Introduction
The use of self-reports in psychology and related sciences and their adapting, translating and validation processes is quite common. Self-reports (tests, questionnaires, scales …) are relevant when proving theories, making decisions about the effectiveness of a psychological treatment, experimentally verifying the impact of independent variables, etc. (Carretero-Dios & Pérez, 2007). Thus, the scores obtained from the scales have implications of great importance for the final result of any research that uses self-reports, as well as for the applied consequences deriving from the professional activity and decisions made based on these tests results (Padilla, Gomez, Hidalgo, & Muñiz, 2007). While there are a number of limitations inherent in these evaluation procedures in psychology (limitations in memory, desirability, understanding …) there are also various standards to minimize said limitations. Following the Standards for Educational and Psychological Testing (AERA, APA, & NCME, 2014), or the remarks of Kolen and Brennan (2014) and Downing and Haladyna's (2006) when preparing a self-report should minimize errors in the estimation of the construct to be evaluated. This should also improve the reliability of the evaluation. If the test does already exist in another language, these standards should be kept in mind, but an additional process should be implemented in order to adapt and translate such self-report so that the translation- related bias is reduced as far as possible.
Protocols and advice on test translation and adaptation have also been crucial, although they have received less attention than general standards. Currently, these adaptations are usually from English into another language i.e. Chinese, Spanish, German inter alia (Guillemin, Bombardier, & Beaton, 1993). In this case, Elosua, Mujika, Almeida and Hermosilla (2014), Hambleton, Merenda, and Spielberger (2005) or Muñiz and Bartram (2007) are some references to take into consideration when translating a test from one language into another. The aforesaid authors include explanations on translation techniques (back translation is usually recommended), cultural equivalence, or specification tables on content equivalence.
Recently, Muñiz, Elosua & Hambleton (2013) improved the traditional Guidelines for Adapting Tests (International Test Commission [ITC]; Hambleton, 1994) in a review of the standards (Hambleton, 1996) where some of their limitations are updated. They further propose - among other changes - replacing the back translation by an expert-agreed double forward translation. This effort seeks to solve the problems posed by increasing transnational collaborations (Bullinger et al., 1998).
Spanish is one of the languages questionnaires are most often adapted to. This language is widely used (414 million native speakers; i.e. more than English. See Lewis, Gary, & Charles, 2014). These speakers are distributed in 33 countries (Lewis et al., 2014) and have large cultural differences. Although it is rare to find adaptations from American English to Australian or British English in English-speaking contexts like in Spanish, these do exist (McDowell, Courtney, Edwards & Shortridge-Baggett, 2005; Sanson-Fisher & Perkins, 1998). An example of these are the PISA test, where translation and adaptation processes are conducted following standardized guidelines (Programme for International Student Assessment, 2010). Nevertheless, it is not clear how they do that process in populations who have the same languages but different expressions and contexts, which undoubtedly may have a serious penalization in the data obtained (Grisay, 2003,2007; Grisey & Monseur, 2007). Because there are no protocols (to the extent of our knowledge) to adapt self-reports from one source language into the same language in another culture, researchers are thus faced with a true challenge (McDowell et al., 2005).
Furthermore, evaluation in sexuality is usually modulated by moral, religious or ethical criteria from the context in which it is evaluated. In addition, intimate aspects of the person are inquired, thus making desirability higher than in other scales. For this reason, it was decided to test this protocol by using a sexuality scale (Sexual Assertiveness Scale; SAS Morokoff et al., 1997), considering that there are differences between the Colombian and Spanish cultures. This is a key variable in psychosexual wellbeing (Santos-Iglesias & Sierra, 2010).
Consequently, the objective of this instrumental research is to adapt the protocol proposed by Muñiz et al. (2013), based on ITC guidelines (Hambleton 1994) in order to obtain new ones for adapting questionnaires from one source language to the same language in two different contexts. To this end, the procedure has been tested by using actual data answered on the SAS.
