Abstract

Background:

Early identification of nicotine dependence symptoms is essential for tobacco control and prevention strategies. The Hooked On Nicotine Checklist (HONC) is widely used to assess early symptoms of nicotine dependence. This study aimed to translate, culturally adapt, and evaluate the psychometric properties of a Moroccan Arabic version of the HONC.

Methods:

A cross sectional study was conducted among a sample of 234 Moroccan smokers. The HONC was translated and adapted following established guidelines. Descriptive statistics and item–total correlations were examined. Internal consistency was assessed using Cronbach’s alpha and McDonald’s omega. Dimensionality was evaluated using confirmatory factor analysis (CFA) with the WLSMV estimator. IRT analyses were conducted using a two-parameter logistic (2PL) model. Assumptions of unidimensionality, monotonicity, and local independence were systematically tested. Item discrimination and difficulty parameters, item characteristic curves, and test information functions were estimated.

Results:

Item–total correlations were moderate-to-high (range: 0.45–0.77). Internal consistency was excellent (Cronbach’s α = 0.905; ωt = 0.908). The one-factor CFA showed good model fit (χ2/df = 1.34; CFI = 0.998; TLI = 0.998; RMSEA = 0.038; SRMR = 0.059), with substantial standardized loadings (0.629–0.952). Monotonicity and local independence assumptions were fulfilled. Discrimination parameters were generally high (a = 1.40–4.69) and difficulties were mostly negative (b ≈ −0.64 to 0.00), suggesting that the scale is most informative at lower to moderate levels of nicotine dependence.

Conclusion:

The Moroccan Arabic HONC is a reliable and valid measure of early nicotine dependence. It can be confidently used for research, screening, and tobacco control interventions.

Introduction

Tobacco use is one of the main avoidable risk factors of non-communicable diseases. In 2020, 22.3% of the global population used tobacco, with nearly 5 times more prevalent use in men compared with women (36.7 and 7.8%, respectively). Around 80% of the 1.3 billion tobacco users worldwide reside in low- and middle-income countries, enduring, thus, the heaviest burden of illness and death linked to tobacco use (1). In Morocco, according to the latest STEPS survey report, 13.4% (12.2–14.6) of Moroccans, aged 18 and older, were current tobacco users. The prevalence was much higher in men (26.9%) than in women (0.4%) (2). These figures underscore the continuing public health importance of tobacco dependence assessment in Morocco, especially considering its implications for cessation, relapse prevention, and the tailoring of intervention strategies.

Tobacco use exposes its users to a wide spectrum of substances, including nicotine, the main chemically active component of tobacco, characterized by its highly addictive potential and considered as the major dependence producing agent, responsible for failure of quitting attempts and maintained use (3, 4). “Loss of autonomy” is a common feature among definitions of substance dependence. Based on this concept, the Hooked On Nicotine Checklist (HONC) was developed (5). The HONC is a theory-derived widely used measurement tool, consisting of 10 items that assess lost autonomy. Diffranza’s Autonomy Theory (5, 6) advances that the onset of dependence occurs when individuals are “hooked” or lose their autonomy over tobacco use (when quitting becomes difficult), and integrates three mechanistic theories of independence; the Self-Medication Theory (7), which postulates that dependence is the result of the use of psychoactive substances to “self-medicate” or to cope with negative symptoms associated with a mental health condition or other issues (HONC item 5), the Incentive-Sensitization Theory (8), which suggests that repeated PASs use renders neural system hypersensitive (‘phenomenon of ‘sensitization’). This hypersensitization leads to generation of “incentive salience” toward drugs and their associated stimuli which become highly salient and “wanted,” causing thus craving, and the Negative Reinforcement Theory, which posits that individuals continue to use drugs to alleviate uncomfortable emotional states, including those associated with withdrawal (9).

