26.9 FITTING REGRESSION EQUATIONS 693
‘The best estimates of a and 8 are sought. Thus to minimize
Do-s =Ti- @ + bP. (26.55)
whete y, are the observed values and §, are the estimated values from Eq. 26.54, take
partial derivatives as follows:
2{3 vi @ + emir} (26:56)
af bi- @+ ax? (26.57)
After carrying out the differentiations and summations, two equations result in @ and
B, called normal equations.
Dy-na- BD x,
Day -oLn-pBst
Solving Eqs. 26.58 and 26.59 simultaneously yields
0 (26.58)
0 (26.59)
_Zy_ BEm oes
a= = ¥- Bx (26.60)
_ Exy— Bu yisn
6 eae (26.61)
Recall the slope is p(a,/a,), or as estimated from sample data
port (26.62)
Se
Also, the unexplained variance in the regression equation is
a= a1 — p?) (26.63)
the square root of which is the standard deviation of residuals (see Fig. 26.8) and is
called the standard error of estimate, These can be estimated from
pe leg — (26.64)
n-2
LZ (26.65)
or
where y, and §, are as defined previously (see Eq. 26.55).
‘Many hydrologic variables are linearly related, and after estimating the regres-
sion coefficients, prediction of y can be made for any value of x within the range of
observed x values. Extrapolation outside the range is often performed but should be