The Spanish version of the Criminal Sentiment Scale Modified(CSS-M): Factor structure, reliability, and validity

The sample consisted of 153 male inmates from one of the sixteen Catalan prisons (Spain), the medium-security institution Brians I. This penitentiary center is one of the largest, housing approximately 1,400 inmates, both men and women, and is run by the Catalan Prison Service. These inmates were selected according to the residential unit they lived in and the type of offence they had committed. This ensured that no crime was overrepresented. The sample of offenders was taken from different regular modules and units. Each participant was under grade 2 regime and free from any punishments at the time. To ensure the sample was valid, it was important that all inmates had a good reading ability – special care was taken to guarantee this. Mean age was 37.3 years ( SD = 10), ranging from 21 to 81. The Spaniards represented 73.9% of the sample, the other 26.1% being foreigners. Since previous studies have pointed out whether the crimes committed were violent or not ( Andrews & Bonta, 2010 ), we focused on this variable, considering a crime as violent only if it had been committed in a physical manner; 38.6% of the inmates had committed a non-violent crime and 61.4% had perpetrated violent offences. The length of the sentences ranged from 1 to 23 years of incarceration, with a mean term of 7.2 years ( SD = 5.7).

In this study we defined a recidivist as an inmate who has been released from prison after serving his sentence and then re-enters prison because he or she has committed a new crime afterwards. Therefore, this variable ignores the number of crimes perpetrated (a person can commit a large number of crimes but enter prison for the first time), focusing only on the offence or offences committed between release and re-entry to prison. That said, 47.1% of inmates of the sample entered prison for the first time to serve a mandatory punishment of imprisonment, and 52.9% had relapsed on the aforementioned terms, returning to prison for the second time or more; 47.7% were serving only one sentence and 52.3% were serving two or more terms. This disregards any prior offences committed before the current entry date. We did not take into consideration the release day, nor the proportion of the overall sentence already served, in order to better randomize the sample.

Spanish version of the Criminal Sentiment Scale Modified (CSS-M) . The Spanish version of the CSS-M consists of the same number of items, 41, as the original English scale. Each of the subscales represent one of the three general categories of criminal attitudes: the subscale Attitudes toward the Law, Court and Police (LCP) comprises the first 25 items, e.g., “The law doesn’t help people (L)”, “You cannot get justice in court (C)”, or “Life would be better with fewer cops” (P). This subscale refers to the category Rejection of Convention, meaning that people who refuse social norms and law-enforcing public institutions are more prone to break the law. The second subscale is Tolerance for Law Violations (TLV), e.g., “A hungry man has the right to steal”, with 10 items related to the category of neutralization. This kind of attitude allows people to let themselves do something socially wrong, making it easier for them to break the law. The third subscale is the Identification with Criminal Others (ICO), with 6 items, e.g., “No one who breaks the law can be my friend”. This category is closely related to the third risk factor of the Big Four, criminal associates, but while the former refers to the inner thoughts to identify oneself as a criminal, the latter implies that one explicitly spends time or not with such criminals. The score of the CSS-M ranges from 0, meaning absence of procriminal attitudes, to 82. Thus, higher scores reflect higher levels of criminal attitudes by the respondent.

The responses of the original CSS-M ( agree, disagree or undecided) were changed to yes, no, or question mark (? ) to avoid misunderstandings among inmate respondents. The acceptance of a prosocial statement or the rejection of a criminal one yields 0 points, whereas an endorsement of a criminal statement (items of reverse score) or the rejection of a prosocial one yields 2 points. An undecided response always yields one point.

