Causal Mapping
1
What’s in this guide?
I Introduction
2
Manifesto: piecing together fragments of causal information
3
Causal maps: a unifying idea
3.1
Causal map
paradigms
3.2
Mini-maps with more than one influencing item
3.3
Semi-formal definition of a causal map paradigm
3.4
Hybrid sets of rules
3.5
Causal
multi-maps
?
3.6
A brief overview of different causal map paradigms
3.6.1
Approaches already called “Causal Mapping”
3.6.2
(Comparative) Cognitive Mapping
3.6.3
Theories of Change for a project or programme,
3.6.4
Programme theories in theory-based evaluation
3.6.5
Fuzzy Cognitive Maps
3.6.6
Contribution analysis
3.6.7
Systems Diagrams
3.6.8
DAGs as promoted by Judea Pearl
3.6.9
Structural Equation Models
3.6.10
Bayesian belief networks
3.6.11
(arguably) diagrams used in Realist Evaluation
3.6.12
(sometimes) diagrams used in Outcome Harvesting
3.6.13
Causal Maps as constructed in QuIP
3.6.14
Mental models
3.6.15
Knowledge graphs, semantic networks
3.6.16
NCA (Necessary Condition Analysis)
3.6.17
Influence diagrams
3.6.18
Paradigms which include the aspect of belief
3.6.19
“System Effects”
3.6.20
QuIP maps
3.6.21
Maps of maps
3.7
Summary
4
Causal mapping
5
The causal map app in four pictures
6
We need a “soft arithmetic” for causal maps
6.1
An ambitious project
6.2
Users of a causal map expect to be able to deduce some kind of comparative information from it
6.3
Asking and answering those kinds of “typical questions” of a causal map boils down to assigning some kinds of numbers to its elements
6.4
Aren’t there strategies to encode causal information without using any kind of number?
7
We need rules about how to encode causal information in a causal map (and decode it again)
7.1
How do causal claims work within ordinary language?
7.2
Understanding the elements of a causal map by agreeing how to make deductions with them
8
Soft arithmetic is also the answer to understanding causal maps
9
Metalanguage: some words for talking about causal maps
9.1
“The cause”, “a cause”, “(causal) influence”
9.2
Nodes, variables, vertices …
9.3
Influence variable, consequence variable, package of variables
9.4
Mechanism, theory ….
II The rules of Soft Arithmetic for causal maps
10
The mini-map coding rule
10.1
Meaning depends also on elicitation context
11
The mini-map coding rule – functional form
12
The extras rule: adding extra information, in particular about the levels or values of the variables
12.1
Interpretation
12.2
Corollary: Ordinary reasoning
13
The juxtaposition rule
14
Rules for joining maps
14.1
The chaining rule
14.2
The shared consequence rule
14.3
The shared influence rule
14.4
The shared arrow rule
15
Rules for joining maps: what counts as “the same” variable?
16
The chaining rule, functional form (zooming out / black-box rule)
16.1
Two special cases
16.1.1
Bare influence
16.1.2
Total control
16.2
Technical note
17
The shared consequence rule, functional version
17.1
Contradictory combinations
17.2
But which case is which?
17.3
Extension to causal
packages
18
The shared influence rule, functional version
19
The shared arrow rule, functional form.
20
The rule for merging arrows. Weight of evidence.
20.1
Problems with evidence
21
The chaining rule with loops
22
The rule for conceptual links
22.1
Variations
22.2
Problems
23
Combining “extra” information like the values or levels of variables
24
Context
24.1
Intersections
24.2
Replacing contexts with variables
25
Maybe the nodes are not variables, maybe they are … propositions, events, schemas?
25.1
Events
25.2
Propositions
25.3
Schemas / Schemata
26
Variable-based and propositional-based approaches to causal diagrams: chalk and cheese??
26.1
The variable-based approach
26.2
More on the proposition-based approach
26.2.1
The items are propositions, not variables
26.2.2
The items have been individually experienced
26.2.3
The propositions have contrastive implications
26.2.4
Two-sidedness of propositions
26.2.5
Aggregation
26.2.6
The items can express changes, states or events
26.2.7
Golden threads
26.3
Statistics?
