The roan allele is partially dominant--it gets expressed if there's one copy, but not completely. It seems like the amount of expression of the roan allele depends on a lot of factors, including the specific problem with the allele, other genes, uterine environment, random chance, and who knows what else.
The way the roan gene works, it controls when and/or how many pigment cells (which are like nerve cells) get sent out from the spinal area of the embryo to spread across the skin. With two normal alleles, everything works well and on time and pigment cells can spread across the entire embryo. With one normal and one broken (roan) allele, the pigment cells leave late and/or in smaller numbers than they should, which doesn't let them cover the entire body. However, how much the cells are affected seems to vary: some roans are nicely roaned all over their bodies, some are mostly white, and some are mostly colored. (The head and butt are normally well-pigmented because they're near the ends of the tube that becomes the spine, and the pigment cells come out of the ends.)
With two roan alleles, the pigment cells never get to leave the tube. That's why lethals have no pigment. Pigmentless ears are very likely to have the nerves die off, causing deafness. The roan alleles seem to screw up the front end of the tube, also interfering with nerve development in the back of the eye, causing blindness. Somehow they affect tooth development too. Different lethals have different problems, although all are white and blind. The different problems are probably because of the same factors as the varying amounts of roaning--other genes, specific problems with particular roan alleles (there may well be more than one, as in all species I know of with problems in this gene, there's at least two deformed variants).