Wpływ liczebności próby i metody zastępowania braków odpowiedzi na miary dopasowania oraz wyniki modelowania ścieżkowego

In the following article the authors describe the problem of influence of a sample size and a method of handling missing values on the results and goodness of fit of the path relation model. In order to estimate the goodness of fit of the model the authors use the indicators which describe the internal (Cronbach’s Alfa, Composite Reliability) and external (R2) stability of the model. By the term “results of the models” the authors mean estimated index values for latent variables and path coefficients of the SEM modeling procedure. In the research the authors analysed outcomes of Partial Least Squares method, used to build a model of Lublin shopping malls sector customers’ satisfaction and loyalty. The research included 43 datasets that varied in a number of observations and a method used for solving the missing values problem. Obtained results not only allowed the authors to statistically verify the main research problem of the study, but also enabled researchers to evaluate practical applicability of the analyzed imputation methods in real market and business consultancy activities. The research showed the supremacy of the Predictive Mean Matching and CART algorithms over other methods in the majority of analyzed ceases. Nevertheless, the differences between obtained results were rather insignificant, so one may assume that there is no visible influence of the used method on the practical interpretation of the obtained model and analyzed phenomenon.