The original version of the HONC (5) has a dichotomous (Yes/No) response format, but a multiple response choices format was afterwards proposed by O’Loughlin et al. (10). Each HONC item reflects a ND symptom. Affirmation of any item (i.e., positive HONC) implies loss of full autonomy over tobacco use (diminished autonomy), while a negative HONC (a score of zero) means enjoying full autonomy over tobacco use and higher scores signify more severe loss of autonomy levels, which indicates greater ND. The original HONC has high internal consistency (=0.94) and good test–retest reliability (kappa = 0.61 (95% CI, 0.35–0.87)). Concurrent validity was supported by the statistically significant positive associations between the total number of HONC endorsed symptoms and the maximum amount smoked (r = 0.65, p < 0.001), and maximum frequency of smoking (r = 0.79, p < 0.001). Additionally, the agreement between endorsement of any HONC item and continued smoking throughout the period of the longitudinal study (OR = 44, 95% CI, 17–114) indicated the HONC’s predictive validity (5).

HONC has been initially developed to be used among adolescent population (5), then, validated among adult population (11). Furthermore, HONC has shown its performance with several forms of nicotine delivery; cigarettes (5, 11, 12), smokeless tobacco (13), waterpipe (14) and has been recently used also for e-cigarettes use (15–18). Moreover, HONC has proven its efficacy to identify early symptoms of dependence even at low-dose and among occasional smokers (19–21).

In addition to established ND measures such as the Fagerström instruments, the HONC offers a complementary assessment focus. Whereas Fagerström-based scales rely on consumption-related indicators (such as, cigarette quantity, smoking frequency, and time to first cigarette) to index dependence severity (22), the HONC was developed to assess a clearly defined construct-diminished autonomy over tobacco use-using simple, easily interpretable symptom counts and a natural cutpoint (0 symptoms). In a direct comparison with the Fagerström Test for Nicotine Dependence (FTND) in adult smokers, the HONC showed advantages including greater sensitivity to the onset and low levels of dependence and clearer interpretability, and was described as particularly useful when cigarette consumption is low (12). Consistently, evidence suggests that diminished autonomy measured by the HONC can appear very early in the smoking trajectory, including among very infrequent smokers and even after minimal lifetime exposure, often before the onset of daily smoking (23). For this reason, the HONC may be especially relevant when the objective is not only to quantify established dependence severity, but also to detect early loss of autonomy that may be insufficiently captured by consumption-weighted instruments.

Given the emphasized need for valid and reliable instruments adapted to the Moroccan context, and considering the advantageous HONC’s flexibility (applicability to many tobacco forms and different populations), brevity and ease of administration, we consider that it will be practical and valuable for use by clinicians and research investigators. Accordingly, the aim of the present study was to translate and culturally adapt the HONC into Moroccan Arabic and to examine its psychometric properties in a sample of Moroccan smokers. More specifically, we sought to examine its internal consistency, test the fit of its hypothesized one-factor structure, assess item-level psychometric performance using item response theory, and evaluate criterion validity through its association with the Fagerström Test for Cigarette Dependence.

MethodsCross-cultural adaptation

Permission to translate, adapt, and use the HONC was obtained from the scale developer. Subsequently, we carried out the cross-cultural translation and adaptation by referring to the recommendations of Hambleton et al. (24) and Beaton et al. (25).

Two native Moroccan speakers with high proficiency in English (English teachers with over 20 years of experience) were summoned for the translation of the HONC. They independently translated the original HONC into Moroccan dialect. Then, the two versions (T1 and T2) resulting from translations were reviewed in a committee meeting, comprising researchers involved in the study, translators, a psychologist, a psychiatrist, and an epidemiologist. Differences were identified and reduced through discussions in order to produce a consensus version. After that, two other independent translators (having the same profile as the first ones), unaware of the original instrument, did the back translation of the consensus version, so that we obtained two back-translated English versions (BT1 and BT2). A second committee meeting was held to compare BT1 and BT2 with the original version, assess equivalences and whether the items preserved their original meaning. Revisions were made through discussions, and consensus was reached producing an approved pre-final Moroccan version. The latter was, subsequently, subjected to pilot testing among a 20-participant group. And since all items were judged understandable, and nothing was mentioned as ambiguous or confusing, there were no adjustments made to the scale after the pilot test.