Method
Participants
Sample included a total of 825 participants who gave their acceptance to participate in this study by an informed consent document. From this sample, 454 participants were Colombian and 371 participants were Spanish. All participants had had a stable partner for a minimum period of 6 months. Information was collected in over 100 different cities, in both countries. Other information about the sample is organized descriptively - and inferentially, if applicable - in Table 1.
Variables | Colombia | Spain | Contrast | |
---|---|---|---|---|
Sex | ||||
Female | 282 | 219 | χ2(1)= .48;p= .48 | |
Male | 170 | 146 | ||
Age | 33.24(11.03) | 35.60(12.94) | t(823) = 2.83;p < .01;d = 0.19 | |
Years of schooling | 16.75(2.94) | 15.64 (4.26) | t(817) = 4.42; p < .01;d = 0.30 | |
Sexual orientation | ||||
Excl. heterosexual | 392 | 315 | ||
2 | 28 | 24 | ||
3 | 5 | 2 | ||
4 | 4 | 5 | χ2(7) = 4.71;p = .69 | |
5 | 0 | 2 | ||
6 | 6 | 8 | ||
Excl. homosexual | 15 | 13 | ||
Asexual | 2 | 1 | ||
Marital status | ||||
Single | 198 | 182 | ||
Married | 143 | 142 | ||
Common-law marriage | 86 | 29 | χ2 (3) = 22.39; p < .01; η2 = 0.10 | |
Divorced | 23 | 16 | ||
Religiousness | ||||
Daily. | 3 | 1 | ||
Once a week. | 77 | 11 | ||
Rarely in a month. | 104 | 16 | χ2 (4) = 130.45; p < .01; η2 = 0.38 | |
Rarely in a year. | 146 | 140 | ||
Never. | 117 | 200 |
Instruments
Sexual Assertiveness Scale (Morokoff et al., 1997). This study was based on the Spanish version of the scale, which has shown adequate psychometric properties (Sierra, Santos-Iglesias and Vallejo-Medina, 2012; Sierra, Vallejo- Medina and Santos-Iglesias, 2011; Vallejo-Medina and Sierra, 2015). This is composed by 18 items which evaluate three different dimensions: Initiation: ability to initiate sexual relations when it is wanted, and to perform them as desired; refusal: ability to reject sexual unwanted practices or contact; and Sexually Transmitted Diseases - unwanted Pregnancy (STD-P). From these items, the nine items that were written backwards were eliminated for this study upon recommendation of the experts due to lack understanding of themselves. The instrument is available in both versions in Appendix A. High scores mean high sexual assertiveness.
Sexual Opinion Survey (SOS; Fisher, White, Byrne, & Kelley, 1988). The brief versions validated in Spain (Vallejo-Medina, Granados, & Sierra, 2014) and Colombia (Vallejo-Medina et al., 2016) were used. This six-item scale is answered on a six - alternative Likert scale, and evaluates attitudes toward sexuality, with a score ranging on a continuum between erotophobia and erotophilia. Reliability of this survey in this study was .85. High scores indicate erotophilia.
Massachusetts General Hospital-Sexual Functioning Questionnaire (MGH-SFQ; Labbate & Lare, 2001). Spain (Sierra, Vallejo-Medina, Santos-Iglesias y Fernández, 2012) and Colombia versions (Marchal-Bertrand et al., 2016) were used. This evaluates sexual interest, sexual arousal, orgasm, erection (for men only) and overall satisfaction in both genders. Reliability in this study was .90. Higher scores indicate a better sexual performance.
Sexuality Scale (Wiedemann & Allgeier, 1993; Soler et al., 2016). The abbreviated scale has been used in this study. This consists of 15 items answered on a Likert scale. They are grouped in three dimensions: Sexual self-esteem, sexual depression and sexual preoccupation. Reliability of the sub-scales is appropriate: the Cronbach α ranges from .85 to .93 and test-retest reliability has significant minimum correlations of .67. Meanwhile, the external validity of the scale has proven adequate in English-speaking contexts.
