The -E means we run the convolution centred on all those nodes in the first and last section corresponding to the half filter width. Thus, it is unbalanced since there are no data before the first one and no data after the last point. So in a sense the filtering gets narrower for the output. Use with caution since nothing can replace knowing the actual values. so we only use -E when it is convenient to have the output range match the input and we are not that concerned about the filtered output being a bit shabby at beginning and end.