Orthogonal Distance Regression with Constraints

Share icon

Wed, 07/14/2010 - 02:24 pm

The documentation on CurveFit gives the impression that it is possible to set constraints when using Orthogonal Distance Regression (ODR). Can anyone tell me how I can adapt the CurveFit code below to ensure that the 'a' coefficient of the regression line is held at the value of zero.

Make/N=1000 Sales, Opmargin, Operprofits
Sales = 1000000 + gnoise(50000)
Opmargin = 0.40 + gnoise(0.01)
Operprofits = Sales * Opmargin
CurveFit /ODR=2 line Operprofits /X=Sales
  Fit converged properly
  y= W_coef[0]+W_coef[1]*x
  W_coef={-12314,0.41217}
  V_chisq= 9.53783e+10;V_npnts= 1000;V_numNaNs= 0;V_numINFs= 0;
  V_startRow= 0;V_endRow= 999;
  W_sigma={6.82e+03,0.0068}
  Coefficient values ± one standard deviation
    a   =-12314 ± 6.82e+03
    b   =0.41217 ± 0.0068


Many thanks.

Eduard Kas
This works very well. See the code below.

k0=0
CurveFit /ODR=2/H="10" line Operprofits /X=Sales /D
  Fit converged properly
  fit_Operprofits= W_coef[0]+W_coef[1]*x
  W_coef={0,0.3999}
  V_chisq= 9.56824e+10;V_npnts= 1000;V_numNaNs= 0;V_numINFs= 0;
  V_startRow= 0;V_endRow= 999;
  W_sigma={0,0.000333}
  Coefficient values ± one standard deviation
    a   =0 ± 0
    b   =0.3999 ± 0.000333


I had figured out the /H="10" flag, but still wrongly assumed that I had to put the k0=0 constraint in the same line of code.

Eduard

July 14, 2010 at 10:31 pm - Permalink