Вы находитесь на странице: 1из 13

49.

Doubling Your Accuracy

An unbiased instrument for measuring distances makes random errors whose


distribution has standard deviation σ. You are allowed two measurements all told
to estimate the lengths of two cylindrical rods, one clearly longer than the other.
Can you do better than to take one measurement on each rod?

Assume that the random errors follow a normal distribution with mean 0 and
standard deviation 1. (1 hundreths of an inch, meter, cm, etc.)

This spreadsheet assumes separate measurements on rods A and B.


It uses NormInv(rand(),0,1) to simulate normal random numbers.

count A B
1 -1.5275 0.1151 -0.1008 = average of measurement errors on A.
2 0.1008 1.0522 1.2113 = variance of measurement errors on A.
3 -0.4304 0.3860
4 3.1356 0.0350 -0.0204 = average of measurement errors on B.
5 0.5137 0.9979 0.7964 = variance of measurement errors on B.
6 -0.2912 -0.2758
7 0.4186 -1.4506
8 -1.2173 0.1868 Note that variances are each about 1.
9 -1.1833 -0.3912
10 -1.5477 -0.3508
11 0.3790 0.0132
12 -0.1580 -0.3504
13 0.4652 0.8392
14 -0.9886 -1.3614
15 1.5648 0.1643
16 1.3362 -0.5698
17 0.5056 -0.8316
18 1.1901 -0.0751
19 -2.6131 1.8333
20 2.2560 -1.7786
21 0.9621 0.3953
22 -0.5131 0.4735
23 -0.6752 0.1083
24 -1.0748 -1.1938
25 0.2145 -0.0178
26 -0.6140 0.6496
27 1.4300 0.2544
28 0.5391 0.3980
29 0.7645 1.7937
30 -2.5806 -2.1700
31 -0.1439 -0.2476
32 -1.6289 0.3669
33 0.8014 0.2813
34 -1.2692 1.1405
35 0.5920 0.5757
36 -1.5867 -0.9384
37 -0.1380 -0.0655
38 -1.1349 -0.2856
39 1.9962 -1.7865
40 0.9319 -1.1004
41 -1.0325 0.5406
42 -0.7100 1.3688
43 -0.1482 1.4702
44 -1.2949 -1.5465
45 -2.2561 -0.6626
46 -0.2090 0.3681
47 0.5384 0.3761
48 -0.5510 -0.7016
49 0.9019 1.5923
50 -0.7840 1.4204
51 0.2053 1.0357
52 0.0446 -0.2278
53 0.5767 -0.0415
54 -0.9949 -0.5970
55 -0.6615 1.0323
56 0.1963 -0.4746
57 0.1729 -0.0555
58 0.8187 0.7586
59 -0.2325 -0.1934
60 -0.9505 1.3424
61 2.1627 -0.4051
62 -0.5453 0.7993
63 0.3709 -0.7757
64 -1.5762 -0.1440
65 0.5982 -0.5577
66 0.6189 -0.3700
67 0.7109 0.9651
68 -2.2737 0.0726
69 -1.8945 0.9364
70 -0.0430 0.3208
71 1.5247 -1.1651
72 -1.3482 -0.1177
73 -0.1127 0.0321
74 -0.3270 0.2580
75 -0.2380 -0.0176
76 1.2822 -0.2959
77 -0.5640 -0.5644
78 -1.1866 -0.0381
79 -0.0112 -0.0460
80 1.5359 0.8629
81 -0.5517 -0.9227
82 -0.5516 0.9068
83 -0.0059 0.6678
84 1.7219 0.4177
85 0.2225 -1.3812
86 1.0685 -0.7939
87 -1.0243 -0.9979
88 -1.1403 -1.0634
89 -0.2278 -1.0815
90 -0.5718 -0.0240
91 -0.3156 0.1764
92 -0.4973 -2.1425
93 -1.8543 0.6506
94 -0.9012 1.0660
95 0.7688 0.8153
96 0.0808 -0.5486
97 -0.0499 1.8455
98 0.8115 -0.4871
99 0.6121 -1.0815
100 1.2361 -1.4367
49. Doubling Your Accuracy

An unbiased instrument for measuring distances makes random errors whose


distribution has standard deviation σ. You are allowed two measurements all told
to estimate the lengths of two cylindrical rods, one clearly longer than the other.
Can you do better than to take one measurement on each rod?

