Modern video encoding systems employ block-based,multi-mode, spatio-temporal prediction methods in order to achieve high compression efficiency. A common practice is to transform, quantize and encode the difference between the prediction and the original along with the system parameters. Obviously, it's crucial to design better prediction and residual encoding methods to obtain higher compression gains. In this work, we examine two such systems which utilize subsampled representations of the sequence and residual data. In the first system, we consider a method for reorganizing, downsampling and interpolating the residual data. In the second system, we propose a new method that employs lower resolution intensity values for spatial and motion-compensated prediction. Both of these methods are macroblock adaptive in the rate-distortion sense. Our experiments show that implementing these methods brings additional compression efficiency compared to the state-of-the-art video encoding standard H.264/AVC.