πŸ“• Comparisons

Pages in this section:
Β 
All sections:

βš’οΈ Comparing maps between particular groups

See this page for showing different groups on the labels:
🎨 Formatters: Labels - Surprise

Filtering by respondent group

We code a causal map on the basis of text data. That text data can be usefully broken up into statements, usually of a length between a paragraph and a page. Each statement usually has β€œadditional data” associated with it, for example the ID or gender of the respondent, the text of a question to which this statement is an answer, the page and name of the document from which this statement comes, etc. When we code a causal claim within a statement, we can associate the resulting link with the additional data. That means that for every link, we should know the additional data, e.g.Β the gender of the respondent, etc.
We call the set of statements corresponding to a particular value of a particular additional data field a β€œgroup”. This definition of β€œgroup” is quite broad and does not have to refer only to respondents, e.g.Β the group for β€œquestion 3” is the subset of all the data relevant to that question.
notion image
It is easy to filter a causal map by this additional data. This idea goes back at least to (Ford & Hegarty, 1984). For example, here is one map filtered to show all and only the links mentioned by with female respondents. We call these theΒ per-value maps, e.g.Β the map consisting of all links mentioned by women. However, often the maps for different groups are quite similar as a large proportion of links are shared. When there are many links as in this example, the resulting filtered maps can be uninformative.
notion image
There may still be a bewildering hairball of links. We can apply techniques likeΒ hierarchical codingΒ to β€œzoom out” of the map, or simply show only the most frequent factors. This map shows the top five factors for women:
notion image
And this map shows only the top five factors for men
notion image
Β