Assume that the random errors follow a normal distribution with mean 0 and
standard deviation 1. (1 hundreths of an inch, meter, cm, etc.)

This spreadsheet assumes measurements on the sum and difference of the rods A+B
and A-B with random errors s and d for each. The errors in estimating
A and B are then (s+d)/2 and (s-d)/2 respectively.

Count s d (s+d)/2 (s-d)/2


1 -0.5999 -0.7490 -0.6744 -0.7117 0.0434 = average error in measuring A.
2 -0.2730 -1.4398 -0.8564 -1.1481 0.5453 = variance of error for A.
3 0.8044 -0.6324 0.0860 -0.2732
4 2.3358 -0.6533 0.8413 0.0940 0.0279 = average error in measuring B.
5 -0.6434 1.3105 0.3335 0.8220 0.5929 = variance of error for B.
6 2.9407 -0.0536 1.4436 0.6950
7 0.0864 0.4363 0.2613 0.3488
8 0.4438 1.7130 1.0784 1.3957 The variances are each about .5, half the variance with
9 1.4300 3.0373 2.2337 2.6355
10 0.4631 0.4256 0.4444 0.4350
11 -0.1634 -0.4468 -0.3051 -0.3760
12 0.5343 -0.3384 0.0980 -0.1202
13 -0.5830 -0.2811 -0.4320 -0.3566
14 2.9012 0.0217 1.4614 0.7415
15 0.4924 -0.3624 0.0650 -0.1487
16 -0.6000 -1.5582 -1.0791 -1.3187
17 -0.6409 -1.1369 -0.8889 -1.0129
18 1.2891 -0.5204 0.3843 -0.0680
19 -1.4959 -0.8455 -1.1707 -1.0081
20 0.3990 1.0232 0.7111 0.8672
21 -0.9115 -0.3416 -0.6266 -0.4841
22 2.9225 -0.1956 1.3634 0.5839
23 0.0297 -0.1842 -0.0772 -0.1307
24 1.8413 0.2013 1.0213 0.6113
25 2.8246 0.6536 1.7391 1.1963
26 0.1659 -0.9270 -0.3805 -0.6538
27 0.1594 -1.0810 -0.4608 -0.7709
28 1.1497 0.7345 0.9421 0.8383
29 -1.3411 0.0719 -0.6346 -0.2813
30 0.7465 -0.7272 0.0097 -0.3588
31 -0.5695 -1.7627 -1.1661 -1.4644
32 -0.0710 -0.6620 -0.3665 -0.5143
33 0.0088 0.1789 0.0939 0.1364
34 1.2675 0.3940 0.8307 0.6124
35 0.8692 1.2818 1.0755 1.1786
36 0.9511 -0.4122 0.2694 -0.0714
37 -0.5383 0.5164 -0.0110 0.2527
38 -0.9700 -0.6043 -0.7872 -0.6958
39 -0.2689 0.0362 -0.1163 -0.0401
40 0.7263 0.4853 0.6058 0.5456
41 0.1285 1.0905 0.6095 0.8500
42 0.2808 -0.4755 -0.0973 -0.2864
43 0.9343 0.4650 0.6997 0.5823