ValidationParticipants

Sample size adequacy was assessed according to methodological recommendations. A commonly cited rule of thumb for factor analysis recommends a person-to-item ratio of at least 10:1 (26). In addition, other authors have suggested that an overall sample size of 200 to 300 participants is appropriate for factor analysis in general (27). For confirmatory factor analysis with binary data estimated using the WLSMV estimator, previous studies have recommended sample sizes exceeding 200 participants (28, 29), while others have suggested a broader range of 200 to 500 participants (30).

For item response theory, sample size requirements are context -dependent and influenced by multiple factors, including item type, the assumed-response model, dimensionality of the model, etc. (31). Published guidance indicates that, for simple Rasch models, sample sizes as small as 100 may be adequate; however, for more complex IRT models, sample sizes generally range from 200 to 500 (32). In addition, some IRT textbooks provide general recommendations for different models, often suggesting minimum sample sizes of approximatively 250 to 500 respondents (31). Sample size requirements also vary according to the underlying purpose of the analysis: larger samples are needed when extremely accurate item-parameters estimates are desired, whereas smaller samples could be sufficient in preliminary evaluations of questionnaire properties (32).

Based on these guidelines, the sample size in the present study (N = 234) was considered adequate for CFA and fell within the lower range reported for IRT analyses.

The validation study was conducted in the addiction treatment center in Fez city. Consecutive current smokers who presented themselves at the center for smoking cessation treatment were recruited. The volunteer smokers were included if they met the following eligibility criteria: (1) current smoking; (2) aged 18 and above; (3) signed the written informed consent. Patients with serious mental illness (such as schizophrenia and bipolar disorder) were excluded, as were participants reporting current use of other psychoactive substances or polysubstance use.

The study period extended from September 2021 to July 2022.

Measures

A self-administered anonymous questionnaire was used to collect data. It required between 5 to 10 min for completion. For the limited number of illiterate participants, the questionnaire was filled out by an investigator through a face-to-face interview. The questionnaire consisted of two sections. In the first one, participants were asked to report their sex, age, area of residence, matrimonial status, and educational level, while in the second one, the following scales were implemented:

HONC: which was subject to validation. In the present study, we used the 10-item dichotomously scored version, as was originally conceived. We preferred it to the multiple-choice version, because it is easily applicable and provides direct information as the total score reflects the number of occurring symptoms. Also, in contrast with what is commonly believed, the multiple-choice version of the HONC did not improve the psychometric properties of the scale (10).

FTCD: to investigate criterion validity, we used the Fagerström Test for Cigarette Dependence (FTCD), originally called the Fagerström Test for Nicotine Dependence (FTND) (33, 34), which is intended to provide a measure of nicotine dependence associated with cigarette smoking. The instrument comprises 6 items that assess the number of cigarettes smoked daily (item 1), time to the first cigarette after waking up in the morning (item 2), the cigarette that the smoker would prefer not to give up (item 3), the time of the day when the smoker smokes the most (item 4), the smoker’s inability to abstain from smoking in places where it is prohibited (item 6) and when he is sick (item 5). The three dichotomous items are scored 0 (No) and 1 (Yes), while the other multiple-choice items are scored from 0 to 3. The test’s total score ranges from 0 to 10; with 0 to 2 indicating very low dependence; 3 to 4 indicating low dependence; 5 indicating medium dependence; 6 to 7 indicating high dependence; and 8 to 10 a very high dependence.

We used the test in its Arabic version, translated and validated by Kassim et al. (35).