26.4
Conclusion
27
Limitations to the proposition-based approach
27.1
Problems
within
the proposition-based approach
27.1.1
Problem with only coding personal observations rather than personal causal knowledge
27.1.2
Problems with establishing common codes using difference of degree
27.1.3
Need for a distinction between “plus” and “minus” links?
27.2
Limitations: Problems with
extending
the proposition-based approach
27.2.1
No way to code multiple causation
27.2.2
Problems with expressing difference of degree
27.2.3
Explicit causal knowledge
27.2.4
Vulnerability to luck / coincidence
27.2.5
Unable to code necessary/sufficient conditions
27.3
Weaknesses of the variable-based approach
27.4
What is to be done? Suggestions for QuIP
28
Clarifying how to ask QuIP questions in contexts when the final outcomes are not necessarily “changes”
28.1
Classic QuIP: the focus on changes
28.2
Even final outcomes don’t have to be changes
28.2.1
One-off events
28.2.2
Maintaining the status quo
28.2.3
Status quo as a divergent counterexample
28.3
Preceding items don’t have to be changes either
28.4
How to focus on the relevant contrast
28.5
Drivers
28.6
Focus on initial items (“drivers”)
29
Coding individual causal fragments using propositions (revised)
29.1
Basic coding rule
29.1.1
Every causal propositional claim is actually three claims
29.1.2
The items have been witnessed
29.1.3
The propositions have contrastive implications
29.1.4
Can include claims about absence of expected states
29.2
Mini Extension: causal chains
29.3
Extension: propositional claims about
additive contributions
29.3.1
Encoding
incomplete
claims about additive contributions
29.4
Examples we can code
29.5
Examples we cannot code
29.6
That’s it
30
Combining causal fragments from different sources using propositions
30.1
Same item
30.1.1
Extension 2: Subsuming items under more general items
30.1.2
Extension 3: Recoding items from different sources as
gradations
of more general items
30.1.3
Extension 3a: allowing negative gradations
30.2
What these extensions still don’t do
30.2.1
Can’t distinguish between AND and OR
30.2.2
Necessary / sufficient conditions
31
Coding causal maps with propositions
31.1
Basic coding rule
31.1.1
Three claims
31.1.2
The items have been witnessed
31.1.3
The propositions have contrastive implications
31.1.4
Can include claims about absence of expected states
31.2
Mini Extension: causal chains
31.3
Mini Extensions: certainty and trust
31.4
How does classic QuIP extend this basic idea?
31.4.1
Extension 1: Combining causal fragments from different sources: same item
31.4.2
Extension 2: Subsuming items under more general items
31.4.3
Extension 3: Recoding items from different sources as
gradations
of more general items
31.4.4
Extension 3a: allowing negative gradations
31.5
That’s still not enough
31.5.1
No way to encode explicit claims of multiple causation
31.5.2
No way to encode implicit claims of multiple causation
31.6
Solution: propositional claims about contributions, and assume additive causation
31.7
What these extensions still don’t do
31.7.1
Can’t distinguish between AND and OR
31.7.2
Necessary / sufficient conditions
32
Types of variable
33
Causal thinking is essentially contrastive thinking
34
Lo/hi Variables, types of variable, and contrasts
35
“For each ..….” variables
36
Coding specific influences – individually and in combination
36.1
Summary of the argument so far
36.2
Encoding causal influences between lo/hi variables
36.3
Breaking down the problem
36.4
Some “mono functions”: functions with a single influence variable
36.5
Some “multi functions”: functions for packages with more than one influencing variable:
36.5.1
SOFTADD
36.6
“Package-free” combinations
36.7
Contradictions?
36.8
Building up more functions from these building blocks
36.9
Coding influences within in the app
37
One function to rule them all?
37.1
One universal type of function for coding causal claims with exclusively lo/hi variables.
38
Coding a claim about the absence of a causal link
39
Strength / importance
39.1
Including the idea of strength in our functional equation
39.2
Direct ways of eliciting how important is one variable’s contribution to another. (Fiona)
40
Causal Inference
40.1
Specifying the map
40.1.1
Exogenous variables and the Grid
40.1.2
Inference in action
41
Upstream (Bayesian) inference
42
Contour
42.1
Flipping
43
Valence
44
Valence and direction
45
Effect: two kinds
46
Interventions and differences
46.1
Shortcuts
47
Contribution
47.1
Why reporting correlation is not helpful
48
Information about the
source
of our causal information
48.1
Applying the “trust” attribute source-by-source
48.2
Including information about sources
49
Clusters of similar maps
50
Maps of maps
51
Probability density functions
III Simplification
52
Simplifying causal maps: aggregation and filtering
52.1
Why simplify a causal map?