44 0.9284 0.4851 0.7068 0.5960
45 -1.1905 0.2621 -0.4642 -0.1010
46 0.4147 -0.5094 -0.0474 -0.2784
47 -0.6309 -0.3444 -0.4876 -0.4160
48 1.2157 -0.1766 0.5196 0.1715
49 -0.1295 -0.9072 -0.5183 -0.7128
50 1.2634 0.6014 0.9324 0.7669
51 -1.0663 -1.1850 -1.1257 -1.1553
52 0.8816 -1.2200 -0.1692 -0.6946
53 -1.0256 -1.4450 -1.2353 -1.3401
54 -0.5669 0.8529 0.1430 0.4979
55 1.1693 0.7942 0.9817 0.8880
56 -0.4005 0.2509 -0.0748 0.0881
57 0.4625 0.8887 0.6756 0.7821
58 -1.4042 0.1561 -0.6240 -0.2339
59 -1.7108 0.6073 -0.5518 0.0278
60 -0.5423 -0.4487 -0.4955 -0.4721
61 1.1298 -0.6890 0.2204 -0.2343
62 -0.5993 -1.4209 -1.0101 -1.2155
63 -0.2104 -1.9474 -1.0789 -1.5132
64 -0.8169 2.1715 0.6773 1.4244
65 -1.1402 0.2326 -0.4538 -0.1106
66 0.2349 0.3935 0.3142 0.3539
67 0.8619 0.9471 0.9045 0.9258
68 0.0870 0.2508 0.1689 0.2098
69 -0.1991 -0.1451 -0.1721 -0.1586
70 -1.2358 -1.1788 -1.2073 -1.1931
71 1.0808 0.4507 0.7657 0.6082
72 0.0940 2.0459 1.0700 1.5579
73 -0.3723 0.2449 -0.0637 0.0906
74 -0.1745 -1.2435 -0.7090 -0.9763
75 -0.0020 0.1998 0.0989 0.1494
76 -0.7313 -1.2634 -0.9973 -1.1304
77 -0.5771 -0.4915 -0.5343 -0.5129
78 1.2972 -0.4239 0.4366 0.0064
79 0.5943 2.4795 1.5369 2.0082
80 -0.8267 0.8052 -0.0108 0.3972
81 -1.0215 -0.3647 -0.6931 -0.5289
82 -0.8542 -0.0360 -0.4451 -0.2406
83 -0.8276 0.9714 0.0719 0.5217
84 -2.3887 1.5561 -0.4163 0.5699
85 0.7954 0.4156 0.6055 0.5106
86 -1.0271 1.8048 0.3889 1.0968
87 -1.7930 0.1520 -0.8205 -0.3342
88 0.5224 0.5051 0.5137 0.5094
89 0.9957 0.6531 0.8244 0.7388
90 -1.3288 -0.1985 -0.7636 -0.4811
91 -0.6143 0.1265 -0.2439 -0.0587
92 -0.8304 -0.1358 -0.4831 -0.3095
93 -1.0744 -0.9175 -0.9960 -0.9568
94 0.0328 0.0867 0.0598 0.0732
95 0.0688 0.0170 0.0429 0.0299
96 -0.0477 0.5638 0.2580 0.4109
97 -0.1454 0.3924 0.1235 0.2579
98 1.2229 -1.3544 -0.0658 -0.7101
99 -0.0540 -0.2921 -0.1730 -0.2326
100 0.2045 -0.4792 -0.1373 -0.3082
.5, half the variance with separate measurements.
49. Doubling Your Accuracy

