We translated and made a cross-cultural adaptation of the APRI questionnaire, followed by an analysis of its reliability and validity in a sample of students that completed the Spanish version of the APRI questionnaire.

The Adolescent Peer Relations Instrument (APRI-BT), which was developed by Parada,15 is a self-reporting tool comprising two subscales: one measuring bullying and the other victimisation. The latter scale, used in this study, assesses three dimensions of being targeted (physical, verbal and social bullying), each of which consists of six items scored on a Likert scale to determine how often the respondent experiences the situation (1=never, 2=sometimes, 3=once or twice a month, 4=once per week, 5=several times per week and 6=everyday). This results in scores whose addition produces overall score on an ordinal scale. High scores represent more frequent victimisation compared to low scores.14 There are no conflicts regarding authorship with the authors of the original English version of the APRI.

We used the forward and backward translation method (Fig. 1)13: two translators who were native Spanish speakers made an initial translation of the original version independently. In collaboration with the research team, the translators revised the two translations and developed a first version of the Spanish questionnaire by consensus. This version was translated back to English, after which the resulting version was compared with the original for the purpose of resolving any disagreements and producing the second version of the questionnaire.

Figure 1.

Cross-cultural adaptation process.

(0.15MB).

Previous to the administration of the instrument, we conducted a pilot study with 10 adolescents aged 12–16 years (six male, four female). We performed individual interviews using probing and paraphrasing methods.16 Once this phase was completed, we had the final version of the questionnaire.

The definitive assessment tool was administered to students enrolled in Spanish compulsory secondary education (ESO) in six secondary schools in a town of the region of Cartagena (South East Spain) (n=1706).

We collected the data in late May 2013. The questionnaires were self-administered by the students, with an approximate duration of 15min.

In addition to the adapted scale from the APRI-BT, we included the three questions on the “social acceptance” dimension of the KIDSCREEN-52.17,18 We also included sociodemographic variables such as age, sex, school year (of the ESO), social class of both parents according to the criteria of the Sociedad Española de Epidemiología (Spanish Society of Epidemiology)19 and parental educational attainment.20

Once we had completed the final version of the Spanish adaptation of the APRI (Appendix 1), we conducted preliminary psychometric analyses. First, we performed a descriptive analysis of the characteristics of the study sample and the questionnaire items. Then we verified the appropriateness of the factor analysis by means of Bartlett's test and the Kaiser–Meyer–Olkin (KMO) test.21 To identify the structure of the initial model, we performed exploratory factor analysis using the least squares method. We selected factors applying the Kaiser criterion (preserving factors with eigenvalues greater than one) and confirmed the results with a scree plot. Since we worked on the assumption that the factors were correlated, we used the Oblimin oblique rotation method to maximise the loading of items onto the different factors.

We compared the ability of the scale to discriminate between the three bullying level groups (a, 0.0; b, 0.1–25.0; c, 25.0–100.0), classified based on the scores obtained in the “social acceptance” dimension from the KIDSCREEN-52, by comparing their averages by means of the nonparametric Kruskal–Wallis test.22,23 Furthermore, we estimated the effect size of the difference between the groups under study (mean difference between groups relative to the standard deviation of the baseline group).

To assess the reliability of the questionnaire by analysing its internal consistency, we calculated Cronbach's alpha with 95% confidence intervals.

We also analysed the floor and ceiling effects (percentage of responses in the lower and upper extremes of the scale, respectively). We determined the acceptability of the instrument based on the missing data in each dimension and the overall response rate. We calculated the scores in the dimensions of victimisation by bullying by adding the points of all items (0=never to 4=constantly). Later on, we transformed these scores into a scale that went from 0 (minimum bullying) to 100 (maximum) to facilitate their interpretation. We used the IBM SPSS 21.0® software for the statistical analysis.

Based on the definition of bullying (abuse sustained through time), we assessed its intensity by assigning 0 points to options in which there had been no bullying (“never”) or it had been experienced infrequently (“occasionally”), while the remaining options were recoded with a value of 1 (“sometimes”), 2 (“often”) and 3 (constantly) points.

