A Test of Random Versus Systematic Ecological Samplig

Of the numerous published papers pertaining to ecological sampling, most are concerned with the size, shape, and number of plots making the most efficient sample, or yielding the greatest precision for the minimum of work. The problem has been attacked empirically, by use of the species-area curve (Cain 1938) or by statistical methods (Clapham 1932, Ashby 1935, Pechanec and Stewart 1940, Penfound 1945, and others). The present investigation is of quite a different nature in that it compares a systematic sample, commonly used by ecologists, with a random sample, required by statisticians, as obtained in a completely enumerated forest stand. When ecologists have to analyze the vegetation of a forest stand, many take 10 to 20 quadrats of 10 meters per side, evenly distributed along evenly spaced lines, a pattern which constitutes, if properly done, a systematic sample. This kind of sample is preferred for its easy applicability in the field. It is also believed to be more likely to include the variations of the stand, and thus to be more accurate, than any scheme of random sampling. In fact, this has been proved in certain instances (Hasel 1938, Finney 1948). However, Finney (1949) has shown that, when there is a strongly marked pattern of variation over the stand, this might not be true. If that pattern is periodic, systematic sampling can be considerably less accurate than some kinds of random sampling. Additionally, the major draw-back of systematic sampling is that it does not offer in itself the basis for the assessment of error. Statistical treatment of sampling data leading to estimates of error is necessarily based on the assumption that all sampling units have been chosen independently and at random. It it thoroughly unsound to assess the error from the different sampling plots of a systematic set where the position of all other plots is determined by the location of the first one.