In order to translate and adapt the CSS-M into Spanish, the most reliable and known procedure, the backward-translation method ( Hambleton, 2005 ) was used. To achieve this, firstly a bilingual person translated the original scales into Spanish, focusing more on the idiomatic meaning than on word-for-word translation. Secondly, another professional who was unfamiliar with the scale made a backward-translation of the reviewed version into English again. Then, to verify that the meaning of the scales was preserved, both English scales versions were compared and reviewed to make cultural and vocabulary adaptations. Afterwards, a group of prison professionals reviewed a defined Spanish version of the CSS-M. This translation technique avoided any flaws or losses in translation compared to a direct one. The CSS-M Spanish version was administered by a team of both psychologists and educational workers who had been adequately trained to administer it. They explained each of the steps of the confidentiality agreement and guidelines scale. Because of the vulnerability of the inmates as a sensitive group, it was important to highlight that answering the CSS-M SV was entirely voluntary. The confidentiality agreement was based on three key aspects: (a) that the information provided by the inmate would not have any impact on their prison life in any case, so that they answer exactly what they think; (b) that the information they revealed through the questionnaire was only under their consent; and (c) that the inmates’ name would not be placed in any public or private document, but managed by a five-digit random code. Inmates were told about the detailed instructions at the very beginning of the questionnaire, remarking that it was not a test with good or wrong responses, that they should try to answer it honestly, and that the material was just for experimental and research purposes. Offenders answered the CSS-M Spanish version voluntarily without any remuneration. Only four inmates refused to do it.

Data analysis was carried out using the programs SPSS version 15.0 ( SPSS Inc., 2006) and Mplus (Muthén & Muthén, 2010 ). They started with an item analysis, which included indices of reliability. It was followed by the analysis of the factor structure conducting an explorative factor analysis (EFA) by an oblique rotation technique and a confirmatory factor analysis (CFA). Finally, several mean comparison analyses, by ANOVA procedure, were also carried out to establish significant group differences.

The 41 items of the CSS-M scale were analysed. In general, most of the items had a mean rating of around 1, indicating that they did not have a skewed distribution. Items 1 and 6 were exceptions, as both were not normally distributed, with means of 0.15 and 0.1 respectively. Tests of normality were conducted, regarding the values of skewness and kurtosis of the variables. Despite the problems with the two items, the reliability of the total scale had not been weakened. Table 1 shows the means, standard deviations, Cronbach's alpha, maximum, minimum, kurtosis, and symmetry for each subscale.

Cronbach's alpha coefficient was used to establish the internal consistency of the scale. The alpha values of the subscales ranged between .61 and .86. The reason the ICO subscale had the lowest coefficient was because it only had 6 items. Even so, it was slightly better than that of the original's (α = .51). The overall alpha coefficient of the Spanish version (α = .89) was very similar to that of the English scale (α = .91) by Simourd and Olver (2002) . Correlations between CSS-M subscales shown in Table 2 were all statistically significant.

An analysis of the structure of the scale was carried out through an Exploratory Factor Analysis (EFA) with an Oblique rotation technique, completed by the default Geomin rotation of the Mplus statistic package. Prior to this, we found that the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was .73 and Bartlett's test of sphericity achieved was statistically significant (χ 2 = 3053.34, p < .001), showing the appropriateness of performing the EFA. This took into account all the 41 items, eigenvalues greater than 1, and loadings above .30. As shown in Table 3 , the most suitable factor solution was a two-factor structure, accounting for 28.4% of the overall. Factor 1 was labelled Sentiments toward the establishment, comprising the majority of the items of the LCP subscale, while Factor 2, labelled Criminality self-benefits, contained most items of both the TLV and ICO scales.

To test the hypothesis that the CSS-M had a two-factor structure, a confirmatory factor analysis (CFA) was performed. The results showed that this model fitted reasonably well to data ( Hu & Bentler, 1999), χ2/df = 1.89, RMSEA = .07 [.06, .08], SRMR = .08. Additionally, cross-validation (splitting the sample into two groups, odd and even participants) exhibited similar results, supporting the stability of the model fit and the generalization to other samples of the same population ( Arce, Velasco, Novo, & Fariña, 2014 ).

A one-way analysis of variance (ANOVA) was completed separately to analyse possible differences of criminal attitudes within some groups: nationality, age, type of offence, recidivism, number of crimes serving, and sentence length (see Table 4 ). Inmates who committed violent crimes showed higher level of criminal attitudes than those who committed offences without violence ( p = .05), while people who committed a crime and entered prison for the first time had lower levels of criminal attitudes than those offenders who relapsed, that is, returned to prison for committing another crime ( p = .034).

However, whether the inmate was currently serving a sentence for committing one or more crimes did result in higher or lower criminal attitudes scores. Hence, the number of crimes they were serving sentence for was an irrelevant variable in determining the inmates’ criminal attitudes ( p = .306). Likewise, to serve a term of ten or more years was also irrelevant ( p = .087). Foreigners and nationals did not have significant differences in criminal attitudes either ( p = .909).