Actually this should be called  as the root mean square deviation,  otherwise it doesn't make very much  sense from the statistic point of view.  It should be also supplemented with the mode - the most probable value of surface height undulations- and the mean- the arithmetic average value of surface height undulations- in order to get some physical feeling about the surface roughness, especially for the random surface roughness-es.  In the case of Gaussian behavior to know the variance (square of RMS) and the mean would be quite sufficient to get the full information about the surface roughness. 

There are with the conventional methodology for determining the roughness, wherein the EDM some problems that must be considered. The main problem isthe that the crater structures and the recrystallized layers result in a surface structure that does not meet the conditions of the conventional machining operations, such as grinding or milling. So if roughness measurements are performed, then the relationship between Ra, Rt and RMS are not comparable to conventional machining. When ED machining of hard metals (WC-Co), it may happen that various measurements on the same working surface leads to very different results.

Under certain circumstances, the slightest mistake is made when RMS value. But I think that in comparative studies and the Ra value may be used.

It is important only in comparisons to maintain the same measuring direction to the machining direction.