Socially desirable responding within the context of privacy-related research: A personality perspective

on the other hand, had the opposite effect. Conclusion: Based on the results, the authors conclude that neurotic individuals seem predisposed towards responding in a socially desirable manner within the context of privacy-related surveys. The authors, therefore, advise researchers within the field of privacy-based personality studies to take care when analysing their results.


Introduction
Social networking sites (SNSs) have become an integral part of daily life, mainly because they facilitate and fulfil (at least to some extent) individuals' needs to belong. They actively encourage their users to share personal information. Whilst SNSs assist geographically separated individuals to stay in touch, there are also significant privacy concerns related to indiscriminate sharing. Such indiscriminate sharing may unwittingly sacrifice online privacy. Given that Facebook is one of the most popular SNSs, the authors have used it as their empirical case. The authors argue this to be an appropriate choice given the recent high profile privacy scandals, which have revealed the true breadth and depth of Facebook's collection and use of personal information (Brown 2020). Overor unwise-sharing exacerbates privacy loss, something that has the potential to be harmful. For example, Househ (2011) wrote about the negative and unintended consequences of sharing health data on Facebook, whilst Taraszow et al. (2010) highlighted the harm that can ensue from teenagers sharing personal information.
Researchers often use self-reported survey responses to study important social media issues, such as those outlined here. There is, however, some anxiety about the reliability of such self-reported survey responses within the context of privacy research (Af Wåhlberg & Dorn 2015). One reason for such anxiety is the tendency for certain individuals to provide socially desirable responses. Although such responses may be related to the privacy paradox, we do not centre our argument thereon. This is especially so because the existence of the privacy paradox is both supported (Dienlin & Trepte 2015) and questioned (Solove 2021). Instead, this study focuses on the impact of Big Five personality traits on a tendency to provide socially desirable responses within the context of self-stated privacy-related survey responses. Our results can inform future privacy-related research by highlighting the personality traits that could lead to socially desirable responses being given to surveys. This could help researchers (and organisations) that make use of self-reported surveys to improve the quality of their data and associated conclusions. This is especially pertinent within context of personality-based privacy decision-making.
This article is structured as follows: after providing a brief background, several hypotheses are developed, followed by an outline of our methodological approach, analysis and results. This is followed by a brief discussion of our results and an outline of the limitations of this study and areas for future research.

Hypothesis development
Many researchers use self-reported surveys to explore privacy-related decision-making. The veracity of the findings relies on respondents being honest and frank in their responses. One aspect that can confound such research is that survey respondents might provide socially desirable rather than honest responses (Milne, Rohm & Bahl 2004). Therefore, to use surveys in personality-based privacy research, such forms of socially desirable responding should be taken into account. In this article, the Big Five personality traits' influence on privacy-related decision-making is considered and empirically situated within the context of Facebook privacy: specifically, respondents' intention to use their Facebook privacy settings. Often referred to using the acronym OCEAN, the Big Five personality traits that this study explores includes: openness, conscientiousness, extraversion, agreeableness and neuroticism (Costa & McCrae 1992).

