ESTIMATION OF EXPECTED VALUES OF ENVIRONMENTAL
POLLUTANTS WITH LOGNORMAL AND GAMMA DISTRIBUTIONS

G. C. White
Los Alamos Scientific Laboratory
Los Alamos, New Mexico

ABSTRACT

Concentrations of environmental pollutants tend to follow positively
skewed frequency distributions.
Two such density functions are the
gamma and lognormal. Minimum-variance unbiased estimators of the expected value for both densities are available.
The small-sample statistical properties of each of these estimators were compared for its own
distribution, as well as the other distribution to check the robustness
of the estimator.
Results indicated that the arithmetic mean provides
an unbiased estimator when the underlying density function of the sample
is either lognormal or gamma, and that the achieved coverage of the
confidence interval is greater than 75 percent for coefficients of
variation less than two.
Further, Monte Carlo simulations were conducted
to study the robustness of the above estimators by simulating a lognormal
or gamma distribution with the expected value of a particular observation
selected from a uniform distribution before the lognormal or gamma
observation is generated.
Again, the arithmetic mean provides an unbiased
estimate of expected value, and the achieved coverage of the confidence
interval is greater than 75 percent for coefficients of variation less
than two.

INTRODUCTION

The concentrations of environmental pollutants have been suggested to
follow positively skewed frequency distributions by numerous researchers.
In particular, Pinder and Smith (1975) investigated the goodness of fit
of the lognormal, Weibull, exponential, and normal distributions to
radiocesium concentrations in soil and biota.
They found that the
lognormal distribution fit the majority of the data sets.
Giesy and

Weiner (1977) found that the lognormal also tended to fit the concentrations of trace metals in fish better than the Weibull, exponential, or

609

Select target paragraph3