9597 - Laserfiche WebLink

Home Browse Search

BIOLOGICAL METHODS sample or a plankton haul. However, 10 this section the term "sample" will be used to denote "a set of observations" - the written records themselves. 1.1.6 Parameter and statistic When we attempt to characterize a popula- tion, we realize that we can never obtain a per- fect answer, so we settle for whatever accuracy and precision that is required. We try to take an adequately-sized sample and compute a number from our sample that is representative of the population. For example, if we are interested in the population mean, we take a sample and com- pute the sample mean. The sample mean is referred to as a statistic, whereas the population mean is referred to as a parameter. In general, the statistic is related to the parameter in much the same way as the sample is related to the pop- ulation. Hence, we speak of population param- eters and sample statistics. Obviously many samples may be selected from most populations. If there is variability in the population, a statistic computed from one sample will differ somewhat from the same statistic computed from another sample. Hence, whereas a parameter such as the population mean is fixed, the statistic or sample mean is a variable, and there is uncertainty associated with it as an estimator of the population parameter which derives from the variation among samples. 2.0 STUDY DESIGN 2.1 Randomization In biological studies, the experimental units (sampling units or sampling points) must be selected with known probability. Usually, random selection is the only feasible means of satisfying the "known probability" criterion. The question of why known probability is re- quired is a valid one. The answer is that only by knowing the probability of selection of a sample can we extrapolate from the sample to the population in an objective way. The probability allows us to place a weight upon an observation in making our extrapolation to the population. There is no other quantifiable measure of "how well" the selected sample represents the population. Thus our efforts to select a "good" sample should include an appropriate effort to define the problem in such a way as to allow us to estimate the parameter of interest using a sample of known probability; i.e., a random sample. The preceding discussion should leave little doubt that there is a fundamental distinction between a "haphazardly-selected" sample and a "randomly-selected" sample. The distinction is that a haphazardly-selected sample is one where there is no conscious bias, whereas a randomly- selected sample is one where there is consciously no bias. There is consciously no bIas because tne randomization is planned, and therefore bias is planned out of the study. This is usually accom- plished with the aid of a table of random numbers. A sample selected according to a plan that includes random selection of experimental units is the only sample validly called a random sample. Reference to the definition of the term, sample, at the beginning of the chapter will remind us that a sample consists of a set of observations, each made upon an experimental or sampling unit. To sample randomly, the entire set of sampling units (population) must be identifiable and enumerated. Sometimes the task of enumeration may be considerable, but often it may be minimized by such conveniences as maps, that allow easier access to adequate representation of the entity to be sampled. The comment has frequently been made that random sampling causes effort to be put into drawing samples of little meaning or utility to the study. This need not be the case. Sampling units should be defined by the investigator so as to eliminate those units which are potentially of no interest. Stratification can be used to place less emphasis on those units which are of less interest. Much of the work done in biological field studies is aimed at explaining spatial distri- butions of population densities or of some parameter related to population densities and the measurement of rates of change which permit prediction of some future course of a biologically-related parameter. In these cases the sampling unit is a unit of space (volume, area). Even in cases where the sampling unit is not a unit of space, the problem may often be stated 2 . .