possible to make this kind of conclusion.
 Then what can be concluded? Statistical calculations
have resulted in a curve, but with tests of significance
which lead one to reject the curve. Even though the
statistical tests are not significant at high levels such
as 1%, 5%, or even 10%, if one were to conclude that the
curve is more significant than the straight line, it would
be a correct conclusion in more than 80% of the cases. In
this specific case, the calculated value of Fc indicates
that the quadratic term is significant at a level between
15% and 20%.
 To make a conclusion based on the data presented and
with theoretical logic and experience, the best evidence
available indicates that the response of IR-22 rice to
nitrogen with 40 kg/ha of P205 and 40 kg/ha of K20 in
Corinto, Cauca, is curvilinear between 0 and 300 kg/ha of
N. Furthermore, it appears that production reaches a
maximum near 270 kg/ha of N. Using a curvilinear equation
it is possible to calculate the profitability of capital
invested in fertilizer and to determine optimum doses for
different combinations of prices for fertilizer and rice.
If, on the other hand, one concludes that the linear
response is the correct response, these kinds of
interpretations cannot be made. The only conclusion that
can be reached from the linear equation is that it pays to
apply 100 or 150 kg/ha of N, but it also pays to apply 200
and 300 kg and even more. This type of conclusion is not
very reasonable.
 It can be seen that theory, logic, experience, and
other a priori and a posteriori information provide much
of the basis for analysis and interpretation of data, and
that statistical analysis is not sufficient by itself. In
summary, in addition to statistical analysis, the
researcher must count on: 1) a priori information such as
theory and experience; and 2) a posteriori information
such as the physical and economic feasibility of the
results.


MULTIPLE REGRESSION--VISIOGRAPHIC

 A visiographic method can also be used to estimate a
response surface from a two-input experiment, if several
levels of each input are included. However, the method is