the threshold structure of the polytomous
Rasch model that dispels any Threshold Disorder Controversy. Educational and
Psychological Measurement, 73, 78124.
Bandura, A. (1995). On rectifying conceptual ecumenism. In J. E. Maddux (Ed.), Self-efca-
cy, adaption, and adjustment: Theory, research, and application (pp. 347375). New
York: Plenum Press.
Bandura, A. (1999). A sociocognitive analysis of substance abuse: An agentic perspective.
Psychological Science, 10, 214217.
Bond, T. G., & Fox, C. M. (2007). Applying the Rasch model (2nd ed.). Mahwah, NJ: LEA.
Ezzati, M., Lopez, A. D., Rodgers, A., Hoorn, S. V., & Murray, C. J. L.Comparative Risk
Assessment Collaborating Group. (2002). Selected major risk factors and global and
regional burden of disease. Lancet, 360, 13471360.
Fuchs, W. J., & Degkwitz, P. (1995). Harm reduction in Europe-trend, movement or
change of paradigma. European Addiction Research, 1, 8185.
Levin, C., Ilgen, M., & Moos, R. (2007). Avoidance coping strategies moderate the relation-
ship between self-efcacy and 5-year alcohol treatment outcomes. Psychology of
Addictive Behaviors, 21, 108113.
Linacre, J. M. (2002). Optimizing rating scale category effectiveness. Journal of Applied
Measurement, 3, 85106.
Linacre, J. M. (2009). A User's Guide to Winsteps Ministep-Rasch-Model Computer Programs.
Chicago: Winsteps.com.
Lopes, P., Prieto, G., Delgado, A., Gamito, P., & Trigo, H. (2010). Rasch-modeling the Portu-
guese SOCRATES in a clinical sample. Psychology of Addictive Behaviors, 24, 355359.
Lopes, P., Prieto, G., Delgado, A., Gamito, P., & Trigo, H. (2011). Anlise das qualidades
psicomtricas da URICA utilizando o Modelo de Rasch. Toxicodependncias, 17, 4764.
 P. Lopes et al. / Addictive Behaviors 39 (2014) 15001503
Lopez, W. D., Krueger, P. M., & Walters, S. T. (2010). High-risk drug use and sexual behav-
iors among out-of-treatment drug users: An aging and life course perspective.
Addictive Behaviors, 35, 432437.
Midford, R. (2009). Drug prevention programmes for young people: Where have we been
and where should we be going? Addiction, 105, 16881695.
Phillips, K. T., & Rosenberg, H. (2008). The development and evaluation of the Harm Re-
duction Self-Efcacy Questionnaire. Psychology of Addictive Behaviors, 22, 3646.
Prieto, G., Delgado, A. R., Perea, M. V., & Ladera, V. (2010). Scoring neuropsychological
tests using the Rasch model: An illustrative example with the ReyOsterrieth com-
plex gure. The Clinical Neuropsychologist, 24, 4556.
1503
Reckase, M. (1979). Unifactor latent trait models applied to multifactor tests: Results and
implications. Journal of Educational Statistics, 4, 207230.
Wilson, M. (2005). Constructing measures. Mahwah, NJ: LEA.
Wright, B. D., & Masters, G. N. (1982). Rating scale analysis. Chicago: Mesa Press.
Wright, B. D., & Mok, M. (2004). An overview of the family of Rasch measurement models.
In E. V. Smith, & R. M. Smith (Eds.), Introduction to Rasch Measurement. Maple Grove:
JAM Press.