Níveis de Lisina Para Suínos de 21 a 42 dias

Anderson e Nelson, 1975

[pic 1]

[pic 2]

Bullock e Bullock, 1994

O modelo quadrático com platô é preferível que o modelo quadrático em todos os casos para predizer a dose ótima de fertilização nitrogenada para o milho.

Cerrato e Blackmer,

Data in Table 3 indicate that, when evaluated by using the R2 statistic, the five models seem to fit the yield data aboul equally well. Because there is little biological basis for selecting one model over other (Mead and Pike, 1975; Nelson et al., 1985), the R2 statistic usually is usled to justify the use of a particular model. The limitatio'n of using the R2 statistic to select a model is further illustrated in Fig. 1, which shows how each of the models fits the data from Site 5 in 1986. Because there is only one correct economic optimum rate of fertilization for each site-year, the observation that models having similar R2 values differ greatly in determined values of economic optimum rates indicates that R2 value is not a reliable criterion for selection of a model for identification of economic optimum rates of N fertilization. The numbers in parentheses in Table 3 illustrate that use of the R2 statistic can result in a false sense of confidence concerning the ability of models to describe responses to N when too few treatments are used. That is, greater R2 values were obtained when only four N treatments were considered than when all 10 treatments were considered. Although these differences can be reduced by adjusting the R2 values for degrees of freedom (Darlington, 1968; Judge et al., 1982; Blackmer and Meisingel, 1990), this adjustment usually has not been made in fertilizer response studies. We believe this is a notable problem because most fertilizer N response studies involve four or fewer N treatments and because the ability of models having high R2 values (e.g., 0.90 or higher) to identify optimum rates of fertilization has not been questioned.

BAKER, 1986

The point to be made is that even with similar samplings of a population and similar dietary controls, two different investigators can arrive at requirement values vastly different simply because they choose to use different methods to predict the require ment. Thus, in many cases authors of requirement papers report that their estimate is considerably different from a previous estimate, when in fact, with similar mathematical approaches the two are, in reality, very similar. Using the data set in table 2 as an example (and most would agree this data set is easier to interpret than most), one could select requirement values of 0.285, 0.316, 0.341 or 0.362%. This represents a variation of 27% from the highest to the lowest estimate. Certainly, if the nutrition community could come to some agreement concerning the definition of "requirement" (e.g., 90 or 95% of maximum response in a curvilinear fit), the literature on requirements would be much easier to interpret.

...