52.2
The tables available
52.2.1
Table of Statements
52.2.2
Table of Variables
52.2.3
Table of Arrows
52.3
The steps taken by the app when simplifying a network according to user commands and/or automatically if requested.
52.3.1
Inference
52.3.2
Calculation of source-by-source data
52.3.3
Merging together variables which are in the same cluster
52.3.4
Merge information about Statements into the arrows which are based on them
52.3.5
Filter sources
52.3.6
Hard-coded calculations
52.3.7
Merge arrows
52.3.8
Calculate QuIP-style metrics, e.g. summaries per domain.
52.3.9
Filter out arrows below a minimum frequency
52.3.10
Add arrow attributes to variables
52.3.11
Filter out variables below a minimum frequency
53
Aggregating and filtering beliefs
54
Aggregation and filtering based on face value
55
Aggregation and filtering based on particular research questions
56
Aggregation and filtering based on metrics
56.1
Network metrics
57
Citation intensity
58
Conspicuous absence
59
Not forwards
60
Homogenity of paths
60.1
(non-) solution 1)
60.2
Solution 2)
60.3
Solution 3)
60.4
Solution 4)
60.5
Solution 5)
IV Visualising
61
Visualising and formatting causal maps
61.1
Conditional formatting
61.2
Hard-coded formatting
61.3
Constructing labels and tooltips
61.4
Focus
62
Coding using the UI (outdated)
62.1
Show that the influence of one variable on another has a particular contour.
62.2
Show that variable X is part of the definition of variable Y.
62.3
Show that variables X and Y are linked by definition
62.4
Show that a set of influence variables influence the consequence variable
collectively
.
V Summary
63
Summary of the rules for inference in causal maps, aka “Soft Arithmetic”
63.1
The inference rules for causal maps
63.2
The mini-map coding rule
63.3
Focus on “lo/hi” variables and functions between them
63.4
Recording the actual values of variables
63.5
The rule for conceptual links
63.6
The rules for coding different types of influence; single influence variable
63.7
The rules for coding different types of influence; packages of multiple influence variables
63.8
INUS
63.8.1
SOFTADD
63.9
Coding the strength of the influence of a package
63.10
Causal claim coding form
63.11
Problems
VI The why question
64
The “
why
question” as a generic method in social research
64.1
Steps to asking “the why question”
64.1.1
Formulate …
64.1.2
Pose the question
64.1.3
Get the answers into a spreadsheet
64.2
When would you
not
want to use this approach?
VII The why question
65
The “
why
question” as a generic method in social research
65.1
Steps to asking “the why question”
65.1.1
Formulate …
65.1.2
Pose the question
65.1.3
Get the answers into a spreadsheet
65.1.4
Start analysing
65.2
When would you
not
want to use this approach?
66
The “
why
question” – asking about changes
66.1
“Changes”
VIII Using the causal map app
67
Using the causal map app – already partly outdated!
67.1
About the tables
67.2
Left-hand tabs for producing and coding your diagram
67.2.1
Import
67.2.2
Code
67.2.3
Merge
67.2.4
Display
67.3
Right-hand tabs for outputs: viewing diagram etc
67.3.1
Diagram
68
Using the causal map app for real-time, collaborative theory construction
IX Case studies
69
Case Study: the Strawberry Line
69.0.1
Theories gathered from activists
69.0.2
Stories gathered from the public
70
Case study: Global Young Academy. Tracing the paths of GYA’s impact
70.1
More details
70.1.1
Aims
70.1.2
Methods
70.1.3
Findings
71
Appendix: previous work
71.1
From “Theorymaker” to “Causal Mapping” and from “Theories of Change” to “Causal Maps”
71.2
Articles
71.3
Presentations
71.4
Poster
71.5
Longer blog posts
71.6
Resources and apps
72
References
Published with bookdown
Causal Mapping
Section 38
Coding a claim about the absence of a causal link
Is it really the same as strength=0. I think not.