Multivariate Statistical Techniques to Manage Multiple Data in Psychology

Introduction: In big-data contexts, multivariate statistical techniques and machine learning methods play a crucial role for the assessment of the interrelations between and within sets of variables. In particular, in social and behavioural sciences, for which the exploration of patterns and mutual interrelation among subject features is needed, a proper use of this technique becomes paramount. Methods: A series of multivariate techniques –clustering, decision trees, principal component, multiple correspondence, partial least discriminate analysis –was applied to a sample of patients with diagnosis of borderline personality disorder (BPD)and bipolar disorder (BD), in order to outline specific socio-demographic and clinical profiles for both the diagnoses. Results: Although the BPD and BD patients are clinically blurred, some features appeared to well discriminate between the two diagnoses. BPD patients are more probably females who have shown self-harm behaviours and/or suicide attempts, while BD are more likely to be males who have never shown self-harm behaviours and have not attempted suicide. Moreover, the assessment variables with more discriminate power were BIS-11, SCL-90 and STAI-T. In particular, patients with SCL-90 total score <36 were more probably BD patients (probability p=87%); whereas patients with SCL90 score >= 36 and a BIS-11 score >=64 were more probably BPD patients (p=83%). Conclusions: The application of multivariate statistical analyses and machine learning techniques allows the definition of specific clinical and diagnostic profiles that can be crucial for taking adequately charge of the patients in a context of precision medicine and an ad-hoc diagnostic and care pattern.