Ethics

The study was approved by the hospital-university ethics committee of Sidi Mohamed Ben Abdellah University (N° 17/21) on April 5, 2022. All participants were informed about the purpose and procedures of the study, their right to decline to participate or withdraw at any time without experiencing any consequences for their access to, timing of, or quality of smoking cessation treatment, and they provided written informed consent. Anonymity and confidentiality of all participants were ensured throughout the study.

Statistical analysis

The psychometric assessment involved Confirmatory Factor Analysis (CFA) and Item Response Theory (IRT) analysis. All the analyses were performed using R (version 4.5.1; R Core Team, 2025).

First, we provided descriptive statistics for the HONC, in terms of item endorsement rates (means), standard deviations (SD), and corrected item-total correlations, to obtain some preliminary information on participants’ responses and examine how each item correlates to other items in the instrument.

Following this, we analyzed the factor structure of the adapted version of the HONC. Because the HONC has a well-established unidimensional factor structure in its original version and throughout many validation studies (5, 12, 36), we did not perform an exploratory factor analysis. Instead, we conducted a CFA to assess the fit of a single-factor model to the data in our context. This approach is consistent with methodological guidance, as CFA is appropriately used when the factor structure is specified a priori based on theory and/ or prior empirical evidence (37, 38). CFA, implemented in the lavaan package (v0.6–20) (39), was set up with the Weighted Least Squares Mean- and Variance-Adjusted (WLSMV) estimator for categorical indicators (based on tetrachoric correlations for binary items). The goodness of model fit was evaluated using the relative chi-square (χ2/df) (values around 3 or lower indicate better fit), CFI (comparative fit index≥ 0.95), TLI (Tucker-Lewis Index ≥ 0.95), RMSEA (root mean square error of approximation ≤0.08) and SRMR (Standardized Root Mean Square Residual ≤0.08) (40, 41). In addition, convergent validity and construct reliability of the latent factor were assessed using average variance extracted (AVE) and composite reliability (CR), calculated from the standardized factor loadings and corresponding error variance values. AVE values ≥ 0.50 were considered indicative of adequate convergent validity, and CR values ≥ 0.70 were considered indicative of satisfactory construct reliability (41).

Criterion validity was tested by assessing the correlation between the HONC’s adapted version total score and the FTCD total score. The association was evaluated using the Spearman’s rank correlation coefficient (Rho). Statistical significance was set at p < 0.05. For interpretation, the following criteria were used; correlations were considered negligible (0.00–0.10), weak (0.10–0.39), moderate (0.40–0.69), strong (0.70–0.89), or very strong (0.90–1.00) (42).

As a subsequent step, we evaluated the main IRT assumptions of unidimensionality, monotonicity and local independence.

Unidimensionality refers to the fact that only one ability is measured by a set of items that make up the test. However, test performance is always influenced by several cognitive, psychological, and test-taking factors, at least to some extent, making it impossible to strictly meet this assumption. For IRT applications, to sufficiently satisfy the unidimensionality assumption, the presence of a “dominant” factor or component that influences test performance is required (43). In this study, unidimensionality was supported by CFA.

Monotonicity means the increasing probability of a higher item score as the level of the latent trait increases, ensuring that item responses meaningfully reflect variations in the underlying construct (44). It was examined using Mokken analysis, with the mokken package (v3.1.2) (45). Scalability coefficients (Mokkens’s H) were calculated at the scale (H) and item (Hi) levels. A Loevinger H coefficient < 0.30 was deemed inadequate (46).

Local independence is the statistical independence of examinees’ responses to any pair of items when the abilities influencing test performance are held constant (43). Investigation of local independence was done with Yen’s Q3 statistic (47). Adjusted Q3 values greater than 0.20 are considered violations of local independence, according to standard guidelines (48).

