Until now we have focused on mapping geographical phenomena at various scales and with various tools and strategies. But, what are these phenomena and how do we gather data on their special nature and property? Well, there are several different types of data and we will cover some of the more common forms one at a time. As for collecting data, there are numerous problems with which to be dealt and these comprise the topic at hand.
Geographic field studies can involve large areas and a variety of spatial phenomena. In many circumstances, it is not possible to investigate every occurence of a phenomenon within the study area because of time and cost constraints. Accordingly, a small percentage or a sample (n) of the total area is often selected from the larger body of data known as the population (N). For example, national population censuses typically ask a few basic questions of every household (100% sample). Many more and detailed questions are asked of only 5% of the households. Does 1 out of every 20th household seem like an adequate sample on which to base generalizations for an entire country? If so, you will doubtless be surprised, or shocked, to learn that even smaller sample sizes are used in some very important studies. Most people, for instance, take comfort in knowing the US Food and Drug Administration conducts clinical studies on medications before they are approved for public use. Few, however, recognize the small size of the samples that are used. Check section 14 of the package inserts for two common drugs, the antidepressant Prozac (n=671), and the sleep medication Ambien (n=462). These samples constitute less than 0.0000015 percent of the US population.
In addition to the sizes of samples, a major concern in sampling is to do so in such a way as to not introduce an unreasonable amount of bias into the study. To wit, testing the effectiveness of Yasmin with an all-male sample would certainly be biased. The term "unreasonable" is used because there will probably always be some bias.
Exploratory sampling is, as the name implies, used in those cases where next to nothing is known of a particular phenomenon area. In effect, it is a first attempt to learn something. For example, let's say you are employed by a rural development agency to find out how many homes in a remote impoverished area lack indoor plumbing. One way of carrying out an exploratory survey might be to drive around and locate those houses with outhouses. While the fieldworker is at it, she or he might even get creative and gather additional data about outhouses such as one hole or two. Exploratory sampling allows researchers to produce maps of things not previously documented. These maps may be crude, but crude is better than nothing, and it can lead to better things.
Reconnaissance sampling is carried out in areas where some information exists, but more is needed. It is usually more systematic than exploratory sampling, and attempts to cover as much area as possible in a brief period at minimal cost. A classic example of this kind of sampling involved the founder of this department, Donald D. Brand, and his mentor, Carl O. Sauer. In the late 1920s they were looking for archaeological sites in northwestern Mexico. They knew, on the basis of exploratory surveys by earlier archaeologists, that sites existed in the area, but nothing more. Sauer and Brand wanted to refine knowledge. According to Brand they drove a Model A Ford which broke down frequently. When the car would fail, Sauer would shout "Brand, fix the car!" and then climb the nearest hill. Once he heard the car start, Sauer would descend the hill with a set of notes about the site on the hill. All their sites correlate with breakdowns.
Extensive sampling involves enough existing background data to formulate a comprehensive research design. Extensive sampling provides generalizations about large areas. For example, extensive sampling of vegetation might reveal regional patterns (e.g., pine forests, grasslands).
Intensive sampling provides more detail about small areas. Using vegetation as an example once again, intensive sampling would reveal individual site variations in numerous locales (e.g., details about the forest-prarie ecotone).
Point sampling involves selecting specific "points" at which to collect specific data. Information about those points can then be used to say something about a phenomena over a broader area. For instance, how might one learn something about soils on a particular plot of land? One way might be to select several points from the plot and collect soil cores at these points. Differences and/or similarites from one point to another can then be discussed.
Area sampling involves collecting data from small areas in order to say something about larger areas. There are two types-fixed and variable. In fixed area samples, the size and the shape of the sample areas remain constant. These are usually circles demarcated by a cord of a specified length (radius), or squares delineated by cords with knots at corner points, or small frames. The sizes of the sample areas changes with the grain and the density of the phenomenon under investigation. For example, studies of grasslands (fine-grained and dense vegetation) can be conducted using sample areas as small as one meter square. Studies of forests (coarse-grained and sparse) need larger areas as some individual trees may encompass one meter themselves.
One method used for determining sample area size is the Phenomena-Area Curve. Using this method, field workers create a graph with area as the x-axis, and the number of phenomena being the y-axis. The number of items found in a 1 m quadrant are plotted on the graph. The number of items in areas measuring 1 x 2 m, 2 x 2 m, 2 x 4 m, etc. are then plotted. The dots are connected and a point just before the curve becomes horizontal is selected. The location of this point on the x-axis is the ideal size of the sample area.
A variation of the fixed plot sample is called Parcelle Mapping. This strategy involves covering the entire region with a grid. The individual grids are uniform in size and shape and determined by the method just outlined. Sampling can involve any of a number of selection processes, but essentially, some grids are studied and others are not.
Some area samples use areas of variable size and shape. Such samples are not unusual in urban and agricultural studies, for example, where city blocks and individual fields are not of uniform dimensions. Not infrequently, variable sample areas are called "releves" or "ocular plots."
Linear sampling is literally nothing more than establishing a transect across the landscape, not unlike a baseline. This can be done in the office on a map or aerial photograph and then followed in the field. Data of one type (e.g., crops) or several types (e.g, physical and cultural) are recorded in reference to stations along this line. Some linear samples involve "belts" rather than lines. Others involve areas along the line, not unlike a combination of linear sampling and Parcelle mapping. Yet others involve points on the line. In those cases where areas or points along a line are used, these areas or points are predetermined before going into the field.
Plotless sampling is a fourth type of sampling unit. It is rather complicated and time-consuming, and, to be frank, if not biased, not worth the trouble. Almost every type of geographical phenomena can be accurately evaluated using either point, area, or linear sampling units.
In addition to considering the type of sample to be used, fieldworkers must understand the theoretical basis of sampling design. There are three basic approaches to designing samples.
Hierarchical sampling involves collecting data from several levels--e.g., block, subdivision, city, county, state, country. It allows for understanding a phenomenon at several scales.
Random sampling is especially important if statistical analyses will be used once the data are collected, however, it runs the risk of not providing uniform regional coverage. For example, some small but very important information can get overlooked. There are well-accepted strategies for selecting sampling locales randomly. In some cases, however, fieldworkers "arbitrarily" or "intuitively" choose sampling sites.
Systematic sampling involves establishing sample points at regular intervals. For example, a fieldworker may elect to assess some phenomenon on the basis of its occurrence or absence at the intersection of all township and range lines. The greatest problem with this approach is that sample points are uniformly distributed and if the phenomenon under investigation occurs regularly all or none of the phenomenon might be included.
Stratified random sampling is the best of all possible approaches. In essence, it combines random and systematic sampling. Typically, the region under study is overlaid with a grid for which coordinates are established. One point, defined by these coordinates, is randomly selected from each grid, thereby resulting in reasonably uniform sample coverage.
There are no hard and fast rules to sampling. The world is simply too complex and varied for any set strategies to work. Field workers have to exercise good judgement. Standard or commonly accepted approaches can be used, but more often than not they will have to be "adjusted" for local conditions, circumstances, and situations. Topics discussed above should be considered as guidelines.