Scoring a forced-choice image-based assessment of personality: A comparison of machine learning, regression, and summative approaches.

Recent years have seen rapid advancements in the way that personality is measured, resulting in a number of innovative predictive measures being proposed, including using features extracted from videos and social media profiles. In the context of selection, game- and image-based assessments of personality are emerging, which can overcome issues like social desirability bias, lack of engagement and low response rates that are associated with traditional self-report measures. Forced-choice formats, where respondents are asked to rank responses, can also mitigate issues such as acquiescence and social desirability bias. Previously, we reported on the development of a gamified forced-choice image-based assessment of the Big Five personality traits created for use in selection, using Lasso regression for the scoring algorithms. In this study, we compare the machine-learning-based Lasso approach to ordinary least squares regression, as well as the summative approach that is typical of forced-choice formats. We find that the Lasso approach performs best in terms of generalisability and convergent validity, although the other methods have greater discriminate validity. We recommend the use of predictive Lasso regression models for scoring forced-choice image-based measures of personality over the other approaches. Potential further studies are suggested.

Measuring Personality through Images: Validating a Forced-Choice Image-Based Assessment of the Big Five Personality Traits.

Selection methods are commonly used in talent acquisition to predict future job performance and to find the best candidates, but questionnaire-based assessments can be lengthy and lead to candidate fatigue and poor engagement, affecting completion rates and producing poor data. Gamification can mitigate some of these issues through greater engagement and shorter testing times. One avenue of gamification is image-based tests. Although such assessments are starting to gain traction in personnel selection, few studies describing their validity and psychometric properties exist. The current study explores the potential of a five-minute, forced-choice, image-based assessment of the Big Five personality traits to be used in selection. Study 1 describes the creation of the image pairs and the selection of the 150 best-performing items based on a sample of 300 respondents. Study 2 describes the creation of machine-learning-based scoring algorithms and tests of their convergent and discriminate validity and adverse impact based on a sample of 431 respondents. All models showed good levels of convergent validity with the IPIP-NEO-120 (openness r = 0.71, conscientiousness r = 0.70, extraversion r = 0.78, agreeableness r = 0.60, and emotional stability r = 0.70) and were largely free from potential adverse impact. The implications for recruitment policy and practice and the need for further validation are discussed.

Game based assessments of cognitive ability in recruitment: Validity, fairness and test-taking experience.

Gamification and machine learning are emergent technologies in recruitment, promising to improve the user experience and fairness of assessments. We test this by validating a game based assessment of cognitive ability with a machine learning based scoring algorithm optimised for validity and fairness. We use applied data from 11,574 assessment completions. The assessment has convergent validity (r = 0.5) and test-retest reliability (r = 0.68). It maintains fairness in a separate sample of 3,107 job applicants, showing that fairness-optimised machine learning can improve outcome parity issues with cognitive ability tests in recruitment settings. We show that there are no significant gender differences in test taking anxiety resulting from the games, and that anxiety does not directly predict game performance, supporting the notion that game based assessments help with test taking anxiety. Interactions between anxiety, gender and performance are explored. Feedback from 4,778 job applicants reveals a Net Promoter score of 58, indicating more applicants support than dislike the assessment, and that games deliver a positive applicant experience in practise. Satisfaction with the format is high, but applicants raise face validity concerns over the abstract games. We encourage the use of gamification and machine learning to improve the fairness and user experience of psychometric tests.