From (Powell et al., 2024)

Some approaches such as that suggested by Markiczy and Goldberg (1995) directly elicit causal links from their sources, perhaps by asking respondents to suggest causal links between a predetermined list of causal factors, and thus, after finishing Task 1, are already in a position to create causal maps.

(See also the dedicated chapters on coding...)

• Task 2 -- Introduction
• Tasks 2 & 3 -- Introduction

More explicitly, qualitative approaches are faced with Task 2: encoding causal claims in the form of explicit causal links and factors. This task is similar to ordinary qualitative data analysis (QDA), whether done manually or using tools like NVivo, Dedoose and AtlasTI. However, these tools are designed to capture general concepts, rather than claimed causal links between concepts, which is what we need for causal mapping. QDA for causal mapping also starts with a corpus of narrative data, but it does not create causal links between independent concepts that might already have been coded using ordinary non-causal thematic analyses. Rather, in causal QDA, the primary act of coding is to highlight a specific quote from within a statement and identify the causal claim made by simultaneously identifying a pair of causal factors: an ‘influence factor’ and a ‘consequence factor’.

The causal factors only exist as one or other end of a causal link and have no meaning on their own. Each claim forms a link in the visual representation of the causal map. The Axelrod school had its own coding manual describing how to highlight areas of text expressing causal connections and code them as links between causal factors, originally inspired by evaluative assertion analysis (Osgood et al., 1956).

Manual causal coding of text data, like ordinary thematic coding, requires a considerable investment of time and expertise to do well. We now use natural language processing to at least partially automate this; however, the process is essentially the same, and discussion of this is beyond the scope of the present article.

Where do the labels for the causal factors come from? As with ordinary QDA and thematic analysis (Braun and Clarke, 2006), approaches vary in the extent to which they are purely exploratory or seek to confirm prior theory (Copestake, 2014). Exploratory coding entails trying to identify different causal claims embedded in what people say, creating factor labels inductively and iteratively from the narrative data. Different respondents will not, of course, always use precisely the same phrases, and it is a creative challenge to create and curate this list of causal factors. For example, if Alice says ‘Feeling good about the future is one thing that increases your wellbeing’, is this element ‘Feeling good about the future’ the same as ‘Being confident about tomorrow’ which Bob mentioned earlier? Should we encode them both as the same thing, and if so, what shall we call it? We might choose ‘Positive view of future’, but how well does this cover both cases? Laukkanen (1994) discusses strategies for finding common vocabularies. As in ordinary QDA, analysts will usually find themselves generating an ever-growing list of factors and will need to continually consider how to consolidate it – sometimes using strategies such as hierarchical coding or ‘nesting’ factors (as discussed in the following section).

The alternative to exploratory coding is confirmatory coding, which employs an agreed code book, derived from a ToC and/or from prior studies. QuIP studies mostly use exploratory coding but sometimes supplement labels with additional codes derived from a project’s ToC, for example, ‘attribution coding’ helps to signify which factors explicitly refer to a specific intervention being evaluated (Copestake et al., 2019b: 257). However, careful sequencing matters here because pre-set codes may frame or bias how the coder sees the data (Copestake et al., 2019a). Again, the positionality of the coder matters just as much when doing causal coding as it does for any other form of qualitative data coding.

Combining Tasks 1 and 2

Tasks 1 and 2 result in a coded data set of causal claims, each of which consists of (at the very least) the labels for a pair of causal factors. Those using a more explicit elicitation approach have been able to skip Task 1.