Openness
Individuals high in openness (OPEN) are naturally inquisitive and intelligent and revel in new experiences (Vishwanath, Xu & Ngoh 2018). They are, however, often labelled as 'anti-establishment'. In other words that which is trendy and new is embraced whereas established practices are questioned. Aside from general characteristics, most studies have found either weak, non-significant or negative correlations between openness and social desirability. For example, Stober (2001) found openness to be uncorrelated with social desirability during the development of their social desirability scale (i.e. SDS-17) -specifically from a validity perspective. Evidence suggests that open individuals are also inclined to overcome those inhibitions that prevent them from acting as expected (i.e. to exhibit their best behaviour). This is particularly important given that Uziel (2010) also reports non-significant results between openness and impression management when evaluating the relationship of the latter concept with social desirability. Similar results are reported by Crant and Bateman (2000), who found no significant correlation between openness to experience (another name for openness) and social desirability. Tangentially, and in an effort to discern answering style from undesirable answering, McCrae and Costa (1983) found little evidence to suggest that openness would provide socially desirable answers. What they did find was a significant negative relationship between openness and the tendency to lie when completing self-reported instruments. Similar significant (and negative) results are reported by Flynn (2005) in a study about interracial attitudes, as well as by Egelman and Peer (2015). Together, the aforementioned provide evidence to suggest that open individuals are, in the least, not likely to provide socially desirable answers -albeit within a more psychometric context. Of course, given the focus on social desirability within the context of Facebook-based privacy behaviour, the authors are also compelled to provide personality-based evidence from a privacy perspective.
Although relatively few studies have been conducted on the direct relationship between openness and Facebook privacy (and social desirability), there are some who have theorised the relationship of openness with privacy settings. For example, Van der Schyff, Flowerday and Lowry (2020) found openness to exhibit a significant positive relationship with Facebook users' attitude towards privacy settings when using Facebook apps. A study on the correlations between the Big Five personality traits and various privacy protective behaviours, found openness to significantly predict specific privacy protection strategies. For example, deleting a tag associated with a photo or video (Gerber, Gerber & Hernando 2017). It has also been demonstrated that openness is significantly related to privacy protective behaviour within the context of people periodically reviewing their privacy settings (Van der Schyff & Flowerday 2021). Given their inquisitive and intelligent nature (amongst other characteristics), it is therefore likely that such individuals will make informed privacy decisions. The authors therefore hypothesise that: H1a: Openness will exhibit a significant negative relationship with social desirability. In other words, individuals who are high in openness are unlikely to provide socially desirable answers.
H1b: Openness will exhibit a significant positive relationship with intention to use privacy settings.

Conscientiousness
Unlike open individuals, those high in conscientiousness (CON) are conformist and adhere to societal norms. Conscientious individuals are also cautious, organised and tend to be competitive (Hao, Yang & Shi 2019). Studies focused on the relationship between conscientiousness and social desirability have found significant positive correlations when completing psychometric evaluations (Ones & Viswesvaran 1996). Stober (2001) also found significant overlap between conscientiousness and desirable responding, specifically when evaluating the discriminant validity of the SDS-17 scale. Similar positive correlations with related concepts such as impression management are reported by Uziel (2010) and Stöber, Dette and Musch (2010), indicating that highly conscientious individuals are wary as to the way their responses may influence favourable impressions. Stöber et al. (2010) also found conscientiousness to be positively correlated with self-deceptive enhancement (SDE) -a means to provide positively biased responses to maintain positive self-esteem. It is therefore evident that highly conscientious individuals may, in certain circumstances, provide socially desirable answers.
As characterised above, individuals high in conscientiousness are cautious when it comes to engaging in privacy protective behaviour. This applies to both organisational contexts and social media contexts. For example, Van der Schyff et al. (2020) found conscientiousness to be positively related to privacy attitudes, specifically within the context of privacy settings when using Facebook apps. Similar results are reported by McCormac et al. (2017), as well as Van der Schyff and Flowerday (2021), who found conscientiousness to be positively related to information security awareness. The latter authors also found a significant positive relationship between conscientiousness and the intended use of Facebook privacy settings. Further research on the security influence of the Big Five personality traits also found a significant positive relationship between conscientiousness and cybersecurity behaviour (Shappie, Dawson & Debb 2019). Given the evidence presented, the authors hypothesise that: H2a: Conscientiousness will exhibit a significant positive relationship with social desirability. In other words, individuals who are high in conscientiousness are likely to provide socially desirable answers.
H2b: Conscientiousness will exhibit a significant positive relationship with intention to use privacy settings.

