Skip to main content
. 2004 Sep 14;1:8. doi: 10.1186/1742-4682-1-8

Table 4.

The effect of the size of the perturbation

Computed 5 % 10 % 50 % 200 %
a10 0 0.0000 0.0000 0.0001 0.0008
a11 -14.6780 -14.3647 -14.1817 -13.1496 -11.3439
a12 0 -0.1466 -0.1429 -0.0671 0.5735
a13 7.3390 7.3414 7.3438 7.3598 7.3735
a14 0 -0.2165 -0.3673 -1.2462 -2.7619
a15 -7.3390 -7.1723 -7.0780 -6.4846 -5.2501
a20 0 0.0000 0.0000 0.0000 -0.0003
a21 14.6780 14.6119 14.5748 14.4207 14.5029
a22 -14.6780 -14.6540 -14.6623 -14.7503 -15.1862
a23 0 -0.0009 -0.0016 -0.0054 -0.0070
a24 0 0.0494 0.0839 0.2494 0.3462
a25 0 -0.0309 -0.0464 -0.1119 -0.0951
a30 0 0.0000 0.0000 0.0004 0.0038
a31 0 -2.3527 -4.5412 -18.2307 -46.8953
a32 0 1.3989 2.6336 9.8422 24.4004
a33 -27.2517 -27.9204 -28.5955 -34.0204 -54.4047
a34 0 1.7491 3.4009 14.0961 39.3252
a35 0 -0.9955 -1.8949 -7.0627 -15.4759
a40 0 0.0000 0.0000 -0.0001 0.0001
a41 0 2.0843 3.7814 14.7316 41.5863
a42 0 -1.0925 -1.7693 -5.5766 -13.2688
a43 18.5664 19.0295 19.4964 23.2397 37.1866
a44 -18.5664 -20.2112 -21.6608 -31.4631 -58.1065
a45 -9.2832 -8.3594 -7.6404 -3.2226 6.5808
a50 0 0.0000 0.0000 -0.0001 -0.0015
a51 0 -0.4026 -0.6581 -2.5848 -10.1097
a52 0 0.1384 0.0830 -0.1317 0.1582
a53 0 -0.0059 -0.0110 -0.0435 -0.0879
a54 18.5664 18.8987 19.1602 21.0620 27.2722
a55 -18.5664 -18.7852 -18.9201 -20.0013 -24.0836

Overall, the estimated coefficients deviate more strongly from the corresponding computed values as the perturbation increases. However, there are substantial differences between variables. The coefficients associated with variable X2, for example, are hardly influenced, while the coefficients associated with X3 are strongly affected. Overall, the method seems to produce the best results for perturbation up to 10%. The datasets for the regression consisted of 401 data points in the interval [0,4] and the method of linearization was option II.