The procedure followed in generating new synthetic composite samples

(n = 1,000) of C_ and C_ were based on the arithmetic means and standard
deviations listed in Table 4 and the stratum areas given in Table 3. No
logarithms were involved.

The synthetic means and standard deviations

thus obtained (n = 1,000) were G_

580.9 +

= 70.38 + 62.32 pCi/g and C

About 15 percent of the synthetic samples of ¢Sand approxi931.9 pCi/g.
mately 35 percent of the synthetic samples of C_ were negative values.

The expected means, based on the arithmetic means given in Table 4 for

six strata, the area data given in Table 3, and the procedure implied by

Equation 5 were Cy = 71 pCi/g and Ce = 579 pCi/g.

The calculator program, described in Appendix I, was modified to operate
on the synthetic means and standard deviations (x + s), discussed above,
and was used to generate synthetic samples of I
as indicated by Equation 1.
The results of this simulation, based on estimates of arithmetic
means and standard deviations, are shown in Table 10.

Table 10.

Synthetic Samples of Ib

(pCi/day Based on Equa

tion 1 and the Following Estimates of Arithmetic Means and

Standard Deviations:

I= 6,526 + 2, 396 g/day, C_

= 70.38 +

62.32 pCi/g, I, = 227 +"670 g/day C5 = 581 + 932 bCi/g

n

x

s

100
100
100
100
100
500

743,668
701,631
546,061
483,138
403,493
575,598

909,937
995,215
893,357
922,872
880,953
921,336

n

100
100
100
100
100
500
|1,000

x

594,596
754,221
487,977
521,907
524,160
576,572
576,085

s

|;

851,972
1,028,018
791,335
910,771
1,045,871
930,832
926,096

The overall estimate, I
= 576 nCi/day (n = 1000), is about 3.5 percent
higher than the overall @stimate shown in Table 9 (557 nCi/day, n = 500)
and approximately 2 percent higher than the estimate (565 nCt/day) based
on the rumen contents of fistulated steers.
The number of negative
values in the total synthetic sample (n = 1,000) was 217, a frequency of

21.7 percent. The expected frequency for a normal distribution based on
u = 576 and o = 926 is about 26.7 percent. The coefficients of variation

indicated by the estimates shown in Table 9, based on the lognormal
hypothesis, are generally smaller than those indicated by the estimates
shown in Table 10, which are based (except for I_) on assumed normal
distributions.
From these results and comparisons, it is obvious that
the simulation model does not depend on the lognormal hypothesis.

503

Select target paragraph3