Extraversion
Individuals high in extraversion (EXT) are outgoing, sociable and assertive (Costa & McCrae 1992). They are also more likely to engage in risky behaviour (Pentina et al. 2016). However, and despite their social nature, some research has found extraversion not to be highly correlated with social desirability. For example, Ones et al. (1996), as well as McCrae and Costa (1983), found low (almost negligible) correlations between extraversion and social desirability. A similar (low) correlation between extraversion and social desirability is reported by Crant (1995) -albeit in a study focused on job performance amongst real estate agents. Tangentially, studies on impression management have also found either nonsignificant or negative correlations with extraversion (Uziel 2010). Especially, in Uziel's (2010) study, impression management is viewed as the capacity to self-regulate impulsive (and arguably irrational) behaviour within social settings, in doing so gaining social approval within those settings as a function of socially desirable responses.
Unlike conscientious individuals, those high in extraversion typically exhibit negative attitudes towards the privacy of their personal information. Recent studies have found this to be particularly prominent within the context of social media use. For example, Van der Schyff et al. (2020) found extraverted respondents to harbour negative attitudes towards the use of Facebook privacy settings. The same applies to highly extraverted individuals' intention to periodically review their privacy settings (Van der Schyff & Flowerday 2021). Related work -specifically that within the field of privacy concerns -has also found extraversion to be either negatively related to privacy concerns (Pentina et al. 2016) or not significantly related at all (Osatuyi 2015). We argue that highly extraverted individuals are likely not to value the privacy of their personal information to the same extent as some of the other traits. The authors therefore hypothesise that: H3a: Extraversion will exhibit a significant negative relationship with social desirability. In other words, individuals who are high in extraversion are unlikely to provide socially desirable answers.
H3b: Extraversion will exhibit a significant negative relationship with intention to use privacy settings.

Agreeableness
Highly agreeable individuals are sensitive and trusting and tend to be concerned about the feelings and well-being of those around them (Costa & McCrae 1992). Crucially, highly agreeable individuals tend to be honest and some studies have found these individuals to provide socially desirable answers (Stober 2001). As with some of the other traits, studies focused on the measurement of impression management (De Vries, Zettler & Hilbig 2014) and selfregulation (Stavrova & Kokkoris 2019) have also found agreeableness (AGR) to correlate with socially desirable answering. It is therefore likely that agreeable individuals will portray more positive or socially acceptable behaviour when conversing with peers in groups where they must regulate what is said. Additionally, and according to Bansal, Zahedi and Gefen (2016), agreeable individuals are likely to avoid behaviour that deviates from that which is acceptable within a certain context. Together with their socially compliant nature (Costa & McCrae 1992), this may indicate that these individuals are likely to value the privacy of their personal information. This may be exacerbated by their sensitivity to upsetting situations (Karim, Zamzuri & Nor 2009) such as the potential loss of personal information in addition to social embarrassment, especially if their personal information is used inappropriately. This also ties in with research by Junglas and Spitzmuller (2006), who found that agreeable individuals harbour privacy concerns -specifically within the context of location-based services. A similar result is reported by Tang, Akram and Shi (2020) -albeit within the context of app use. Given such concerns, a positive privacy attitude is likely. In fact, recent research has found direct evidence of such positive privacy attitudes. For example, Van der Schyff et al. (2020) found that highly agreeable individuals exhibited positive attitudes towards the use of Facebook privacy settings, specifically within the context of Facebook apps. It is therefore hypothesised that: H4a: Agreeableness will exhibit a significant positive relationship with social desirability. In other words, http://www.sajim.co.za Open Access individuals who are high in agreeableness are likely to provide socially desirable answers.
H4b: Agreeableness will exhibit a significant positive relationship with intention to use privacy settings.