Two items required modification to achieve cultural equivalence: “tropiezan contra mí” (“bump into me”) rather than “chocan contra mí” (“crash into me”), “me ponen motes” (“they give me nicknames”) rather than “me llamaron por nombres que no me gustaban” (“I was called names I didn’t like”). During the interviews, two children agreed on having difficulty remembering the number of times that a situation had occurred each month or week. This is why we substituted the categories in the original Likert scale by a less ambiguous scale. The new scale has five categories of frequency that range from “never” to “constantly”. Considering the boom of emerging technologies and the growing importance of bullying through technological platforms in recent years,24 our research team considered that we also needed to include an item on cyberbullying.

Of the 1706 students in the initial sample, 1428 participated in the study, which corresponded to a response rate of 83.70%. Fifty-three percent (731) were male and the mean age was 14.6 years (SD=1.2 years) (Table 1).

Most participants belonged to the least privileged classes: 67.7% of fathers worked in semi-skilled and unskilled manual labour, and more than half of the mothers were housewives or currently unemployed.

Of the 1428 study participants, only 14 did not complete the questionnaire, with an overall acceptance rate of 99.0%. The lowest acceptance corresponded to the item about bullying through social networks (98.1%). In the other extreme, items on physical victimisation by being pushed and the item that dealt with the damage to personal property were the items with the highest specific-response rate (98.9%). The remaining items in the questionnaire were answered by a mean of 1407 participants.

The percentage of missing values for one or more items in each dimension and in the overall scale amounted to less than 5% (2.2% in the physical victimisation dimension and 4.3% in the verbal/social dimension) (Table 2). If we were to exclude the cyberbullying item, the percentage of missing values in the verbal/social subscale would drop by nearly one point (3.6%).

The ceiling effect was negligible in the two dimensions of the scale (0.1%), although the floor effect was considerable (60.6% [physical victimisation] and 36.5% [verbal/social victimisation]) (Table 2).

The Kaiser–Meyer–Olkin measure of sampling adequacy (0.963) and Bartlett's sphericity test (P<.0001) showed that the items were interrelated and that the sample met the assumptions for factor analysis. Both the scree plot and the exploratory factor analysis suggested a two-factor structure.

After performing oblique rotation, we observed that all items exhibited high factor loadings ranging between 0.92 and 0.56; the assumed correlation between factors was evinced by an r coefficient of 0.77 between items (P<.001). The first-order factor, which we labelled “social and verbal victimisation”, groups the items pertaining to social exclusion and verbal abuse. This factor is nearly identical to the “verbal” and “social” dimensions of the original scale. However, item 2 in the physical victimisation subscale (being pushed) loaded more strongly onto the social and verbal victimisation factor (0.44) than the physical victimisation factor (0.22), which comprises all the “physical bullying” items in the original dimension. Something similar occurred with item 17 on the exclusion from leisure activities and item 19 on cyberbullying, with loadings of 0.58 and 0.42 onto the second-order “physical victimisation” factor in spite of their logical place in the “verbal and social victimisation” dimension (first-order factor) (Table 3). This factor structure is maintained with only slight differences in item saturation after excluding item 19 on cyberbullying.

The questionnaire can discriminate between subjects experiencing different degrees of bullying (Table 4). The mean scores were lowest in the no-bullying group both in the two dimensions and the overall scale, increased significantly in the moderate bullying group, and peaked in the severe bullying group (P<.001). The effect sizes were also considerable, and ranged between a minimum of 0.7 and a maximum of 3.2.

The internal consistence of the overall scale was very high, with a Cronbach's α coefficient of 0.94 (CI, 0.936–0.945). The dimensions in the scale were also highly consistent, with a Cronbach's α of 0.85 (CI, 0.83–0.862) for the “physical victimisation” factor—including item 2, being pushed—and an even higher value for the “social and verbal victimisation” factor—including items 17 and 19 pertaining to the exclusion from leisure activities and to cyberbullying—of 0.92 (CI, 0.914–0.926) (Table 5). This consistency was maintained and did not improve with the removal of items from the overall scale or subscales (Table 5).

The percentile scores for bullying intensity we obtained in the possible range of 0 to 100 were low. Males scored higher than females in all dimensions for any given percentile. The difference between male and female respondents in the assessment of bullying averaged 1.8 points at the 85th percentile, and 28.2 points at the 97th percentile (Table 6). Only 6.3% of female participants had experienced physical bullying, a percentage that doubled in males (12.1%).