where wu and of are the mean and variance of y = In x, respectively.
Finney %1941) derived a minimum variance unbiased estimator (MVUF) for

EX because of the large bias of the maximum likelihood estimator (MLE).

Finney's estimator is

E(x) = exp(¥) 8, (8/2)
where y is the arithmetic mean of the log transformed x values, s? is
the variance of the log transformed x values, and the function g is the
infinite series

g(t) =l1+

(n-1)t + (n-1) >t?

n

+

no(nt1)2!

(n-1)°t?

n> (ntl) (n43)3!

+...

A finite sample confidence interval can be estimated for E(x) by Cox's
directmethod (Land, 1972).

A confidence interval is calculated for the

value y + 1/2 s* and then antiloged to achieve an interval on E(x).

This interval ig asymmetric, but has the desirable property that the
lower confidence bound cannot be negative.
Cox’s direct method was
shown to be easily the best of the approximate methods considered by
Land and was recommended by him when dealing with large sample sizes and
moderate values of s*,
Land's exact method is not used because of the

computational diffictlties involved.

Estimation of the EX for the gamma distribution is much simpler than for
the lognormal distribution.
The EX of the gamma distribution is the
product of the parameters a and 8, where the probability density function
is

f (x) =

x

T (a) 8”

exp (-x/B) xo dx

(x > 0; a > 0,

The maximum likelihood equations to estimate ao and 8 are

Wette, 1969)

A

¥(a)} + log B = y

8 > Q)
(Choi and

UA

, anda gp=x

where Y(t) is the psi (diagamma) function.

Hence we see by the invari-

ance property_of MLE's that x is the MLE of FX for the gamma distribution.
In addition, x is also an MVUE of EX.
This result is known because (1)

the gamma distribution is a member of the exponential family, (2) the
set of minimal sufficient statistics (MSS) is

612

Select target paragraph3