Procedure
Preliminary SAS adaptation of the items from Spain to Colombia was made by four psychologists who were born in Colombia, or who were Colombian residents; all of them had at least a master's degree, and had been studying and living in Spain for at least two years. Using a table of specifications of the items, the psychologists had to indicate whether the items were fully understood, sounded strange, or whether they could be stated otherwise. This task was performed individually; it was compiled by the researchers of this study using a color table where the items were identified. Any items which had not been modified were identified in green; items which had been modified by only one of the psychologists were yellow, and those which had more than one modification, were red. Once the status of the items was identified, two sexuality experts met with the adapters in order to review the status of the items following Muñiz et al.'s (2013) guidelines. When the item was green, it was reported that the four experts had agreed not to modify it; however, they would read it again to ensure that the item had been understood correctly. Items that had a discrepancy (marked in yellow) were read again and it was stated that a different wording was considered more appropriate by one of the adaptation experts. At that point, a small debate was conducted under the guidance of the sexuality experts, until all the participants agreed on a final wording (adapter and sexuality experts). Something similar happened when the items were marked in red. Suggested modifications were presented and debated in order to obtain the best solution.
During this adaptation process, an error occurred when adapting item number 4 where some content of the item was missed. Thus original item is: “I refuse to let my partner touch my breasts if I don't want that, even if my partner insists” and in the Colombian adaptation the - “if I donʼt want that”- fragment was skipped. The problem was mitigated by modifying the wording in the middle of the evaluation. This error has undoubtedly affected the results of this research and it will be corrected in the future. So, to control a possible change of contents of an item suggested by the adapters, adaptation should be confirmed with a backward “translation”; not as an adaptive method itself, but to ensure the avoidance of any alteration in the content of the item. Thus, it is recommended that once experts and adapters have agreed on the new items, at least one expert who did not partake in the process of adaptation assesses the equivalence of the content between the first Spanish version and the adapted version.
Once there was an agreement on the adapted version, a total of four different psychometrics and / or sexuality experts evaluated the items’ properties. The following characteristics were evaluated: Representativeness and Ownership of the item to the sexual assertiveness construct; Understanding the item in the Colombian version; a single Interpretation (no ambiguity); and item Clarity (how concise it is). To this end, a table of specifications of the items and the ICaiken program (Soto & Segovia, 2009) were used. The ICaiken program allows to obtain the confidence interval for the Aiken V (1985). Experts scored the property of each item in a range of 1 (Nothing) to 4 (Very).
The sampling was conducted similarly in both countries; it was incidental and the evaluation was performed on-line (from October 23, 2014 to February 24, 2015). Questionnaires were designed in Typeform© and distributed through personal and Facebook contacts. The average time to answer the survey was 13 minutes and 18 seconds.
Data Analysis
A cut point below .50 in the lower limit (CI = .95%) of the Aiken V has been considered as a criterion of item inadequacy (Merino Soto & Livia Segovia, 2009).
EQS 6.1. multivariate software was used to calculate Factorial Invariance (FI). This was evaluated with a forward, multistage procedure, under a Mean and Covariances Structures procedure (MACS) as recommended (Byrne, 2009). This procedure allows for strong evaluation of invariance, as compared to the Covariance Structures Analysis (COVS), which only allows for weak evaluation of the FI (Meredith, 1993). Moment analysis and Maximum Likelihood - Robust method (Ml, Robust) were used. The latter is a robust estimator when non-multivariate normality is observed. Forward FI will be performed in four steps:
Configural invariance (invariance will be evaluated without restrictions in the model);
Metric or weak invariance (the factorial weights will be restricted);
Strong invariance (the intercepts will be restricted); and
Strict invariance (the variances of errors will be restricted).
