2. Punton & Welle (2015) Straws-in-the-
wind, Hoops and Smoking Guns: What
can Process Tracing Offer to Impact
Evaluation?
http://www.ids.ac.uk/publication/straws-in-the-wind-hoops-and-
smoking-guns-what-can-process-tracing-offer-to-impact-
evaluation
Stedman-Bryce (2013) Health
for All: Towards Free Universal
Health Care in Ghana
http://policy-
practice.oxfam.org.uk/publications/effectiveness-
review-health-for-all-towards-free-universal-health-
care-in-ghana-306376
2
3. Process Tracing:
A qualitative method for assessing causal
inference within a single case
Sits in the ‘evaluator’s
toolbox’ alongside
• Theory-based evaluation
• Contribution analysis
• Realist evaluation
3
4. Why process tracing?
0 Useful where pathways of change are uncertain
0Tells you WHY change happened
0 Rigorous approach through ex-post design without a
control group
0Time intensive
0 Requires the outcome to be known
0 Uncharted territory
4
5. A brief history of process-
tracing
0 A method stemming from the social and political sciences.
Major contributors: Alexander George, Andrew Bennett,
David Collier
0 Originally used to explain historical events, e.g. the cause of
the Cuban Missile Crisis (Allison and Zelikow 1999)
0 A ‘contested method’
0 Beach and Pederson’s book (2013): detailed theory and
step-by-step guidance
0 Limited examples of its application in evaluation
NB: See Punton & Welle (2015) for references
5
9. The Universal Health Care
Campaign in Ghana
0 Collaborative advocacy effort by civil society to
promote universal free access to health care
0 Core funding from Oxfam until 2013
0 Lobbying, mobilising, media and research
0 Effectiveness review in 2012-13, using elements of
process tracing
Stedman-Bryce (2013) Effectiveness
Review: Health for All: Towards Free
Universal Health Care in Ghana
http://policy-
practice.oxfam.org.uk/publications/effectiveness-review-
health-for-all-towards-free-universal-health-care-in-
ghana-306376 9
10. A (simple) example of a process
tracing mechanism
Universal
Health Care
campaign
Free
universal
health care
Civil
society
Conduct
coordinated
advocacy
activities for
free
universal
health care
Part 1
Public
Part 2
Become
aware of
the
limitations
of current
health care
financing
Public
Part 3
Demand free
universal
health care
from
government
actors
Govern-
ment
actors
Part 4
Increasingly
support free
universal
health care,
based on
desire for
public
support
Govern-
ment
Part 5
Amend
policies and
processes to
move
towards free
universal
health care
Activity
Entity
10
11. Differences between a mechanism
and a theory of change…
0 No leaps in logic between the parts
0 Every part should be necessary
0 Every part should be observable and empirically
measurable
11
12. Collecting evidence
• Examples of coordinated
advocacy linked to campaign (e.g.
events, platforms, meetings)
• Interview data suggesting that
the campaign increased
knowledge about and interest in
healthcare reform among civil
society actors
• Number of advocacy events held
before the campaign started,
compared to number held during
campaign
Hypothesis: civil society conducts coordinated advocacy for free
healthcare as a result of the Campaign
12
13. Collecting evidence
• Examples of coordinated
advocacy taking place before the
campaigns
• Interviews suggesting that the
campaign made little difference
to activities
Alternative hypothesis: The advocacy campaign made little difference
to the coordination or activities of CSOs
13
14. “We balance probabilities
and choose the most likely. It
is the scientific use of the
imagination.” – The Hound of
the Baskervilles
14
15. Assessing evidence: four tests
• Analogous to a criminal trial
• The evidence for each hypothesis
is weighed….
• …in order to increase the
researcher’s confidence in the
hypothesis.
• Bayesian probability logic
followed to assess the strength of
each part of the causal chain.
• Evidence examined using
concepts of necessity and
sufficiency.
15
16. Assessing evidence: four tests
Straw in the wind test
Evidence: hotel
receipts, suggestive
text messages
- Neither necessary
nor sufficient
Hypothesis: John shot Mary
Hoop test
Evidence: John lacks
an alibi
- Necessary, but not
sufficient
Smoking gun test
Evidence: John was
found holding a gun
over Mary’s body
- Sufficient, but not
necessary
Double decisive test
Evidence: A tamper-
proof CCTV camera
recorded the murder
- Necessary and
sufficient
16
17. Assessing evidence: four tests
Certainty of evidence
Uniquenessofevidence
Low
(evidence is insufficient
for h )
High
(evidence is sufficient
for h )
High
(evidenceis
necessaryforh)
Low
(evidenceisnot
necessaryforh)
Hoop tests Double-
decisive
Straw inthe
wind tests
Smoking
gun tests
17
18. Assessing evidence: four tests
Straw in the wind test
Hypothesis: The Campaign significantly increased the
capacity of member CSOs to plan and work together on
the issue of free health care
Evidence:
1. Campaign members were given the opportunity to
discuss and debate a report on the complex
National Health Insurance Scheme, in order to
increase their understanding on the topic
2. Before the report was published, there was little
coordinated advocacy for free universal healthcare
Increase the plausibility of the hypothesis but do not
prove it or disprove alternative hypotheses. BUT
together they provide stronger evidence than they do
alone 18
19. Assessing evidence: four tests
Hoop test
Hypothesis: The Campaign significantly increased the
capacity of member CSOs to plan and work together on
the issue of free health care
Evidence: Until the campaign started, there was little
notable coordinated advocacy on this issue.
