# Sampling design and plot design

For any sampling study, there are three basic design components that have to be planned: (1) sampling design, (2) response design and (3) estimation design. All three need to be clearly defined before implementing an inventory!

## Sampling design

Usually a sampling exercise in forestry starts with the selection of locations at which certain observations should be made. This first "design element" is therefore the definition how (and how many) locations (dimensionless points) should be placed inside the area of interest (or in more statistical words: the sampling frame). Typical examples are simple random sampling or systematic sampling (see the category sampling design).

## Response design

defines how the observation unit “response”. As the sampling design only defines a dimensionless point that is not suitable to make any observations, one needs a "rule" to define what to observe around this location. It is also called plot design or observation design or field protocol. In the forest inventory context, the response design defines the plot design to be used and the measurements that are to be taken on that plot. The observation units don't have to be areal plots in all cases. It is also possible to define e.g. a number of nearest trees that should be included at each location. These designs are known as distance based plots, k-tree sampling or "plotless density estimators" in ecological literature. Also two dimensional lines can be used as observation units (see line sampling). All in all, the response design describes what is to be done on the selected location that is defined by the sampling design.

## Estimation design

defines the formula (the estimators) that are used for estimation. One is not free to choose any formula but the estimator needs to be compatible to both sampling and response design. It cannot be enough stressed how important it is to have the estimation design completely clear before starting to go out and collect data. There are too many cases in which data were collected according to some complex sampling or plot design - and at the end it was not known how to do a proper analysis of the data.