The overall fit indices used were the Root Mean Square Error Approximation (RMSEA) and its 90% confidence interval, as well as the Comparative Fit Index (CFI). Values lower than .08 for the RMSEA and higher than .95 for the CFI will be considered of good fit. In addition, the CFI will be the main indicator used to evaluate the FI; the fact that the CFI does not decrease by more than .01 compared to the previous model Cheung & Rensvold, 2002) shall be regarded as evidence of invariance. Finally, the Akaike Information Criterion (AIC) will also be reported. a considerable increase in this indicator will be indicative of absence of FI.
The SPSS 20.0 was used in order to evaluate the presence of DIF by way of multinomial logistic regression. This technique allows to detect uniform and non-uniform DIF in polytomous items (Hidalgo & López-Pina, 2004). If the contribution of DIF is uniform by Mean Square Error (99%), we will have a uniform DIF; if the contribution of Model 3 is significant (99%), DIF is not uniform. Given the large sample size, it is expected to find meaningful relationships; thus, so the DIF report will be complemented by a measure of effect size, which will measure the magnitude of DIF - the Nagelkerke's ΔR2 (R2 < 0.035 = negligible DIF; 0.035 < R2 < 0.070 = DIF moderate; R2 > 0.070 = DIF high; Jodoin & Gierl, 2001). A purification process will also be implemented in stages for DIF items that have shown moderate or high levels. To this end, we will conduct a new regression eliminating all items with DIF from the scale. Thus, we will be able to determine whether the presence of DIF was attenuating, aggravating or concealing up the presence of more DIF (Differential Item Functioning).
The other results were obtained with the use of SPSS 20.0.
Results
Table 2 shows that the items adapted to the Colombian version have adequate qualitative properties, including an excellent content validity. This will indicate that a proper adaptation process of the items was performed. It is clear that this process failed to show the error in item 4.
Exp1 | Exp2 | Exp3 | Exp4 | M | Aiken | % agreement | LL 95% | UL 95% | ||
---|---|---|---|---|---|---|---|---|---|---|
Item 1 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 1 | 1 | 1 | 1 | 100 | |||||
Understanding | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Item 2 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 1 | 1 | 1 | 1 | 100 | |||||
Understanding | 3 | 4 | 3 | 4 | 3.5 | .83 | .55 | .95 | ||
Interpretation | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Clarity | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Item 3 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 1 | 1 | 1 | 1 | 100 | |||||
Understanding | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Item 4 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 2 | 2 | 2 | 2 | 100 | |||||
Understanding | 3 | 4 | 3 | 4 | 3.5 | .83 | .55 | .95 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 3 | 4 | 4 | 4 | 3.75 | .91 | ‥64 | .98 | ||
Item 5 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 2 | 2 | 2 | 2 | 100 | |||||
Understanding | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Item 6 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 2 | 2 | 2 | 2 | 100 | |||||
Understanding | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Item 7 | Representativeness | 3 | 4 | 3 | 4 | 3.5 | .83 | .55 | .95 | |
Ownership | 3 | 3 | 3 | 3 | 100 | |||||
Understanding | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Item 8 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 3 | 3 | 3 | 3 | 100 | |||||
Understanding | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Item 9 | Representativeness | 3 | 4 | 4 | 4 | 3.75 | .91 | .64 | .98 | |
Ownership | 3 | 3 | 3 | 3 | 100 | |||||
Understanding | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Interpretation | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 | ||
Clarity | 4 | 4 | 4 | 4 | 4 | 1 | .75 | 1 |
Note: Exp: Expert; LL: Lower limit; UL: Upper limit.
Table 3 shows how the other psychometric properties of the items are suitable, except for item 4 of the Colombian version. Total item correlations of all items are above .30, and the Cronbach's alpha is always lower than the alpha of the corresponding sub-scale if the item is removed, except for items 1 of both versions. SD's are observed which indicate proper distribution of scores.