Not enough on its own to prove the hypothesis, BUT
evidence must pass the test in order to keep the
hypothesis under consideration
19
20. Smoking gun test
Hypothesis: The National Health Insurance Authority
in Ghana revised methodology for calculating
membership because of pressure created by the
campaign
Evidence: Ghana delegation at international meeting
stated that the campaign’s report ‘prompted us to revise
our figures’ (despite initially publically dismissing the
report)
Evidence is sufficient to give high confidence that the
hypothesis is true. BUT this type of evidence is hard to
come by.
Assessing evidence: four tests
20
21. The end goal…
0 Researcher can express level of confidence in each
part of the mechanism
0 Mechanism is only as strong as its weakest part
0 If the researcher is confident in the whole
mechanism…
0 …the researcher can express confidence that the
mechanism holds in that particular case: that A
caused B, and why.
21
22. Why process tracing?
0 Useful where pathways of change are uncertain
0 Rigorous approach through ex-post design without a
control group
0Tells you WHY change happened
0Useful in a mixed methods design, or in combination
with e.g. Contribution Analysis?
0 Time intensive
0 Requires the outcome to be known
0Uncharted territory
0 Risk of one weak link causing the whole mechanism to
be rejected
22
23. Questions?
melanie.punton@itad.com
Punton & Welle (2015) Straws-in-the-
wind, Hoops and Smoking Guns: What
can Process Tracing Offer to Impact
Evaluation?
http://www.ids.ac.uk/publication/straws-in-the-wind-hoops-and-
smoking-guns-what-can-process-tracing-offer-to-impact-
evaluation
23
Notas do Editor
PT is a qualitative method in that it takes a qualitative approach to assessing inference – it collects, interprets and weighs up evidence and gives a qualitative rather than quantitative conclusion. BUT both qualitative and quant data can be used to reach these conclusions.
PT is a case based approach. A ‘case’ might be an intervention, a project, a country, an organisation, a process or even an individual.
It’s comparable to other theory based approaches to evaluation – including contribution analysis and realist evaluation
So how does process tracing work? It works by opening up the ‘black box’ between an intervention, and an outcome. What is it about an intervention that causes it to lead to a specific outcome?
This is the fundamental difference between process tracing and experimental and quasi-experimental evaluation designs:
Methods based on counterfactual frameworks (such as RCTs) use control group to isolate the effect of an intervention on the treatment group. Causality inferred through comparison the outcomes of two or more (identical or very similar) cases, in which one received the intervention and the other did not. The black box of the intervention remains unopened
In a generative causal framework, causality is inferred through explaining why an intervention lead to an outcome: generating evidence of what happens in the black box and how that connects the intervention to the outcome. This is the same framework underpinning realist evaluation.
This is why process tracing doesn’t need a control group in order to rigorously attribute change.
So how does process tracing open the ‘black box’?
Inside the black box is the ‘causal mechanism.’ The concept of ‘causal mechanisms’ is central to process tracing.
A causal mechanism explains what it is about intervention A that leads to outcome B - the ‘causal force or power’ that gives rise to the outcome.
The concept of the causal mechanism is central to the generative causal framework that underpins process tracing.
This concept also exists in other fields – for example in realist evaluation. The concept has quite a specific meaning in Beach and Pederson’s version of process tracing….
A causal mechanism in process tracing looks like this. It elaborates all the steps between A (the intervention), and B (the outcome), opening up the black box and breaking down what is inside into the smallest feasible number of PARTS.
Each part consists of ENTITIES conducting ACTIVITIEs. Each part gives rise to the next part in the mechanism.
What does this actually look like in practice?
I’ll now work through an example of what a causal mechanism might look like in practice, using the case study of the Universal Health Care Campaign in Ghana.
This is an example of a causal mechanism based on the Universal Health Care Campaign. A caveat – it’s a simplified example adapted from the Ghana case, designed to show what a mechanism might look like. In actual fact Stedman Bryce’s paper does not consider the full mechanism, because the campaign was still ongoing – so it only tested part of the mechanism.
The whole mechanism is a THEORY about how the intervention resulted in change.
Constructing a mechanism has obvious parallels with developing a theory of change.
Evidence can be qualitative and quantitative.
Usual principles apply – reliability, validity, potential biases, the need for triangulation.
For each part of the mechanism, the researcher should also collect evidence to test alternative hypotheses – what else might have happened, at each point in the chain?
Evidence is necessary to prove a hypothesis if it must exist in order for us to keep the hypothesis under consideration
Evidence is sufficient to prove a hypothesis if it is enough on its own to prove the hypothesis
The researcher constructs these tests based on the evidence they expect to find. Following Bayesian logic, the researcher gives greater weight to evidence that she expects to be less probable, based on previous knowledge of the phenomenon:
The more unlikely it is that you will find a piece of evidence given a specific hypothesis, the more the evidence will increase the probability of the hypothesis being true, if the evidence does in fact exist.
More surprising evidence results in larger increases in our confidence in a hypothesis than less surprising evidence.
Constructing tests is highly analytical and contextual – and there are risks of subjectivity. However, the tests provide a transparent and systematic approach to weighing up qualitative evidence.
Three examples from the Universal Health Care Campaign to show how the tests can be applied in practice.
If the evidence fails to support any one part in the mechanism, the whole mechanism must be rejected. There is insufficient evidence to show that A lead to B in the way specified in the mechanism (although A may have led to B in some other way).