Neuroticism
The final trait in the OCEAN personality model is that of neuroticism (NEU). Highly neurotic individuals are negative, suffer from impulsivity, are nervous and tend to be preoccupied with worrying (Costa & McCrae 1992). From a social desirability perspective, a number of studies report a positive relationship between neuroticism and social desirability or socially desirable answering (Thomsen et al. 2005). Similar results are reported by both Heaven and Shochet (1995) and Jackson and Francis (1998) who found neuroticism to be significantly related to the Lie Scale -albeit only for men.
Individuals high in neuroticism often experience more threats and heightened levels of anxiety. These individuals also tend to be concerned about information privacy (Junglas & Spitzmuller 2006;Van der Schyff et al. 2020). Such concerns align with recent evidence, which suggests that neurotic individuals perceive the periodic review of Facebook privacy settings positively (Van der Schyff & Flowerday 2021). As a result, it is argued that individuals high in neuroticism will value the privacy of their Facebookbased personal information. This, in turn, may result in the increased use (and review) of their Facebook privacy settings. Their anxious nature may further exacerbate not only the fear of losing control of their personal information but also the possibility of suffering security (and privacy) violations as a result of not making adequate use of these settings. Based on the given discussion, the authors hypothesise that: H5a: Neuroticism will exhibit a significant positive relationship with social desirability. In other words, individuals who are high in neuroticism are likely to provide socially desirable answers.
H5b: Neuroticism will exhibit a significant positive relationship with intention to use privacy settings.

Methodological approach
After receiving ethical clearance from the primary author's institutional ethics committee (ref: CIS18-10), primary data were collected as part of a large cross-sectional survey.

Data collection and screening
A total of 651 responses were collected from citizens of the United States of America above the age of 18 who are active Facebook users. The qualifying criteria were used as part of an Amazon Mechanical Turk (AMT) human intelligence task.
To improve the quality of questionnaire responses several criteria were applied, which are summarised as follows: • Response completeness: Twelve incomplete responses were eliminated. • Completion time: Questionnaire responses had to have been completed in no less than 6 min. This ensured that our resulting statistical analysis did not include responses from respondents who were 'speeding' (Zhang & Conrad 2014). Fifty-eight responses that were completed in less than 6 min were eliminated. • Attention trap correctness: The survey included two attention trap questions, both of which had to have been correctly answered for the response to be included in the analysis. Only five respondents incorrectly answered both attention trap questions leading to their elimination.
After applying the aforementioned criteria, 576 (n = 576) valid responses remained and were subsequently used in the development of the structural model. See Table 1 for a complete outline as to the distribution of the sample demographics.

Measures used
All the questionnaire items were adapted from existing instruments. To evaluate an individual's personality traits, the 44-item Big Five Inventory (BFI) (John & Srivastava 1999) was used. To evaluate an individual's level of social desirability (modelled as SD in our structural model), a shortened (8-item) version of the original 33-item Marlowe-Crowne scale was used (Ray 1984). An individual's intention to use privacy settings was evaluated by adapting items from similar behavioural information privacy instruments Taneja, Vitrano & Gengo 2014).

Analysis and results
To develop the structural model, SmartPLS v3.3.3 was used (Ringle, Wende & Becker 2015). The choice of partial least squares structural equation modelling (PLS-SEM) was driven by two factors. Firstly, the primary data were not normally distributed and secondly, PLS-SEM is particularly adept at evaluating recursive models of an exploratory nature (Hair et al. 2017). To the authors' knowledge, no studies have used a social desirability scale in this manner.