Country | Dimension | Item | M | SD | Asymetry | Skewness | ri-tc | α-ítem | α | M | SD |
---|---|---|---|---|---|---|---|---|---|---|---|
Colombia | Initiation | SAS1 | 2.57 | 1.07 | -.33 | -.73 | .36 | .82 | .72 | 6.88 | 3.24 |
SAS2 | 2.34 | 1.44 | -.36 | -1.23 | .69 | .44 | |||||
SAS3 | 1.98 | 1.50 | .00 | -1.48 | .63 | .52 | |||||
Refusal | SAS4 | 0.59 | 0.98 | 1.85 | 2.51 | .26 | .71 | .61 | 3.87 | 3.13 | |
SAS5 | 1.56 | 1.58 | .46 | -1.41 | .53 | .34 | |||||
SAS6 | 1.76 | 1.53 | .29 | -1.46 | .53 | .34 | |||||
STD-P | SAS7 | 1.25 | 1.50 | .86 | -.80 | .82 | .84 | .89 | 3.75 | 4.23 | |
SAS8 | 1.49 | 1.67 | .58 | -1.39 | .79 | .86 | |||||
SAS9 | 1.05 | 1.50 | 1.14 | -.33 | .79 | .86 | |||||
Spain | Initiation | SAS1 | 2.45 | 1.16 | -.26 | -1.16 | .41 | .83 | .75 | 6.78 | 3.28 |
SAS2 | 2.32 | 1.34 | -.20 | -1.34 | .70 | .53 | |||||
SAS3 | 2.01 | 1.48 | .06 | -1.51 | .67 | .57 | |||||
Refusal | SAS4 | 1.37 | 1.48 | .65 | -1.09 | .63 | .81 | .83 | 5.22 | 4.02 | |
SAS5 | 1.96 | 1.65 | .09 | -1.67 | .71 | .74 | |||||
SAS6 | 1.92 | 1.53 | .08 | -1.56 | .72 | .73 | |||||
STD-P | SAS7 | 1.48 | 1.65 | .59 | -1.37 | .81 | .87 | .90 | 4.49 | 4.72 | |
SAS8 | 1.76 | 1.79 | .28 | -1.75 | .82 | .86 | |||||
SAS9 | 1.35 | 1.72 | .74 | -1.28 | .81 | .87 |
Note. M: Mean; SD: Standard Deviation; ri-tc: item total corrected correlation; α-ítem: Cronbach alpha if item is deleted; α: Cronbach alpha.
Subsequently, Factorial Invariance was tested. Mardia's test had a value of 6.27 and 7.40. Table 4 shows a suitable fit for the configural model. Adequate fit indices can be seen in the model of weak invariance; however, the AIC decreases slightly and the ΔCFI is six thousandths above .01. Therefore, the level of weak invariance could not be regarded as accepted. At this point, and in accordance with the modification indexes (Lagrange multiple test), partial invariance without item 4 constriction, was conducted. With this change, both good fit indexes and invariance indicators were observed. Continuing the progression, we find strong invariance which obtains adequate fit indices and it has appropriate indicators of factorial invariance. Finally, strict invariance, which is the last level of invariance tested here, was also fulfilled, thus closing the process of invariance. In addition, standardized weights of scales and related R2 can be observed in Table 5. DIF analysis showed no bias in the functioning of items - except item 4 - measuring assertiveness in both countries (see Table 6).
Level of invariance | S-B χ2 | df | p | AIC | RMSEA | CI (90%) RMSEA | CFI | ΔCFΊ |
---|---|---|---|---|---|---|---|---|
Configural invariance | 80.88 | 42 | .00 | -3.11 | .049 | .032 - .065 | .986 | – |
Weak invariance | 135.02 | 51 | .00 | 33.02 | .065 | .052 - .078 | .970 | -.016 |
Weak -item 4 | 97.28 | 50 | .00 | -2.71 | .049 | .034 - .064 | .983 | -.003 |
Strong invariance | 174.17 | 59 | .00 | 56.17 | .071 | .059 -.083 | .982 | -.001 |
Strict invariance | 178.18 | 68 | .00 | 42.18 | .065 | .053 - .076 | .983 | .001 |
Note. S-B χ2: Satorra-Bentler scaled chi-square; df: degree of freedom; ΔCFI: CFI increment; Weak -item 4: Item 4 was not constricted.
