mm43, GPS is 100 meters accuracy "guaranteed" for the C/A channel. Generally you can do much better if your dynamics are not too large. If every transmission sends say 500 meters resolution position data how closely could they plot the plane's impact point, or at least the last point at which it is known to have missed a transmission?
If you send difference from last full report then you have pretty decent resolution within two BYTEs. A third BYTE with altitude might be nice for reconstructing the blow by blow of the accident. But it's probably not as big as win as the N/S and E/W BYTES.
You could get better position resolution with fancier coding - delta from straight line projection of the last two points, for example. But there is a point at which it all gets silly. A 0.8 km diameter circle is a whole lot better for searching than any 900 square miles search area. And that should be possible using only the printable ASCII character set.
(And it appears I must have left "full" out of "delta from last full position report." I figured they were the position reports. The deltas were delta reports. And as pointed out deltas on deltas isn't good. That was so obvious to me I forgot it would not be to many other people, I guess.)
{^_^}