“Items selection for building classification trees”.

RICHARD'S, MARÍA MARTA; SOLANAS, ANTONIO; ANDRÉS, AMARA; MANOLOV, RUMEN

Universidad de Oviedo, España. Modalidad de presentación: Póster

Congreso; III European Congress of Methodology.; 2008

European Association of Methodology and Society for Multivariate Analysis in the Behavioural and Social Sciences (SMABS)

Several psychometric tests have been proposed to measure personality dimensions and to assess personality disorders. However, it is not common to carry out studies in which the classification level of accuracy is estimated. The purpose of this work is to select items from the Millon’s Clinical Multiaxial Inventory by means of classification trees in order to identify those items associated with a higher probability of suffering from Ischemic Cardiovascular Acute Episodes (ICAE). A sample of 313 participants was divided into two  groups (clinical and control), which were balanced by sex, age, socioeconomic and educational levels, and were used to extract classification rules. The analyses carried out were: a) Global analysis for each of the most significant MCMI-II personality scales (Richard´s & Solanas, 2008) were used to choose the items with larger propensity values and more useful to detect people who suffer from ICAE; b) Once the most relevant items were identified for the 9 scales, sensitivity item analyses were separately conducted for each of the 9 personality scales in order to discard redundant items. In order to carry out the sensitivity analysis, each item selected in the step a) was extracted and then the percentage of correct classification was computed again. This process was separately carried out for each scale. If the reduction of accuracy was less than 3%, the item was excluded; c) Finally, two separate analyses by gender with the refined set of items were carried out to compare the classification rates of this analysis with the global analysis. The results show that a reduced set of items are sufficient to reach a significant level of individuals correctly classified. Additionally, this reduced set of items can be used by clinical psychologists to implement low and efficient time-consuming systems for identifying those individuals who belong to the clinical population.