One of the very first decisions required when carrying out a corpus-based analysis is to determine what your unit of analysis is. This is a crucial decision because it determines the object of your research; if you do not identify the unit of analysis properly, you may not be able to address the research questions that you have posed.

As you see in this book, corpus-based studies usually have one of two primary research goals: describing a linguistic structure and its variants (as in Part I of the book), or describing some group of texts (as in Part II of the book). Thus, the unit of analysis in corpus-based studies is typically one of two kinds: either occurrences of a linguistic feature or a text. These units of analysis are called the “observations” for the study. Thus, in a study that characterizes a linguistic structure, each observation is an occurrence of the structure in question. In a study that seeks to describe a group of texts, each observation is a text.

Suppose that we plan a study to analyze variants of that-complement clauses and to determine which contextual factors are most strongly associated with each variant. The study should be designed so that each occurrence of a that-complement clause is a separate observation.