Items | Weight (λ) | Error | R2 |
---|---|---|---|
Colombia | |||
SAS1 | .41 | .91 | .17 |
SAS2 | .88 | .46 | .78 |
SAS3 | .77 | .65 | .59 |
SAS4 | .31 | .95 | .10 |
SAS5 | .74 | .67 | .55 |
SAS6 | .74 | .67 | .55 |
SAS7 | .85 | .51 | .73 |
SAS8 | .89 | .45 | .79 |
SAS9 | .82 | .56 | .68 |
Spain | |||
SAS1 | .44 | .89 | .19 |
SAS2 | .89 | .45 | .79 |
SAS3 | .80 | .59 | .64 |
SAS4 | .70 | .71 | .49 |
SAS5 | .81 | .57 | .66 |
SAS6 | .83 | .55 | .69 |
SAS7 | .85 | .51 | .73 |
SAS8 | .89 | .45 | .79 |
SAS9 | .87 | .49 | .76 |
Stage 1 | Stage 2 | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Country Model 2 | Model 3 | Country Model 2 | Model 3 | ||||||||||
Ítem | χ2(1) | p | R2 | χ2(1) | p | R2 | χ2(1) | p | R2 | χ2(1) | p | R2 | |
Initiation | 1 | 1.08 | 0.29 | 0.002 | 1.50 | 0.22 | 0.003 | ||||||
2 | 0.20 | 0.65 | 0.001 | 0.63 | 0.42 | 0.001 | |||||||
3 | 0.62 | 0.43 | 0.001 | 0.01 | 0.89 | 0.000 | |||||||
Refusal | 4 | 45.05 | 0.00 | 0.072 | 4.17 | 0.41 | 0.006 | 62.92 | 0.001 | 0.101 | 9.18 | 0.002 | 0.014 |
5 | 5.37 | 0.20 | 0.009 | 8.41 | 0.00 | 0.014 | 2.94 | 0.08 | 0.005 | 1.63 | 0.2 | 0.003 | |
6 | 25.14 | 0.00 | 0.041 | 4.09 | 0.04 | 0.006 | 2.94 | 0.08 | 0.005 | 0.58 | 0.44 | 0.001 | |
STD-P | 7 | 0.02 | 0.87 | 0.001 | 3.44 | 0.06 | 0.005 | ||||||
8 | 0.25 | 0.61 | 0.001 | 4.77 | 0.29 | 0.008 | |||||||
9 | 0.41 | 0.52 | 0.001 | 5.68 | 0.02 | 0.010 |
Note. Stage 1 = Initial regression; Stage 2 = Purified regression. Model 1 regression with DIF abscence, Model 2 is a regression with a grouping variable (uniform DIF) and Model 3 is added an interaction between group and total scoring in the test (nonuniform DIF)
Initiation: stage 1: Model 1 = χ2(1) = 0,2; p =.64 R2 = 0.01
Refusal: stage 1: Model 1 = χ2(1) = 24,46; p = .000 R2 = 0.019. Stage 2 Model 1 = χ2(1) = 6,59; p = .10 R2 = 0.011
STD-P: Stage 1: Model 1 = χ2(1) = 8,18; p = .004 R2 = 0.013.
Indicators of external validity are shown in Table 7. Correlations are low and significant for most cases. As observed, differences between correlations in Spain and Colombia are very similar. Guidelines for adapting questionnaires in the same language source inter-culturally are described at Table 8.
