maximum yield per hectare of rice. Solving the equation
of the response surface for 270 kg of N shows that maximum
expected rice production would be 6995 kg/ha. This point
can be plotted on the curve.
 Economically, a more important value would be the
amount of nitrogen which would maximize profit for the
farmer with abundant resources. Once again, referring to
a previous discussion, this value can be found by equating
the derivative of the response surface to the ratio: price
of nitrogen/price of rice. Remember that the prices must
be in the same units of measurement as the equation. At
the time of the study, IR-22 rice had a value of $2.80/kg
and the price of nitrogen as urea was $4.40/kg, both
expressed in Colombian pesos. Therefore, the following
relationship can be found:

 dY/dN = 12.25 0.0455 N = 4.40 / 2.80,
 from which N = 235

In other words, 235 kg/ha of N will maximize profit for
farmers with the prices as stated. Once again, solving
the equation of the response surface for N = 235, the
value of Y = 6970 can be found. Although it is possible
to produce 6995 kg/ha of rice with 270 kg of N, it is not
profitable to apply more than 235 kg/ha, because the cost
of the additional 35 kg/ha of N is greater than the value
of the additional 25 kg of rice.
 It is possible to use the visiographic method to
calculate response surfaces for more than three levels of
one input, but the results are less precise. It is
necessary to draw a curve through the data points visually
in a form that approximates a quadratic curve and appears
to be representative of the data. Once the shape of the
curve satisfies the researcher, three points along that
curve can be chosen and the same calculations made as
above. When the equation of the estimated response
surface is found, if it does not satisfy the researcher,
it is possible to begin again with a new curve drawn
visually. Obviously, this method is subject to human
error and the bias of the researcher. However, with
practice it is possible to achieve acceptable results (see
Fox, 1968, p.102.) even though no estimates of sampling
error will exist.