After unidimensionality was confirmed, IRT analyses were conducted using the mirt package (v1.45.1) (49) in R. To estimate item discrimination (a) and item difficulty (b) parameters for each HONC item, a two-parameter logistic (2PL) model was used. The fit of the 2PL model was compared to the Rasch (1PL) model using the likelihood-ratio test (LR test). Item Characteristic Curves (ICCs), item information functions, and the Test Information Function (TIF) were analyzed to evaluate measurement precision across latent trait levels.

The reliability of the data was assessed through Cronbach’s alpha (α) (50) and McDonald’s omega total (ωt) (51), computed from a one-factor model using the psych package (v2.5.6) (52). Additionally, IRT-derived marginal reliability was also computed.

ResultsBackground characteristics of participants

The participants’ background characteristics are summarized in Table 1. The mean age was 28.09 ± 9.24 years (range 18–61). Young people (aged 30 or below) constituted more than half of the study population (66.7%). Overall, our sample was predominantly male (more than 85%), single (75%), educated (only 2% who were illiterate) and living in urban area (more than 80%).

Variablen (%)SexMale203 (86.8)Female31 (13.2)Matrimonial statusMarried41 (17.5)Single176 (75.2)Divorced17 (7.3)Widowed0 (0.0)Educational levelIlliterate5 (2.1)Primary school36 (15.4)Middle/high school138 (59.0)University55 (23.5)Living areaRural16 (6.8)Urban190 (81.2)Sub-urban28 (12.0)Mean ±SD (range)Age (in years)28.09 ± 9.24 (18–61)

Background characteristics of the participants.

Descriptive statistics

All participants completed the questionnaire. Thus, no missing data were reported.

Before conducting the confirmatory factor analysis, an item-level analysis was performed. Item-level descriptives and point-biserial item–total correlations for all 10 items of the HONC are summarized in Table 2. Endorsement rates (means) ranged from 0.50 to 0.75 (SDs 0.44–0.50).

ItemMeansSDCorrected point-biserial item-total rHONC10.710.460.616HONC20.710.460.640HONC30.710.450.645HONC40.710.460.744HONC50.660.470.713HONC60.500.500.450HONC70.710.460.647HONC80.670.470.687HONC90.710.460.730HONC100.750.440.770

Means, SDs and item-total correlations for the HONC items.

Mean = Endorsement rate; SD = standard deviation; Corrected point-biserial item-total r = Corrected point-biserial item-total correlation.

Point-biserial item-total correlations were moderate to high, ranging from 0.45 to 0.77 and supported the strength of each items’ association with a common construct. HONC 6 exhibited minimal correlation with the total score (r = 0.45), but still within acceptable bounds (values ≥ 0.30 are rated as acceptable) (53).

Confirmatory factor analysis

CFA was performed to confirm the unidimensional structure, derived from the conceptual framework underlying the instrument. Estimated using WLSMV estimator on the tetrachoric correlation matrix of the HONC dichotomous items, this analysis showed excellent fit of the model to the data: (χ2 (35) = 46.71, p = 0.089; CFI = 0.998; TLI = 0.998; RMSEA = 0.038 [90% CI 0.000–0.064]; SRMR = 0.059). The CFA model’s path diagram is provided in Figure 1.

Path diagram of the one-factor CFA model (standardized loadings).

As reflected in Table 3, standardized factor loadings were uniformly strong, ranging from 0.629 (HONC 6) to 0.952 (HONC 10) and all estimated loadings were statistically significant (p < 0.001), indicating that all items converge on the latent construct at adequate levels. A rule of thumb is that standardized loading estimates should be 0.5 or higher (41).

ItemB (Unstd.)SEzpStd. loadingR2HONC11.000–––0.8090.655HONC21.0090.06515.59<0.0010.8170.667HONC31.0290.06116.83<0.0010.8330.693HONC41.1460.06019.03<0.0010.9270.860HONC51.1060.06118.15<0.0010.8950.801HONC60.7770.07410.57<0.0010.6290.396HONC71.0450.06915.15<0.0010.8450.715HONC81.1180.06417.49<0.0010.9050.819HONC91.1360.06517.47<0.0010.9190.845HONC101.1760.06817.34<0.0010.9520.906

CFA results for the one-factor HONC model.