Colombia | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
---|---|---|---|---|---|---|---|---|
Assertiveness initiation (1) | 1 | .22** | .16** | .30** | -.33** | -.01 | .16** | .08 |
Assertiveness refusal (2) | .14** | 1 | .22** | -.02 | -.03 | -.23** | .06 | -.28** |
Assertiveness STD-P (3) | .04 | .23** | 1 | -.02 | -.08 | -.07 | .06 | -.03 |
Sexual self-esteem (4) | .27** | -.07 | -.10* | 1 | -.51** | .05 | .09 | .23** |
Sexual depression (5) | -.28** | .05 | .02 | -.50** | 1 | .14** | -.09 | -.42** |
Sexual preoccupation (6) | .06 | -.15** | -.03 | .03 | .12* | 1 | .11* | .20** |
Attitudes toward sexuality (7) | .28** | -.04 | -.02 | .13** | .01 | .27** | 1 | .01 |
Sexual Functioning (8) | .25** | -.17** | .02 | .36** | -.51** | .22** | .02 | 1 |
Note.
*= p < .05;
**= p < .01
|
Note. It is worth mentioning that these guidelines are a suggestion and do not replace the use of international standards in adapting tests.
Discussion
This paper presents a proposal that can be a guide for translations of scales from one language into the same language - which is commonly done amongst Iberoamerican countries (Cova, Bustos, Rincón, Grandón, Saldivia, & Inostroza, 2016; Londoño, Peñate, & González, 2016; Moyano-Díaz, Páez, & Torres; 2016; Ruiz et al., 2016; Ruiz, Suárez-Falcón, & Riaño-Hernández, 2016). This proposal has shown good results in adapting the brief SAS to Colombia, based on the Spanish version. Even though there are aspects to be improved, we believe that this work could provide guidance for adaptations amongst other countries and topics.
This research has proposed a 9-point guide to adapt tests from one language to the same language in another culture, minimizing adaptation issues. When using an instrument, it is necessary to reduce its distortion vulnerability due to factors that may affect measuring, like language (Arafat, Chowdhury, Qusar, & Hafez, 2016). However, it has been taken into account when adapting and translating processes, measurement errors happen when an adapted instrument is applied on a different population who speaks same language (Grisay, 2003, 2007; Grisey & Monseur, 2007). Thus, not only language is to be taken into account, but also cultural differences and linguistic expressions that can be dealt with in any given context. Table 8 - the main result - provides a number of specific guidelines that can be used as supplementary guidance in these adaptation processes.
Present guidelines were proved with the Spanish adaptation of the SAS (i.e., Spanish from Spain to Spanish from Colombia). The process was successfully carried out, in observance of the appropriate psychometric indicators. Problems in item 4 of the Colombian version are consistent in all the analysis (DIF, Invariance and psychometric properties) which show the relevance in keeping up with the guidelines proposed in this research. Invariance analysis is only defendable if item 4 is not constricted. Despite the adaptation problem that was mitigated in the middle of the evaluation, overall psychometric properties of the scale are good with a consistence similar to that observed in the original version (Morokoff et al., 1997), the Spanish version (Sierra et al., 2011) or others (Santos-Iglesias & Sierra, 2010). In addition, there was also enough external validity of the SAS with its respective relations, as observed before (Haavio-Mannila, & Kontula, 1997; Hurlbert, Singh, Menendez, Fertel, Fernández, & Salgado, 2005; Ménard & Offman, 2009); Santos-Iglesias, & Sierra, 2010; Santos-Iglesias, Sierra & Vallejo-Medina, 2013). In turn, this would be an indicator of an adaptation process.
In spite of the fact that a complete methodology was used, the adaptation process was not exempt from any mistakes. Hence, despite having obtained adequate content, construct and external validity, as well as adequate levels of reliability, the adaptation process was not completely accurate. For the avoidance of errors, it is recommended to include a validation of the adapted form following the nature of the backward translation. While this adaptation method could be flawed on a standalone basis, in this case it would help to ensure that the adapted form is true to the original version.
It is worth mentioning that this work is, to the extent known, the first study conducted with this objective. Accordingly, the limitations are great and despite the effort, the method will have to be replicated using other measuring instruments and in other cultures. Moreover, this kind of adaptation from Spanish to Spanish could only be performed using initial or adapted versions which have been made following appropriate guidelines; otherwise, the result would be just a good adaptation of a scale which has been poorly adapted.