A selection of remarks on the measurement of correlations between variables of a panel data structure

Many articles featuring panel data modelling tend to begin their considerations with an introduction of the Pearson linear correlation coefficients matrix between the analysed variables. The aim of the article is to prove such an approach unsuitable in the analysis of panel data dependencies. Instead, an attempt has been made to propose a more appropriate measure – a correlation coefficient between the empirical and fitted values of the dependent variable of the estimated panel model (with fixed or random effects) in relation to the variable whose dependency towards the dependent variable is being studied. Pearson’s linear correlation coefficient does not reflect the basic advantage of panel data, which is the ability to provide information about the dependencies of the studied phenomena simultaneously in time and space. The fact that one observation relates to object i during period t and another to object j during period t + 1 is irrelevant for the calculation of the coefficient. Pearson’s coefficient, however, can be used when conducting sub-calculations in panel data analysis. The presented considerations have been illustrated by the calculations of the relationships between the structure of capital and the profitability and size of 17 construction companies listed on the Warsaw Stock Exchange in the years 2009–2018 (170 observations) which created a balanced panel. A specification of the advantages and disadvantages of the proposed solution was formulated on the basis of the calculations.