Epilepsy is a chronic disorder of the brain that affects 1% of world population. The occurrence of epileptiform discharges (ED) in electroencephalographic (EEG) recordings of patients with epilepsy signifies a change in brain dynamics and particularly brain connectivity. In the last decade, many linear and nonlinear measures have been developed for the analysis of EEG recordings to detect the direct causal effects between brain regions. In many cases the number of EEG channels (the time series variables) is large and the analysis is based on short time intervals, resulting in unstable estimation of vector autoregressive models (VAR models) and subsequently unreliable Granger causality measure. For this, restricted VAR models have been proposed and in our recent study it was found that optimal restriction of VAR for the estimation of Granger causality was obtained by the backward-in-time selection method (BTS). We use the concept of restricted VAR models in measures both in time and frequency domain, namely restricted conditional Granger causality and restricted generalized partial directed coherence. We test the two measures in their ability of detecting changes in brain connectivity during an epileptiform discharge from multi-channel scalp electroencephalograms (EEG).