Measurement model evaluation
To evaluate the measurement model, the authors first ensured that all the items exhibited acceptable outer loadings. To this end, outer loadings were assessed using a PLS-specific approach consisting of two decision criteria advocated by Hair et al. (2017): • Eliminating items with outer loading below 0.4. • Eliminating items with outer loadings between 0.4 and 0.7 if they do not increase the average variance extracted (AVE) value of the latent variable in question.
After assessing the outer loadings, it was ensured that all the remaining items loaded significantly (i.e. t-values in excess of 1.96) on their intended latent variable. The magnitude of the AVE value of each latent variable was also assessed. None of the items exhibited a t-value below 1.97 and all the latent variables exhibited an AVE in excess of the accepted 0.5 threshold. Based on these factors it was determined that the questionnaire satisfied convergent validity. To ensure that our model satisfied the criteria for discriminant validity, three techniques were used. Firstly, the heterotrait-monotrait (HTMT) ratios were assessed, which were all below the accepted threshold of 0.85. Secondly, the values on the diagonal of Table 2 (presented in bold), commonly referred to as the Fornell-Larcker criterion, were assessed. As a final means to assess discriminant validity, it was ensured that each item loaded the highest on its intended latent variable. Based on our assessments the model was declared valid from a convergent and discriminant perspective.
To assess the reliability of the model the Cronbach's alpha (CA) and composite reliability (CR) value of each latent variable was inspected to ensure that they were within accepted thresholds. Although the CA value for the social desirability latent variable is slightly below 0.7, the authors argue this to be an acceptable outcome given that Hair et al.
(2017) explicitly stated thresholds of 0.6 to be acceptable when creating path models of an exploratory nature. Furthermore, evaluation of the influence of social desirability within the context of Facebook privacy is exploratory. The variance inflation factor (VIF) value of each item was inspected to check for signs of multicollinearity. None of the items exhibited VIF values in excess of 5, thus eliminating multicollinearity. See Table A1 in the Appendix for a complete outline of this study's instrument and associated descriptive statistics.

Structural model evaluation
Two indicators were used to assess the quality of the structural model. Firstly, the in-sample predictive power of both endogenous latent variables (expressed as R 2 ) was inspected. In short, the R 2 value of an endogenous variable explains to what extent the related exogenous variables influence the variance in the endogenous variable in question.
Here, it was found that the personality traits evaluated explained 26.6% (R 2 = 0.266) of the variance in the social desirability latent variable and 10.7% (R 2 = 0.107) of the variance in the latent variable intention to use privacy settings. Secondly, the out-of-sample predictive power (Stone-Geisser's Q 2 ) of the endogenous latent variables in the model was assessed (social desirability = 0.12 and intention to use privacy settings = 0.05, respectively). Given that both these Q 2 values were in excess of 0, it was concluded that the model has an adequate amount of predictive power (Stone 1974).
As per Table 3, there is no support for hypotheses 1a, 3a or 4a. There is, however, support for hypotheses 1b and 2b. In other words, both openness and conscientiousness significantly influence intention to use privacy settings. Hypothesis 2a is also supported, indicating that conscientiousness is significantly related to social desirability. This provides some  http://www.sajim.co.za Open Access evidence to suggest that within the context of intended use of Facebook privacy settings, conscientious individuals may provide socially desirable answers. In addition, hypothesis 3b is also supported, indicating that extraverts are not inclined to use privacy settings. The results of this study also provide support for hypothesis 4b. Therefore, together with the result of hypothesis 4a, there is evidence to suggest that within the context of Facebook privacy decision-making, agreeable individuals tend not to provide socially desirable responses (i.e. they don't 'fake good'). Given this study's focus on the intended use of Facebook privacy settings, this suggests that these individuals' intentions are more likely to result in similar actual behaviour. In other words, they would likely use Facebook privacy settings. The results also provide support for hypotheses 5a and 5b. In particular, the significant positive relationship between neuroticism and social desirability indicates that neurotic individuals may provide socially desirable responses within the context of privacy decision-making. Considering this result within the context of the significant positive relationship between neuroticism and intention to use privacy settings, it can be argued that neurotic individuals are not likely to make actual use of their Facebook privacy settings, despite their intention to do so. See Figure 1 for a visual illustration of our research model.

