Recent studies suggest that the reported tornado counts prior to 1990s do not fully reflect actual occurrence of tornado in the U.S. The short-lived nature of tornados and the limitation of tornado database, which primarily relies on human observations has made it complicated to do climatology analysis using the historical records. This study presents a systematic analysis to examine the potential non-meteorological factors influencing tornado counts during 1950-2018. We first explore the governing spatio-temporal patterns in tornado counts across the US using a Principal Component Analysis and investigate possible linkage between the dominant modes of tornado variability and large-scale climate oscillations as well as population shifts. Statistically significant time trend was observed in the first principal component (captured 62% of variance), but none of the climate indices were found to be correlated with it. In the next level, a statistical model using a hierarchical Bayesian framework is constructed to evaluate characteristics of tornado variability across US states and quantify its relation with influencing regional characteristics. The model accounts for changes in observational practices and estimates the counts simultaneously. Results indicate that there is significant association (at 90 percent confidence interval) between reported tornado and population shift for almost all the states in tornado-prone regions. The significant relationship with climate indices like El Niño- Southern Oscillation and the North Atlantic Oscillation demonstrates considerable spatial variability across the country showing negative and positive connections contingent on the region (e.g. during La Niño, tornados in southeastern US increases while it decreases in Northern areas). Results also show significant association between increasing tornado and negative phase of PDO in southeast areas except for Florida.