User defined data scaling

There is yet one other form of scaling that I recommend.  Users can define their input data in a fashion that scales the model data and the answer.  In particular, one should try to develop units for the input data so that the largest value expected for decision variables and shadow prices is under a million and in the thousands if possible.

Carefully consider the units for your data.  For example, in US agriculture about 325 million acres are cropped and there is a 9-10 billion bushel corn crop.  Thus in setting up production data I could enter land in 1000's of acres and all other resources in 1000's of units.  I might also treat the corn crop in millions of bushels.  The data will be simultaneously scaled so with resource endowments in 1000's then corn yields are divided by 1000.  The net affect then is a corn production variable in the units of millions.  Consumption statistics would need to be scaled accordingly.  Money units can also be in say millions or billions of dollars.

Such data scaling generally greatly reduces the disparity of coefficients in the model.