Cultural adaptation to Spanish and psychometric analysis of the Woman Abuse Screening Tool (WAST) in Peruvian adult women

Cultural adaptation phase

Translation

Between December 2022 and January 2023, the English version of the Woman Abuse Screening Tool (WAST) scale was translated into Peruvian Spanish by two independent translators. After the translation was completed, a consensus was reached among the translators and researchers so that the translations were reviewed to determine a single version in Peruvian Spanish. To verify the quality of the translation, a reverse translation (from Spanish to English) was used, where two additional translators were hired; a consensus was reached to verify the similarities with the original version (in English) and the final version of the WAST translation was approved.

Cultural adaptation to the Peruvian population

For this stage, a validation and cultural adaptation form was initially elaborated, considering some indicators such as relevance, representativeness, clarity and cultural equivalence.³³ In addition, the expert judges were asked to make qualitative suggestions to improve the items. Four expert judges (two psychologists and two physicians) with at least three years of experience in dealing with cases of intimate partner violence participated in the study. The research team had an interaction (verification rounds) from two to four times with the expert judges individually. In these interactions, the goal was to present a version with the improvements made until a favourable review was obtained from the judges.

In order to verify the clarity and ease of understanding of the WAST items in Peruvian Spanish, a focus group was organised (on a virtual platform). This activity was carried out by a psychologist dedicated to scientific research with a master’s degree and experience in qualitative interview methods. The researchers who were in Cajamarca, Ancash, and Lima invited women to participate in public parks in urban areas and in the community hall in rural areas. Initially, the researchers presented the study and the objectives, and invited the participants to participate in the study. Women over 18 years of age who had been living with their spouse in the last year were included, and those who did not have Peruvian nationality were excluded. The focus group was recorded with the consent of all participants. 12 women over 18 years of age participated in the focus group, 3 of them indicated that they had completed primary education and 9 that they had completed secondary education, and 5 lived in rural areas and 7 in urban areas. Initially, the WAST scale was presented in a virtual format in Google Forms and the moderator requested the participants to complete the scale. Subsequently, the moderator used a guide of open questions related to the clarity and understanding of the WAST items referring to the content analysis, which was developed for the focus group. Further details of the cultural adaptation process can be found in the flow chart (see Ref. 33).

Psychometric analysis phase

Participants

Participants were selected by non-probability snowball sampling (n = 628 women). The participants were women over 18 years of age, living in urban and rural areas of different departments in Peru, who reported having a partner with whom they were living (at the time of the study) in a cohabiting or married condition, at least able to read Peruvian Spanish, and having Peruvian nationality.

Instruments

The Woman Abuse Screening Tool (WAST) scale assesses indicators of intimate partner violence. This instrument has 8 items and a single dimension, which are designed to determine the degree of tension and difficulty in the partner relationship, as well as the presence of violent episodes, which can be answered on a Likert scale, from 1 to 3, with 1 being the option of least intensity or frequency. In its original version, it reported adequate discriminant validity when correlating the WAST with the Abuse Risk Inventory -ARI (Rho = 0.90; p= 0.00), and also showed an optimal alpha coefficient of 0.95.¹²

The Patient Health Questionnaire (PHQ-9) consists of 9 items that correspond to DSM-IV depressive symptoms.¹³ Its response options are directed to the last two weeks, considering a Likert-type scale: 0 = not at all, 1 = several days, 2 = most days, 3 = almost every day. From its nine items a score between 0 and 27 can be obtained. The Spanish version of the PHQ-9 has shown validity (e.g., goodness-of-fit as an one-dimensional measure: CFI = 0.936; RMSEA = 0.089; SRMR = 0.039), as well as adequate reliability (α = ω = 0.87) in a Spanish-speaking Peruvian population.¹⁴

The Generalized Anxiety Disorder-7 (GAD-7)¹⁵ scale assesses generalised anxiety disorder during the last two weeks according to the DSM-V. It presents 7 items ranging from 0 (not at all), 1 (some days), 2 (more than half of the days) to 3 (almost every day) and the score can vary from 0 to 21. It also shows an adequate internal and external structure, taking into account an one-dimensional model (χ²¹⁴ = 31.717, CFI = .995, TLI = .992, RMSEA = .056, SRMR = .026) as well as adequate reliability (α = .92).

Covariates

Some variables were considered to characterise the population, as well as to study the measurement invariance of the one-dimensional model according to age (in years), sex (female, male), educational level (primary or secondary, higher), work activity (housewife, employee or worker, own business) and place of residence (urban area, rural area).