Discussion
It is important to emphasise that the authors are not arguing the extent to which each of the various personality traits influence Facebook privacy behaviour as a direct interaction with an individual's level of social desirability (e.g. as an interaction term ). Instead, firstly they used the eight-item Marlowe-Crowne scale to determine those traits within the Big Five that relate significantly to a tendency to provide socially desirable responses to privacy-related surveys. If the relationship is significant, it was also deemed appropriate to evaluate the significance of the relationship of these traits with the veracity of responses expressing an intention to use the privacy settings. For example, if a positive relationship is exhibited in relation to social desirability, those individuals who are characterised by that personality trait will tend to provide socially desirable answers and their privacy-related questions might not reflect their actual intention. The opposite is true when a negative relationship is exhibited in relation to the tendency to provide socially desirable responses. Their stated intention to engage with privacy settings can be taken at face value, given that it is less likely to be a socially desirable response.
The results of this study indicate that conscientiousness, agreeableness and neuroticism are significantly related to both social desirability and intention to use privacy settings. However, it is the significant negative relationship between agreeableness and social desirability that is of particular interest. This may, for example, indicate that the agreeable individuals are more likely to provide honest responses (not faking good). This applies specifically, to the privacy-related decision-making context. In other words, an agreeable individual's intention to use their privacy settings is likely to be honest. Within the context of this study's main objective, this implies that agreeable individuals are likely to provide less socially desirable self-reported responses when completing privacyrelated surveys. We found the converse to be true for those who score high on neuroticism. In essence, neurotic individuals are likely to intend using their privacy settings, but may not actually do so. This, in turn, suggests that these individuals are more likely to provide socially desirable responses when completing privacy-related surveys.
This finding calls the existence of the privacy paradox into question, confirming Solove's (2021) reservations. Consider that the paradox manifests when people's stated intentions do not match the privacy-preserving actions they engage in. As such, neurotics might well give socially desirable responses, saying that they intend to take privacy protective actions (i.e. misrepresenting their intentions).

Practical and research implications
Based on the results, it might be advisable for privacy researchers to include questions that evaluate respondents' personality traits. This is likely to complicate the analysis process and lengthen the surveys. There are, however, established short form BFI scales that have proven to be reliable in similar research. This is likely to be an acceptable trade-off, given that it will enable privacy researchers to improve the trustworthiness of survey responses. For example, some respondents' responses (e.g. those high in neuroticism) cannot be relied on to be accurate. Having said  this, the results of this study do not indicate the threshold, within the neuroticism subscale, at which the relationship between neuroticism and social desirability will significantly impair the veracity of responses. Determining this threshold would be an interesting topic of future research. Being able to pin down this threshold would enable a more granular approach when eliminating untrustworthy responses.
In this article the authors have highlighted a particular confound that makes stated privacy-related intentions less reliable indicators of actual future privacy-related behaviours. It seems that, in order to truly understand privacy behaviours and related decision-making, all data from self-reported instruments should not be expected to be truthful. It is possible that excluding responses from highly neurotic respondents would improve data quality. A better option would be to find ways to observe actual future privacy-related behaviours, especially where an intervention is being evaluated, which encourages people to engage with privacy-related settings. Unless unreliable responses are removed it cannot be concluded, with any degree of confidence, that the data, which includes self-reported intentions, is indeed a reliable measure of privacy decisionmaking.

Limitations and future research
This study is subject to several limitations -all of which point to interesting areas for future research. For example, a large cross-sectional survey was used, which does not provide much latitude to generalise about other similar contexts. This ties in with the second limitation because participants from only the United States of America were considered for this study. This likely presents only one specific view of privacy setting use: especially how it affects the data collected from apps and websites that are integrated with Facebook. For example, many websites and apps enable individuals to use their Facebook credentials to authenticate using the Facebook Login developer feature.

Conclusion
In this article, the extent to which the Big Five personality traits influence the provision of socially desirable responses within the context of self-reported privacy-related intentions, specifically, the intention to engage with Facebook privacy settings was investigated. It is clear from the results that socially desirable responding may indeed have influenced other studies. Our results can help to improve the reliability of self-reported surveys that evaluate personality traits as part of privacy-related studies.