PARTIAL LEAST SQUARES REGRESSION ANALYSIS: EXAMPLE OF MOTOR FITNESS DATA / PARCIJALNA REGRESIJA METODOM NAJMANJIH KVADRATA: PRIMJER IZVEDEN NA PODACIMA IZ MOTORIČKOG FITNESA

Abstract

ABSTRACT
Based on the research example, the article attempts to describe the partial least squares regression (PLS) as a tool used for modelling the explanatory variables for the prediction of the dependents. The research was carried out on the fitness data of 52 children, nine anthropometric variables were used as predictors, while the dependents were composed of five motor fitness tests. A two-component model was obtained where a small fraction of the dependent variation (R2Y = .20) was explained by predictors (R2X=.64). The Q2 indicator of the predictive capability of the model was rather low (.16). The main advantages of the PLS were demonstrated: the simultaneous handling of multiple independents and dependents.
Key words: partial least squares regression, PLS, modelling, anthropometric variables, motor fitness data prediction

-----

SAŽETAK

U ovom se radu, na temelju primjera istraživanja, nastoji opisati parcijalna regresija metodom najmanjih kvadrata (PLS) kao metoda za modeliranje eksplanatornih varijabli u predviđanju zavisnih varijabli. Analiza je izvedena na podacima 52 djece, uključujući devet antropometrijskih varijabli koje su predstavljale prediktore, dok su zavisne varijable bile sastavljene od pet testova motoričkog fitnesa. Dobiven je model s dvije komponente u kojem su prediktori (R2X= .64) objasnili mali dio varijabilnosti u zavisnim varijablama (R2Y = .20), pokazatelj kvalitete predikcije modela Q2je bio nizak (.16). U istraživanju je prikazana glavna prednost PLS metode istovremeno uključivanje nekoliko nezavisnih i zavisnih varijabli.
Ključne riječi:parcijalna regresija metodom najmanjih kvadrata, PLS, modeliranje, antropometrijske varijable, predikcija motoričkog fitnesa