You are on page 1of 5

VARIABILITY AND BIAS

The three major sources of variability in this method : sampling variability, estimation variability, and measurement error

The major sources of bias : preferential and clustered sampling, and bias in estimation and measurement

Sampling Variability and Bias


Since geological, geochemical, and other techniques can be used to identify promising areas, the search for minable coal often results in preferential sampling aimed at finding the most economically viable resources. Once a mine site has been established, numerous closely spaced holes are drilled to help develop the mine plan. Over 50% of the drill hole data are clustered in mined areas located in the south-central region of the bed, while no drilling occurred in much of the northern and northeastern parts of the bed.

We seek to develop a method to place a confidence interval on the volume of coal in each of the geologic assurance categories. Many of the usual assumptions, including that of a random sample, will not be met. If the highly drilled areas predominate and contain thicker and less variable coal than is present in the rest of the bed, then a confidence interval estimate on the mean volume of remaining coal would be too narrow and biased upward. In some instances it is possible to test for biases. Often, however, such tests have little power because drilling is sparser in the areas where the resources need to be estimated. Some bias correction procedures adjust the sample mean (mean thickness) and/or the variance. Reducing the influence of these oversampled areas by subsampling is risky because, particularly in this study, we need information that only can be obtained from closely spaced points.

In the estimation process we assumed that thickness of the coal beds/zones varied continuously and that once we removed the large-scale trend, the residual thickness Z(u) is such that the random function increments Z(u)Z(uCh) are stationary of order two. Discontinuities sometimes exist and may be caused by channeling, splitting of a bed into two or more parts, and tapering. These factors may be more of a concern when economic filters are applied to the resource base. However, if drill holes that were terminated have been incorrectly classified as completed, variability and bias could be affected. Any sig-nificant relationship between local means and standard deviations, referred to as the proportional effect, may invalidate the intrinsic hypothesis. Graphical tech-niques will be used to investigate this phenomenon. In this paper we refer to coal beds and zones. The Harmon is a bed or single stratum. A zone refers to two or more beds. We would prefer to analyze beds; however, due to geologic complexity it is not always possible to identify individual beds.

Estimation Variability and Bias


The method that we recommend requires the computation of a sample semi-variogram and fitting a model. Coal thickness in most beds/zones displays a trend. We refer to this as large-scale variability. Cressie (1991) showed that over- or under-specifying the trend could bias the estimation of the semivariogram. Goovaerts (1997, p. 143) cautioned that, the semivariogram of estimated residuals, (h), strongly depends on the algorithm used to estimate the trend component. Consider the representation of a random variable Zat locationu(Cressie, 1991, p. 290) as = + ,

You might also like