Substituting missing data

V sodobnem svetu so anketni vprašalniki, z namenom ocenjevanja lastnosti ciljne populacije, vedno bolj popularni. Zaradi problema manjkajočih podatkov pa je analiza le-teh lahko problematična. Obstaja veliko metod, s katerimi se spopadamo s problemom manjkanja podatkov. Za praktični del sem se odločila, da uporabim metodo popolnih primerov, deterministično imputacijo in večkratno imputacijo z linearno regresijo. Cilj diplomske naloge je ugotoviti, kako se metode obnašajo na določenem vzorcu podatkov in katera nam da najbolj optimalen rezultat. Zanimalo me je, ali lahko v splošnem določimo boljšo metodo ali je odvisno od celotne zgradbe ankete in odgovorov. Iz rezultatov lahko ugotovimo, da je odvisno tudi od drugih dejavnikov, katera metoda je bolj učinkovita. Pomembno je predhodno znanje teorije, da lahko določimo metodo, ki jo bomo uporabili za določen primer.In modern world, survey questionnaires with the intention of assessing properties of the targeted group of people, are becoming more and more popular. But because of the missing data, the analysis of these surveys might sometimes be problematic. There are many different methods that can help us cope with missing data. For the practical part, I have decided to choose a method of complete-case analysis, single imputation with linear regression and multiple imputation with linear regression. The goal of dissertation is to find out, how different methods work on the certain samples of data, and which method gives us the most optimal result. I was interested if we can generally determine the best method or does it depend on the whole structure of the survey and its answers. From the results we can see that determining the best method also depends on other factors. Beforehand knowledge of the theory is important to determine which method will be used for certain examples