B = unstandardized loading; SE = standard error; z = Wald z statistic; p = two-tailed p-value; Std. Loading = standardized factor loading; R2 = squared multiple correlation. Estimator: WLSMV (robust DWLS). All estimated loadings were significant (p < 0.001).

Convergent validity and composite reliability of the one-factor model were further supported by an AVE of 0.736 and a CR of 0.965, exceeding the recommended thresholds of 0.50 and 0.70, respectively.

Criterion validity

Significant moderate positive correlation was found between HONC total score and FTCD total scores (Spearman’s ρ = 0.57, p < 0.001), showing the concordance between the two measures (higher levels of loss of autonomy over tobacco use as measured by the HONC, were associated with higher levels of nicotine dependence assessed by the FTCD), and supporting criterion validity.

Item response theoryAssumptions’ check

Prior to conducting IRT analyses, we tested its three assumptions; (a) unidimensionality, (b) local independence, and (c) Monotonicity.

Unidimensionality

The one-factor CFA model estimated with WLSMV fit the tetrachoric correlations well (χ2 (35) = 46.71, p = 0.089; CFI = 0.998; TLI = 0.998; RMSEA = 0.038 [90% CI 0.000–0.064]; SRMR = 0.059), and all standardized loadings were ≥ 0.60, indicating that the items are adequately represented by a single latent dimension; thus, the unidimensionality assumption is fulfilled.

Monotonicity

Table 4 results show strong Mokken scalability for the overall scale (H = 0.557, SE = 0.037), which indicated the scale is strong. According to Mokken, a scale is considered weak if 0.30 ≤ H < 0.40, a medium scale if 0.40 ≤ H < 0.50, and strong if H ≥ 0.50. Moreover, Hi > 0.30 (acceptability threshold) was achieved for all the items of the HONC (Hi = 0.496–0.663) (46). Therefore, the scale can be safely treated as unidimensional and can order the respondents on the latent continuum. Besides, as illustrated by the #vi column, no monotonicity violations were detected for any HONC item, meaning that the probability to endorse an item increases as the latent trait “loss of autonomy” increases. This meets the monotonicity requirement for unidimensional IRT.

ItemHiSE#vi#zsigHONC10.4960.04700HONC20.5150.04900HONC30.5220.05400HONC40.5890.03900HONC50.5950.04100HONC60.5220.05800HONC70.5200.04800HONC80.5710.04200HONC90.5790.04000HONC100.6630.05300Scale H0.5570.037

Mokken scalability and monotonicity diagnostics for HONC items.

Hi = item scalability (Loevinger’s Hᵢ); #vi = number of monotonicity violations; #zsig = number of statistically significant violations.

Local independence

We examined local independence using Yen’s Q3 statistic. It is the linear correlation between the residuals of each pair of items. There should be essentially no correlation between the residuals of two distinct items if participant ability is the only latent feature that determines the likelihood of correctly answering items. Item pairs with a value of Q3 > 0.20 should be flagged, according to Yen and Fitzpatrick. A high residual correlation might indicate that the response to one item influences the response to the other, or both items measure another unintended construct (54, 55).

The highest positive Q3 found was 0.196 for item pair HONC 8-HONC 9 (Table 5), which is below the standard 0.20 threshold. No item pairs exceeded the Yen’s Q3 cutoff, supporting the local independence assumption.

PairQ3HONC8-HONC90.196HONC7-HONC80.124HONC7-HONC100.119HONC1-HONC30.118HONC1-HONC20.115

Highest residual correlations (Yen’s Q3).

Item response theory analysisModel specification and rationale

For dichotomous items, IRT typically offers three common models: the one-parameter logistic model (1PL, Rasch), the two-parameter logist