Procedures

For data collection, approval was first sought from the Ethics Committee. Then, three trained interviewers presented the data collection instrument by means of a URL or QR code printout leading to the survey in Google Forms (with the following structure: informed consent, WAST, PHQ-9, GAD-7). The instrument initially presented informed consent and only those who agreed to participate proceeded with the survey. Data collection was conducted in the departments of Cajamarca, Ancash and Lima. Each interviewer approached potential participants in public parks, considering only those who indicated that they lived with their spouse in the same household during the data collection period (January to April 2023). Once identified, they proceeded to present the study and once the data collection was completed, they asked the participant to provide an account of another adult woman living with her partner (married or cohabiting). This procedure was repeated until the end of the data collection period.

Statistical analysis

At the descriptive level, we used descriptive statistics of the items (mean, standard deviation, skewness and kurtosis). Subsequently, a confirmatory factor analysis (CFA) was performed assuming that the WAST is a one-dimensional model, due to its uniform (one-dimensional) behaviour in all previous studies. Also, we used the weighted least square mean and variance adjusted (WLSMV) estimator because this estimator is unbiased for instruments with ordinal categorical items and non-normal distribution.¹⁶ We report the standardised betas of the model and standard measures of goodness-of-fit: the χ² for the model versus baseline, considering values < 3 acceptable; the comparative fit index (CFI), which is adequate when >0.90; the Tucker-Lewis Index (TLI), which is acceptable when >0.90. In addition, the standardised root mean square residual (SRMR) and the root mean squared error of approximation (RMSEA) were considered adequate with values ≤ 0.08.¹⁷

The next step was to assess measurement invariance across groups defined by housing location (rural, urban), using a multi-group CFA. The CFI (ΔCFI) and RMSEA (ΔRMSEA) values are used as the main criterion to compare models with more restrictions against models with fewer restrictions. Initially, the models assumed configural invariance (i.e., similar factor structure across groups) as the base model, progressing to metric invariance (i.e., similar factor loadings and factor structure across groups), strong invariance (i.e., similar thresholds, factor loadings, and factor structure across groups), and strict invariance (i.e., similar residual variances of items, thresholds, factor loadings, and factor structure across groups). Between each model, we examined whether ΔCFI < 0.01 or ΔRMSEA < 0.01 to establish whether the more restricted model was appropriate.¹⁸

Optionally, MIMIC (multiple indicator, multiple cause) models were fitted for the assessment of measurement invariance by age, education level, years of living together and work activity. The invariance of the intercepts of the indicators and the mean differences of the latent dimensions were evaluated, all through groups according to these variables. Each covariate is evaluated separately, comparing for each of them two types of models: 1) a saturated version where the covariate explains all the observed items, but not the latent dimensions, and 2) a version of the invariant intercept model where the covariate explains all the latent dimensions, but not the items. Similarly, the fit indices reported above are reported and interpreted.¹⁸

Reliability was also estimated using Cronbach’s Alpha¹⁹ and McDonald’s Omega.²⁰ All analyses were performed in R Studio version 4.0.4, with the packages “lavaan”,²¹ “lavaan.survey”,²² “semTools”,²³ “semPlot”.²⁴

Ethical considerations

This study was evaluated and approved by the Ethics Committee of Universidad Cesar Vallejo with report 022-CE-FCS-UCV-21, with approval date 9 December, 2021. All the ethical principles of research involving human subjects from Declaration of Helsinki,²⁵ such as autonomy, confidentiality and fairness, were respected both in the sample for the psychometric analysis and the focus group participants. Participants were given an information sheet regarding the study (via Google Forms) and confirmed their consent via clicking ‘yes’ on the form. Authorisation to use and adapt the original instrument was obtained from the copyright owner through email.

Confirmatory factor analysis

It has been identified that the one-dimensional model presents adequate goodness-of-fit indices (CFI = 0.998; TLI = 0.991; RMSEA = 0.103; SRMR = 0.075), considering the CFI and TLI with values > 0.95, as well as the RMSEA and SRMR with values < 0.10 estimating those as adequate fits and in the case of the RMSEA as moderately adequate (Table 2). Likewise, the factor loadings were between λ = 0.79 and λ = 0.96 with values > 0.30 (Figure 1).

Figure 1. SEM graphic of the WAST in Peruvian women.

Measurement invariance

The configural invariance reported that the model structure is similar based on place of residence (rural/urban) (p > 0.05). Additionally, the CFI and TLI values were greater than 0.95 and the RMSEA and SRMR less than 0.10, so we proceeded to analyse the metric, strong and strict invariance, the goodness-of-fit indices were within the expected parameters for an adequate fit and their delta values (Δ CFI and Δ RMSEA < 0.015) reported values below the parameter for an adequate strict invariance (Table 3).

