Cross-cultural adaptation of the Danish version of the Big Five Inventory – a dual-panel approach

2.1 Design

This paper follows the guidelines for reporting reliability and agreement studies [31]. Using a multistep approach involving a centralized review process has been considered an important factor for securing the best possible translation [32]. Therefore, the translation process was divided into three steps; a translation phase and then a two-step validation phase, which was evaluated by the group that performed the initial translation. Lastly, the test-retest reliability and internal consistency of the Danish version of the BFI was evaluated.

2.2 Ethical considerations

According to the Danish Act on Research Ethics Review of Health Research Projects, research that only involves interviews and questionnaires does not require approval by the Ethics Committee. Nevertheless, no sensitive personal data were collected in this study. The study was reported to the Danish Data Protection agency.

2.3 The Big Five Inventory

When filling out the questionnaire, the responder indicates how well he/she agrees with short statements by help of a 5-point Likert scale (1: Disagree strongly, 2: Disagree a little, 3: Neither agree nor disagree, 4: Agree a little and 5: Agree strongly). The total score is calculated for each of the five subscales which in turn consists of several items (Extroversion: 8 items, Agreeableness: 9 items Conscientiousness: 9 items, Neuroticism: 8 items and Openness: 10 items). For some of the questions, the score is reversed before calculating the final score [21].

2.4 Translation

The translation was performed using a dual-panel approach, following the recommendations by Swain-Verdier and colleagues [33]. The translation process was conducted by a bilingual panel (panel 1) and a panel consisting of laymen (panel 2). The bilingual panel consisted of 4 persons with fluent Danish and English language skills (Table 1). Members of the bilingual panel independently translated the English version of the BFI [21] into Danish before attending a consensus meeting led by one of the group members (TSP). During this meeting the translations were compared and if there was any disagreement regarding the translation, this was discussed until consensus was reached. This version of the questionnaire was then administered to a group of 8 laymen (panel 2; Table 1) who were asked to qualify the wording of the translated questionnaire with particular focus on whether the translation would be understood by laymen in clinical and non-clinical populations. Prior to reviewing the questionnaires, the purpose of the study was explained by one of the group members (MHP). After reviewing the questionnaire independently, the panel members participated in a focus group-interview led by MHP who, following the interview, reported any suggestions for alternate wording back to the leader of the bilingual panel (TSP). Lastly, the bilingual panel met again to discuss the suggestions put forwards by panel 2 and implemented as considered appropriate.

For the two panels involved in the translation process, it was assumed that the saturation of data could be obtained by using the proposed methods [34] and therefore, no formal power calculations were performed. Nevertheless, the study aimed at recruiting between 5 and 10 individuals for each panel in consistency to what previous studies with similar aims have done [35], [36].

2.5 Face validity

Although the main purpose of the Big Five Inventory is to quantify five different personality traits [21], there is growing interest in investigating whether specific traits could be related to clinical conditions [9], [10], [12]. With this in mind, a group of 9 persons undergoing rehabilitation for various clinical conditions reviewed and filled out the questionnaire independently (panel 3; Table 1). Subsequently, they participated in a focus group interview led by MHP with the aim of recording any feedback regarding the questionnaire. The feedback from the interview was then reported back to the bilingual panel who integrated the suggested changes into the final version of the questionnaire as deemed appropriate.

2.6 Test-retest reliability, internal consistency and factor solution

Personality traits are likely to be stable in the short-term, while changes are likely to occur across the life span [37], [38]. To determine the reliability and internal consistency of the Danish version of the BFI, 100 people were asked to fill out the questionnaire with a 7-day time interval. A Principal Component Analysis (PCA) with a five-factor varimax rotation was performed to evaluate the construct validity i.e. how well the measured items (indicator variables) loaded onto the different subscales (constructs) of the questionnaire.

An a priori power calculation was not performed but this group size was estimated to be sufficient based on previous studies with similar aims [39], [40].

2.7 Analysis

To investigate the relationship between the individual items in each sub-scale and thereby internal consistency of the questionnaire, the Cronbach’s coefficient α level was determined for each of the five subscales. The coefficient ranges between 0 and 1, where numbers closer to 1 indicate a better relationship between each item within each respective subscale. Previously, it has been suggested that a correlation between 0.7 and 0.95 is acceptable [41], [42], [43]. In case some of the scales revealed α values lower than 0.7, an additional analysis was performed to determine whether any single item in the subscale was responsible for a low value.

To determine the reliability of the questionnaire (test-retest consistency), the Intraclass Correlation Coefficient (2.1) was calculated for each of the five subscales. Here, coefficients were considered low (0.26–0.49), moderate (0.50–0.69), high (0.70–0.89) and very high (0.90–1.00) [44]. To account for the difference between the two measurements and the systemic errors within the two measurements, the value of SEM_consistency was derived by dividing the SD of the mean differences between two measurements (SD_difference) by 2, as recommended by de Vet et al. [45]. The smallest detectable change was calculated to determine the minimum change needed for the change to be considered a true change (i.e. not because of measurement error) [42]. For calculating the smallest detectable change, the following formula was used: Standard error of measurement * 2 * 1.96 using Microsoft Excel 2016 (Microsoft Inc., Redmond WA, USA).

All statistical analyses were performed using SPSS^® (IBM Corp. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp).

3.1 Translations and face validity

Both panels provided feedback that resulted in changes in the final version of the translated questionnaire (Appendices i and ii). In general, the participants in both panels considered the questionnaire to be too long and were concerned that elderly people and persons in pain would struggle with filling out the questionnaire in one go.

