Table iii also shows that the solvent parameters
for these three structures are remarkably similar. This similarity is consistent
with the notion that ``water is water''. It seems reasonable that solvent
will be excluded in a similar fashion from all protein molecules, and because 
describes the transition from solvent to protein it should be similar from
one structure to another.
However not all mother liquors have the same electron density. 
should vary from structure to structure. The structures in Table
iii are listed in order of increasing solvent electron density. Thermolysin
crystals are stored in water with 7% DMSO and little else. T4 lysozyme
crystals are immersed in 2M phosphate buffer. Although mass density is
not strictly proportional to electron density the electron density of T4
lysozyme mother liquor should be higher than that of thermolysin.
Because the structures were refined without the low-resolution data
and without solvent models, there is a correction to the average temperature
factor in both sets of scale factors for each structure. In each case the
average temperature factor increases when the solvent model is added. An
additional trend can be observed. The increase for the thermolysin inhibitor
which was refined to 1.6Å resolution was 2.8Å 
while the increase for the bacteriochlorophyll containing protein (refined
to 1.9Å) was 6.0Å
. The lower the resolution of the data set, the more important is the contribution
of the lowest resolution data.
It is true that the R value calculated with both low-resolution data and a solvent model will be higher than the R value for the same coordinate set calculated with the low-resolution data eliminated. As seen in Table iii the R value for the thermolysin:inhibitor complex using the 20 to 1.6Å data is 17.5%. When the inner resolution limit is set to 5.5Å the R value drops to 17.1%. Some investigators may consider the lower R value justification for rejecting these data. This is incorrect reasoning. In a statistical analysis, the model can never be used to assess the data. It is clear that the lower R value does not represent a better model -- The same model is used in both cases. The lower R value is simply an artifact of the method used in its calculation.