Validity based on the relationship with other variables

Using the Spearman correlation coefficient, it was found that the WAST scores were related to the PHQ-9 (Rho = 0.620; p = 0.00) and GAD-7 (Rho = 0.634; p = 0.00) with a direct relationship and a strong relationship strength Rho > 0.5. This indicates that higher WAST score, the higher the scores on the PHQ-9 and GAD-7 measures, which confirms external validity (Table 5).

Reliability

The reliability of the Woman Abuse Screening Tool (WAST) was high, reporting internal consistency coefficients of α = 0.899 and ω = 0.916 as optimal, based on which the instrument is considered accurate in measuring intimate partner violence (Table 2).

Validity of the internal structure

A confirmatory analysis was used for internal validity, where a one-dimensional model reported a good fit, considering the CFI and TLI with values > 0.95 and the RMSEA and SRMR with values < 0.08; in the case of the RMSEA it turned out to be moderately adequate. These results are similar to a study conducted in Peruvian university students where it was also found to be unifactorial and with similar indices (χ² = 48.037; GFI = 0.997; CFI = 0.994; TLI = 0.992; SRMR = 0.049; RMSEA = 0.058).¹¹ Overall, the goodness-of-fit indices support the validity of the internal structure of the WAST in Peruvian adult women. However, the error rate, such as the root mean square error of approximation (RMSEA), was above the acceptance threshold, which could be due to the sample size, since mostly women with lower indicators of intimate partner violence were included.²⁶ This could be overcome in a future study by including a proportionate number of adult women who have experienced violence compared to those who were not victims.

Measurement invariance

Our findings support that the WAST shows measurement invariance across groups of women with differences in age, place of residence, education, years of cohabitation and work activity. This is mainly due to the fact that the values of the difference indices such as the comparative fit index and the root mean squared error of approximation, resulted in expected values (< 0.01) for an adequate adjustment for the characterisation variables analysed at all levels of invariance up to strict invariance.²⁶ This study would be the first to report the invariance of the instrument, as previous studies focused more on other aspects of its validity.²⁷ This finding suggests a quality of the WAST in assessing indicators of intimate partner violence in subgroups of women which supports the usefulness and generalizability of the WAST²⁸ as a screening instrument for clinical use and further research in adult Peruvian women.

Reliability

Regarding the reliability of the WAST in Peruvian women, the results showed adequate scores on Cronbach’s alpha and McDonald’s Omega coefficients. Such strong internal consistency guarantees an accurate measurement of intimate partner violence taken from the same factor loadings by means of McDonald’s Omega.²⁰ Likewise, previous studies reported similar results. For example, a study conducted in a Canadian population reported an alpha coefficient of α = 0.75,⁸ while in in the United States, the alpha coefficients were α = 0.78 in English and α = 0.80 in Spanish.²⁹ In a Chilean population, the reliability coefficient was α = 0.92,¹⁰ in Indonesia α = 0.801,⁹ while in the French population it was solidly reliable α = 0.95,⁷ and finally a study with Peruvian university women reported an optimal reliability α = 0.86 and ω = 0.90.¹¹ Therefore, despite having a different culture in each country, it was observed that the WAST measures have similar internal consistency, which provides strong evidence of cross-cultural internal consistency.

Validity based on the relationship with other variables

Using Spearman’s correlation coefficient, it was found that the WAST scores were correlated to the PHQ-9 and GAD-7 scores with an adequate coefficient (Rho > 0.50), so it can be assumed to have external validity. Our finding is similar to a study by Brown et al. (2000) who analysed results of the WAST and the ARI in London, Ontario, Canada and its surroundings (risk inventory) where adequate correlation coefficients were also reported (r = 0.69, p = 0.1). Similarly, a study in Hong Kong, China showed a significant correlation with anxiety (r = 0.34, p < 0.00) and depression (r = 0.30, p < 0.000).³⁰ It is therefore possible that further studies could use it to study intimate partner violence even in populations suffering from anxiety and depression in addition to other possible variables.

Public health relevance

The Woman Abuse Screening Tool (WAST) is one of the most widely used instruments in primary health care in different parts of the world for the early detection of cases of intimate partner violence due to its easy scoring based on its eight items (where the higher the score, the greater the presence of intimate partner violence).^31,32 Therefore, in Peru, the Ministry of Health could implement it as a screening tool in primary health care and community mental health centres.

Limitations and strengths

It is important to state some limitations of the study, one of them is the type of sampling, as the findings are not generalisable, but the results are valid for the population studied. In addition, the collection was carried out in places with higher traffic of people, so it is possible that in areas with little traffic the results would be different; however, this does not detract from the importance and scientific value of the data analysed.