3.2 Internal consistency, test-retest reliability and factor solution

The Cronbach’s α ranged from 0.66–0.84. Only the value for the subscale agreeableness was below the desired cut-off value of 0.70 (Table 3). The removal of the first question in the scale (Tends to find fault with others [Har tendens til at finde fejl hos andre]) raised the α level to 0.71 (Table 4). For the test-retest reliability, the ICC-values were high to very high ranging from 0.86 to 0.95 (Table 2). The smallest detectable change ranged from 1.13 to 1.70 (Table 3).

Based on the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity, the assumptions for performing a PCA were met (KMO: 0.659, Bartlett’s: 0.000001). The five factors all had an eigenvalue above 1 (2.57–8.03) and the five factor solution explained 48.6% of the variance with factor loadings ranging from −0.067 to 0.843 with an average value of 0.443. Cross-loadings ranged from −0.465 to 0.843 with an average value of 0.039. Two of the sub-scales (agreeableness and conscientiousness) loaded primarily on the same factor (Appendix iii; Factor loadings above 0.3 are marked with bold as factor loadings below 0.3 are considered minimal [46]). Moreover, the sub-scale for openness loaded similarly on two factors.

4.1 The translation process and feedback

Using a dual panel process for translating questionnaires is advantageous, as it is preferred by laymen participating in the translation process without compromising the content or internal consistency as compared to the forward-backward translation method [47], [48]. Both panel 2 and panel 3 considered the questionnaire to be too long (Appendices i and ii). Moreover, panel 2 was concerned that the length might affect the ability of elderly people and/or people in pain to use the questionnaire (Appendix i). This was to some extent confirmed by the comments from panel 3 (Appendix ii). These comments warrant a translation and cross-cultural adaption of the short version of Big Five Inventory (BFI-10) [23].

4.2 Face validity

A patient group was chosen to participate in the third panel to get an impression of how laymen with clinical conditions responded to the translated questionnaire. The questionnaire is designed to measure the five most commonly acknowledged personality traits, but it is not capable of e.g. quantifying function or severity of a physical condition. Including a group with a range of underlying clinical conditions was nevertheless considered important to secure the face validity of the questionnaire especially when considering the increasing interest in the interaction between personality and clinical conditions such as chronic pain [9].

4.3 Internal consistency, test-retest reliability and factor loadings

Surprisingly, the Cronbach’s α value for agreeableness was below the desired 0.7 level (Table 2). This contrasts what has been reported in previous translations such as the Dutch [26] and Norwegian [29] versions. Nevertheless, translations into other languages have shown discrepancies in the internal consistency of the same subscale [28], [49], [50], [51]. It is unlikely that this can be related to the translation of the questionnaire as only one of the comments from panel 2 or 3 were related to this subscale (questions 7; Appendices i and ii). Removing the first question from the subscale items did improve the α level sufficiently while removing other questions did not (Table 3).

The test-retest reliability was between high and very high which is comparable to what was seen in the Italian [27] and Dutch [26] translations of the BFI.

The findings from the principal component analysis were different from what was seen previously [30] where items from two sub-scales (Agreeableness and Conscientiousness) loaded on the same factor in an almost equal manner (Appendix iii). Although the assumptions for performing the principal component analysis were met, it is likely that the study was underpowered to detect a five-factor structure considering that it is recommended that the ratio between number of subjects per item should be between 3 and 10:1 [52–54]. Moreover, factor analyses run on a subject-item ratio similar to this study (2:1) have been shown to produce correct factor solutions in only 1 out of ten cases [55]. Considering that a five-factor structure was detected in versions translated into languages culturally and geographically similar to Denmark [28], [29], it is likely that a larger sample would be comparable to the findings in the Norwegian (n=389) and German (n=480) versions. This however, warrants further investigations of the construct validity of the Danish version of the BFI.

4.4 Conclusion

This is the first time the Big Five Inventory has been translated and cross-culturally adapted to the Danish language. The translated version demonstrated high to very high test-retest reliability and, for most parts, an acceptable internal consistency. The construct validity was however different from versions translated into languages geographically and culturally similar to Danish. The length of the questionnaire was a concern for laymen with and without a clinical condition, which highlights the need for a Danish version of the short version of the Big Five Inventory to increase feasibility of its use in research and clinical practice.

4.5 Implications

The current results indicate that the Danish version of the Big Five Inventory can be a useful tool for evaluating the Big Five personality traits in Danish populations. This can be valuable for many reasons e.g. when assessing people in pain. The fact that pain is a subjective experience, which can be affected by multiple domains [56] is why there is a broad consensus regarding people in pain needing to be assessed within a biopsychosocial framework [57]. Although the biopsychosocial approach is broad and inclusive, it has recently been considered inadequate, as it does not account for the dynamic nature of emotional and physical health [58], [59]. In fact, it has been suggested that certain personality traits may play a role in this dynamic relationship which may imply that the concept of a “pain personality” exists [9]. Moreover, the authors concluded that assessing personality traits might be important, especially in patients where the best available care was unhelpful.

When assessing people in pain and attempting to determine whether personality traits may contributed to their condition, it is important to note that normative values for personality do not exist. It is possible to compare individual scores to data from large populations such as those included in the German [28] or Dutch [26] translations. Here however, it needs to be acknowledged that these studies were not designed to provide normative data. Assessing personality traits in pain populations may however indicate patterns or relationships with clinical variables such as pain intensity, disability or coping strategies.