An unbiased instrument for measuring distances makes random errors whose


distribution has standard deviation σ. You are allowed two measurements all told
to estimate the lengths of two cylindrical rods, one clearly longer than the other.
Can you do better than to take one measurement on each rod?

Assume that the random errors follow a normal distribution with mean 0 and D= A− B+ d
standard deviation 1. (1 hundreths of an inch, meter, cm, etc.)
S= A +B +s +e
This spreadsheet assumes measurements on the sum and difference of the rods A+B 1
and A-B with random errors s and d for each, and random error e that comes from
( S+ D )= A+
2
aligning A and B when measuring the sum of the rods. Now the errors in estimating A and B
are (s+d+e)/2 and (s-d+e)/2 respectively. 1
( S− D )= B+
2
Count s d e (s+d+e)/2 (s-d+e)/2
1 0.5925 1.6990 -0.5458 0.8729 -0.8262 -0.1163 = average error for A.
2 1.6989 0.4707 0.2612 1.2154 0.7446 0.7424 = variance for A measurements.
3 -1.0499 1.2898 0.6704 0.4552 -0.8346
4 0.0153 -2.4481 0.2882 -1.0723 1.3758 -0.1494 = average error for B.
5 0.1941 -1.0946 0.4765 -0.2120 0.8825 0.6919 = variance for B measurements.
6 -1.1302 2.7376 -1.5791 0.0142 -2.7234
7 -0.0071 1.3277 -1.7336 -0.2065 -1.5342
8 0.9114 -0.1365 0.6095 0.6922 0.8287 The variances are somewhat larger now, ab
9 0.6489 1.0219 0.3261 0.9984 -0.0234 the variance with two separate measuremen
10 -0.8082 0.4744 0.3156 -0.0091 -0.4835
11 0.9915 0.1249 0.5929 0.8546 0.7297
12 -2.0485 -0.3195 -1.3364 -1.8522 -1.5327
13 1.4826 -0.1576 -0.5007 0.4121 0.5697
14 -0.1982 -0.9895 0.7575 -0.2151 0.7744
15 -0.3139 -1.5915 -1.9251 -1.9152 -0.3237
16 -0.3288 0.8268 1.4558 0.9769 0.1501
17 -0.7439 -0.9091 -1.1074 -1.3802 -0.4711
18 1.5681 -0.3197 -0.7545 0.2469 0.5666
19 -0.1016 0.4546 -1.0123 -0.3297 -0.7843
20 0.1118 -0.7936 0.2203 -0.2308 0.5629
21 -0.4178 -2.4311 -0.3861 -1.6175 0.8136
22 0.1405 -0.1275 1.0219 0.5175 0.6450
23 0.8774 -0.1209 1.5179 1.1372 1.2581
24 -0.0038 1.2769 -0.0174 0.6279 -0.6490
25 -0.0866 0.0590 -1.1691 -0.5984 -0.6573
26 1.2435 0.0817 1.3699 1.3476 1.2658
27 -0.5818 1.7734 -0.3974 0.3971 -1.3763
28 0.8883 0.6581 0.4475 0.9970 0.3388
29 0.4333 -0.1801 -0.2439 0.0046 0.1847
30 0.9166 -1.5020 -0.7783 -0.6818 0.8202
31 -2.5636 0.1834 -2.0976 -2.2389 -2.4223
32 0.1473 0.3182 0.6030 0.5342 0.2161
33 1.1888 -0.9330 -0.5146 -0.1294 0.8036
34 -1.7184 -0.6415 0.7307 -0.8146 -0.1731
35 0.1199 -0.0112 -0.4578 -0.1745 -0.1634
36 -0.2042 -0.5670 -0.6799 -0.7256 -0.1586
37 1.4688 0.7757 -0.3099 0.9673 0.1916
38 -0.3134 1.1476 0.9572 0.8957 -0.2519
39 -0.7069 -0.2937 -1.5056 -1.2531 -0.9594
40 1.2413 1.0311 -1.0175 0.6275 -0.4036
41 0.5724 0.0486 0.0445 0.3327 0.2841
42 -1.0126 0.2487 -1.3707 -1.0673 -1.3160
43 -0.9498 -0.2992 0.3954 -0.4268 -0.1276
44 -2.1689 -0.4727 0.0848 -1.2783 -0.8057
45 -0.2371 -0.3409 -0.6640 -0.6210 -0.2801
46 -1.9082 0.7627 -1.1497 -1.1476 -1.9103
47 0.5641 -0.8873 -1.8377 -1.0805 -0.1931
