Analytical units force us to be specific about what we do. How much text do we code? What context do we take into consideration? What documents or text segments do we include or exclude from our analysis? Thinking about them can definitely help us improve the quality of our analysis.
When looking into the methodological literature about Qualitative Data Analysis, we see analytical units everywhere. Here, we will first take a look at the purpose of analytical units and then compare the analytical units of two prominent (German) authors, Udo Kuckartz and Philipp Mayring.
What are analytical units and what are they good for?
Analytical units serve the purpose of explaining what, when and how to code. When “coding”, – that is, when we apply categories, codes or tags to segments of text – we have to follow a set of rules in order to achieve systematic, non-random and reproducible results. When we start coding with any QDA-software, we will quickly run into many questions: How long should our coded segment be? Should it be just one word? A couple of words? A sentence or even an entire paragraph?
For example: Above, we want to know what the devil looks like in historical documents. What should we code? Just one word “dog”? A couple of words “a black dog?” half a sentence “and he looked like a black dock” or even an entire paragraph? (I spare you that example!)
What is the best way? Only we can decide on how much context we want to preserve and how granular we need to be. But there are more questions just around the corner!
Another common question is: What context is taken into consideration when we decide upon what category to apply? With the example from above we look again at witchcraft documents for a start:
“6. Admits, she cannot tell, how often she has magically travelled to the Hainburg-mountain and to the crossroads.”
Are we allowed to code that as “witches’ meeting”? There is nothing said about a meeting here. But we might know that there was one from either I) a different part in the same document. II) a different (even contradicting?) document within the same corpus, III) our own knowledge or secondary literature.
For such questions we should strive to create general answers, these are our analytical units. The purpose of them is to force us to explicitly address such questions. Unfortunately, in the methodological literature there are different variants of them and they can be quite confusing.
Philipp Mayring (2015)
In an attempt not to misinterpret Mayring (which is not an easy thing to do) we look here at his very brief definitions (German) and my translation of it. (Mayring 2015, p. 61)
When I read this the first time, one and two sounded simple and the third one was confusing. Let’s try to understand what Mayring means.
According to Mayring, the coding unit is the smallest part of the material that can be analysed. He says that this will often be one “proposition” so the smallest unit that can make sense on its own. This is the minimum size of text that can be coded.
The second one, the context unit, seems at first glance to be the opposite, the largest piece of text that goes under one unit. However, it becomes clearer later that by “context unit” he rather means how much context we are allowed to take under consideration, when evaluating under what code/category a single coded segments belongs (see above my example with the witches meeting). In any case, it makes sense to think about both: The outer limit of our coded segment and the context we need to understand it.
The third, evaluation unit, is confusingly described. After talking to him in person (which was admittedly not much help), my understanding of this unit is, that it says what parts of the material are relevant for the analysis and in what sequence they should be analysed. Where do we start with our analysis? Do we look at the entire text or just at parts of it? When creating our codes inductively we of course have to look at the entire text, we cannot leave anything untouched. However, in other cases this might not be necessary. For example, when analysing newspapers, we might not be interested in advertisements, or instead we might be interested in nothing else! In any case, we will not need to analyse the newspaper in all its entirety.
Udo Kuckartz (2014)
Kuckartz presents other units than Mayring does. Fortunately, his languages is much clearer. In principle he distinguishes four different units (Kuckartz 2014, p. 49-48):
- Sampling unit
- Unit of Analysis / Recording unit
- Coding Unit
- Context Unit
The Sampling unit is the selection of data/source that we make from the entirety of potential sources out there. We can take a random sample, follow a certain quota or choose them as we see fit. For example, from thousands of newspapers we decide to look at 10 specific newspaper editions from Paris and 10 from Berlin, from January 1st 1982.
Unit of Analysis / Recording unit. While the sampling unit decides weather or not something is included in our research, the unit of analysis turns our data into individual segments or groups. Continuing the example above the unit of analysis might be each individual newspaper or (on a more detailed level) the individual articles that we are considering independently from one another. Most commonly the unit of analysis are our “documents” within the QDA-software (see Rädiker; Kuckartz 2019, p. 24).
The Coding Unit is the individual text segment that we assign our code to. For the coding unit there has to be a clear criteria what triggers a text to be coded. Under this “coding unit” Kuckartz also covers considerations about its formal criteria (length of the segment) and also the type of its creation (inductive/deductive or data-driven/concept driven).
The Context Unit with Kuckartz is also the largest amount of information that might be taken into consideration when evaluating under what category/code an individual segment belongs. (See again the example of the witches meeting above.)
What do we take from this? Who should we follow? Do we really need to define all that?
Well, we are certainly well advised to think about each of these units in relation to our own research. Are they relevant for us? Which of them are important in our case? However, they should not cause us sleepless nights, let’s focus instead on what we take away from them:
- We should consider, what data we are including in our research and how we choose it (sampling unit).
- We should consider, what different types of data we have and how we want to treat them (unit of analysis).
- We should consider, the size of our coded segments, what is the minimum and what is the maximum (coding unit).
- We should consider, what context we are using (or not using) when coding our documents (context unit).
- We should consider, what part of our data/documents is indeed relevant for us and in what order we intend to analyse it (evaluation unit).
As we saw, we can learn much from analytical units and there are good reasons to take them under serious consideration. Although they might be confusing in the beginning, it can be illuminating to think about each of them and many of them should definitely go into the methods chapter of our paper. However, if you feel unsatisfied with them, don’t take them as something given but feel free to change them for your needs.
(For my part, I have defined “scope” and “quantity” as important analytical units myself, we will look into this in a later article.)