Sampling design also involves choices of the interrelated factors of sample size/fraction and the sampling unit (shape and size). They are interrelated because of statistical considerations and because of the costs of doing archaeology. In statistics, sample size is usually expressed as the number of sampling units that will be, or have been, investigated. In popular political polling, the results frequently mention, for example, that 4123 people responded out of 10 000 questionnaires; 4123 would be the sample size while 10 000 would be the sampled population. In archaeology, sample size is best thought of as sampling fraction because it is usually expressed as a fraction or percentage, for example, 10% sample, meaning that 10% of the project area, site, or region is being investigated. Sample size refers to the number of sampling units needed to obtain a certain level of precision and can be determined by solving statistical formulas for sample size after possible precision levels and other archaeo-statistical factors are determined.
Sampling unit is the methodological unit, usually spatial in nature, through which archaeologists look in order to observe the past. Sampling unit consists of the size and shape. For example, quadrats (square or rectangular) and transects (rectangular) are two types of sampling units in regional survey. Rectangular trench excavations (manually dug or by backhoe) and square grid units (large block excavations, e. g., 25 ft on a side, 625 sq. ft) are analogous shapes in site excavation. A larger number of smaller sampling grid units increases the cost of locating the units and works best for descriptive CRM/PPG 16 surveys. For distributional research questions concerning relations and associations among various data classes, larger, square sampling units are preferred, but such units are difficult statistically.
Sample size is also related to sample comparisons through the concept of significance. As sample size increases for two samples being compared, the probability of any given difference between the samples decreases. In other words, it is easier to show statistical differences through significance with larger sample size; samples may in fact become too large. On the other hand, when sample size is small, it may be difficult to show that two samples are statistically different. A ‘via media’ solution balances the costs and benefits of large sample sizes/fraction against large sampling units and against the stated research objective.
World History
