Top-down or bottom-up qualitative coding?

In framework analysis, sometimes described as a top-down or 'a-priori' approach, the researcher decides on the topics of interest they will look for before they start the analysis, usually based on a theory they are looking to test. In inductive coding the researcher takes a more bottom-up approach, starting with the data and a blank-sheet, noting themes as the read through the text.

Obviously, many researchers take a pragmatic approach, integrating elements of both. For example it is difficult for a emergent researcher to be completely naïve to the topic before they start, and they will have some idea of what they expect to find. This may create bias in any emergent themes (see previous posts about reflexivity!). Conversely, it is common for researchers to discover additional themes while reading the text, illustrating an unconsidered factor and necessitating the addition of extra topics to an a-proiri framework.

I intend to go over these inductive and deductive approaches in more detail in a later post. However, there is also another level in qualitative coding which is top-down or bottom-up: the level of coding. A low 'level' of coding might be to create a set of simple themes, such as happy or sad, or apple, banana and orange. These are sometimes called manifest level codes, and are purely descriptive. A higher level of coding might be something more like 'issues from childhood', fruit, or even 'things that can be juggled'. Here more meaning has been imposed, sometimes referred to as latent level analysis.

Usually, researchers use an iterative approach to qualitative analysis, going through the data and themes several times to refine them. But the procedure will be quite different if using a top-down or bottom-up approach to building levels of coding. In one model the researcher starts with broad statements or theories, and breaks them down into more basic observations that support or refute that statement. In the bottom-up approach, the researcher might create dozens of very simple codes, and eventually group them together, find patterns, and infer a higher level of meaning from successive readings.

So which approach is best? Obviously, it depends. Not just on how well the topic area is understood, but also the engagement level of the particular researcher. Yet complementary methods can be useful here: the PI of the project, having a solid conceptual understanding of the research issue, can use a top-down approach (in both approaches to the analysis) to test their assumptions. Meanwhile, a researcher who is new to the project or field could be in a good position to start from the bottom-up, and see if they can find answers to the research questions starting from basic observations as they emerge from the text. If the themes and conclusions then independently reach the same starting points, it is a good indication that the inferences are well supported by the text!