48 -1.7859 1.3674 0.6573 0.1194 -1.2480
49 -0.1772 -0.1688 1.3755 0.5148 0.6836
50 0.5793 0.5134 0.2739 0.6833 0.1699
51 0.9022 -0.8723 -1.2526 -0.6113 0.2610
52 -1.0438 -1.2635 0.9948 -0.6562 0.6072
53 0.3548 0.0568 -1.0606 -0.3245 -0.3813
54 0.9570 1.2382 -0.1657 1.0147 -0.2234
55 0.3699 1.0513 0.9033 1.1622 0.1110
56 -0.7438 1.1024 1.2985 0.8286 -0.2738
57 -0.8707 0.3056 0.6192 0.0271 -0.2785
58 1.3319 0.4963 -0.7074 0.5604 0.0641
59 -1.5551 -1.7028 0.3688 -1.4446 0.2583
60 -0.5077 -0.3966 0.2334 -0.3354 0.0611
61 -0.4678 -0.2062 -0.5367 -0.6054 -0.3992
62 -1.1816 -2.4927 -0.2324 -1.9534 0.5394
63 0.1790 1.3967 -0.3602 0.6078 -0.7890
64 0.0146 -0.4179 -0.0313 -0.2173 0.2006
65 0.9971 0.1970 -1.6834 -0.2447 -0.4416
66 -0.4843 -2.1856 -0.0458 -1.3579 0.8277
67 -0.0434 -0.0979 -1.8111 -0.9762 -0.8783
68 -1.1192 -0.6523 -0.6042 -1.1878 -0.5356
69 -0.1098 1.0593 -0.6073 0.1711 -0.8882
70 -0.8153 2.4748 -0.2693 0.6951 -1.7797
71 0.1806 -0.0372 1.0665 0.6050 0.6422
72 1.0580 -1.0211 -0.0686 -0.0159 1.0052
73 -0.7661 0.5513 -0.6445 -0.4297 -0.9809
74 0.0686 1.8488 0.2790 1.0982 -0.7506
75 -1.7895 0.4520 -1.1879 -1.2627 -1.7147
76 -0.0879 -0.3613 -1.1212 -0.7852 -0.4239
77 0.6255 -1.2599 -0.5863 -0.6103 0.6496
78 0.8668 -0.1758 -0.9559 -0.1325 0.0434
79 -0.3799 -0.5664 -1.3781 -1.1622 -0.5958
80 -1.0465 -1.6421 0.7336 -0.9775 0.6647
81 -0.3858 0.5237 0.2367 0.1873 -0.3364
82 -0.4210 0.2933 -1.6932 -0.9105 -1.2038
83 0.4192 0.2930 -1.0096 -0.1487 -0.4417
84 0.4482 0.4965 -0.6722 0.1362 -0.3603
85 -1.8668 -0.0826 1.5286 -0.2104 -0.1278
86 -0.6198 1.2253 -0.7351 -0.0648 -1.2901
87 0.8178 0.0814 0.3186 0.6089 0.5275
88 1.2187 0.4548 -0.3315 0.6710 0.2162
89 1.0242 0.9040 0.9600 1.4441 0.5401
90 -1.0328 0.5459 2.1515 0.8323 0.2864
91 0.1820 -0.4186 0.9045 0.3340 0.7526
92 -1.5702 0.5856 0.7621 -0.1112 -0.6969
93 0.3238 -0.1994 -0.9046 -0.3901 -0.1907
94 0.8398 0.0360 -1.2308 -0.1775 -0.2135
95 -1.1697 0.2854 -0.7852 -0.8347 -1.1201
96 -0.2803 -0.6186 -1.8082 -1.3535 -0.7349
97 -1.2066 -0.6248 -0.0235 -0.9275 -0.3027
98 0.3979 -1.0052 1.6746 0.5336 1.5388
99 1.1717 0.0004 0.9818 1.0770 1.0766
100 0.7528 0.7213 0.5831 1.0286 0.3073
D= A−B+ d
S= A +B +s +e
1 1
( S+ D )= A+ ( s+ d +e )
2 2
1 1
( S− D )= B+ ( s− d+ e )
2 2
e error for A.
e for A measurements.

e error for B.
e for B measurements.

omewhat larger now, about .75, but still less that


wo separate measurements.
A Note on normal random numbers.

The web site: http://www.exceluser.com/explore/statsnormal.htm talks about the Box-Muller approach to


to generate normal random numbers, and states:

The Box-Muller approach suggests that Excel users should use this formula to calculate a random number from a normal distr

=SQRT(-2*LN(RAND()))*SIN(2*PI()*RAND())

See also http://mathworld.wolfram.com/Box-MullerTransformation.html


Muller approach to

m number from a normal distribution:

Вам также может понравиться