Bayesian approaches to integrated qualitative and quantitative data (with Alan Jacobs)

• Open access version
• CausalQueries Package
• WZB CausalQUeries Shiny app

• Qualitative and quantitative inference: Malaria example
• The key idea behind the formalization of process tracing is: process tracing is submission of case level data to a causal model
  
• Key idea of mixing: Model updating is submission of population data to a more fundamental model
• Both types of data can be simultaneously applied

See APSR: Humphreys and Jacobs (2015)

• We can go a lot deeper if we marry:

  • Potential outcomes framework
  • Graphical causal models
  • Bayesian methods

• Approach:

  • Write down structural causal model (DAG)
  • Figure out parameters needed to specify all possible probability distributions over primary strata
  • Update with data
  • Query model using do calculus

Package with Till Tietz, Lily Medina, Georgiy Syunyaev

Replication of Chickering and Pearl (1996): interest in average causal effect with imperfect compliance.

model <- make_model("Z -> X -> Y; X <-> Y") 

plot(model)

model |>
  update_model(lipids) |>
  query_model(query = "Y[X=1] - Y[X=0]",
              given = c("All",  "X==0 & Y==0", "X[Z=1] > X[Z=0]"),
              using = "posteriors") 

Remark: See promising avenues here for more historical analysis and also for mixed methods impact evaluation. Qualitative research has been systematically undervalued in many social sciences and there is much more scope for integration than often presumed.

• Happy to do an ad hoc seminar on graphical causal models and causal querying
• Interested in applications also, e.g. in historical analysis, mechanisms analysis, mixed methods analysis

Formal design declarations for transparency and diagnostics

• Open access version
• DeclareDesign Package
• WZB DeclareDesign Shiny app

Very wide class of research designs involve choices about four things: Model, Inquiry, Data strategy, Answer strategy

If you are able and willing to specify these in advance then your design becomes diagnosable:

• Is it unbiased?
• Is it powered?
• Are inferences valid?

Remark: Remarkable lack of formality about how to specify and evaluate research designs

The dream: that this gets used to facilitate design development, pre-registration, replication, and funding decisions.

TCD: Happy to do a seminar at some stage on design declaration at diagnosis

• Older social choice theoretic work
• Newer more structural work

Political violence and endogenous growth Humphreys (2022)

Endogenous growth model where factor productivity depends on public goods production. Public goods can be produced via cooperative or coercive technologies.

• Political Salience and Regime Resilience Schweighofer-Kodritsch, Huck, and Humphreys (2023)

Conditions under which increased attention to politics can stabilize or threaten existing systems.

• Power: Connecting causal inference with non-cooperative power indices

• Human shields: strategic logics for when fighting groups mingle among civilians

Increasingly interested in combining formal models with experimental data.

Currently using Bayesian methods and MLE

• Fearon, Humphreys, et al. (2018) Van der Windt et al. (2019) Wilke and Humphreys (2020)
• Understanding intersectional discrimination with Kasara and Foong
• Understanding political inequality with Bosancianu and Garcia
• Understanding bureaucratic incentives with Voors, Prato, Jaiteh, Windt

Keen to learn more about this

Remark: The “credibility revolution” has resulted in theory taking too much of a backseat. But social science is really about understanding why things work.

Structural approaches are a promising but dangerous approach: let’s proceed with caution.

Generally involve partnerships to figure out what works or how things work via randomized assignment.

Generally done in partnership with various groups:

• Local and national governments
• Local NGOS
• Governmental or non governmental aid organizations
• International Organizations

This is a real area of strength at TCD and at TIME especially.

• Political communication interventions in Uganda with NDI (Grossman, Humphreys, and Sacramone-Lutz 2020)
• Accountability interventions with AFLI (Uganda NGO) on parliamentary behavior
• “Metaketa” studying accountability interventions across 7 countries with EGAP and others (Dunning et al. 2019)
• New: Sierra Leone study on bureaucratic incentives facing Community Animal Health Workers with Ministry of Agriculture

Remark: Received wisdom seems naive. Responsiveness to information weak. Economic rationales often more compelling.

Community Driven Reconstruction studies seeking to assess the effects of development interventions on social cohesion and the quality of local governance

• With IRC in Liberia (Fearon, Humphreys, and Weinstein 2015)
• With IRC and CARE in DR Congo (Humphreys, Sierra, and Van der Windt 2019)
• With the World Bank in Indonesia (quasi experimental)
• Now: With GIZ in Lebanon

Remark: Very pessimistic about scope for outside interests to alter local governance in meaningful ways.

• S{~a}o Tom{'e} and Pr{'}ncipe forum experiment with UNDP and government of S{~a}o Tom{'e} and Pr{'}ncipe (Humphreys, Masters, and Sandbu 2006)
• Kampala experiments with KCCA (city of Kampala)

Remark: Conditions certainly exist for meaningful vertical communication. Irish experiments here look very impressive.

Remark: The knowledge base supporting many types of development interventions is very weak. There are large investments in unproven strategies. Many interventions do little and outsiders cannot (and often should not) by trying to change structures from the outside.

Most development is not driven by aid but by endogenous processes and we should be doing more to understand these.

• Older lab-in-the-field work on ethnic bias in Uganda
• Field experiment on housing discrimination in NYC with city government
• Field experiment on violence against women in India with DFID
• New: Field experiment on trans-Saharan migration from Nigeria

New lab-in-the-field project with Kasara and Foong in Kenya

An older theme for me:

• Causes of civil wars Humphreys and Ag Mohamed (2005)
• Violence during war (Humphreys and Weinstein 2006)
• Violence measurement (Van der Windt and Humphreys 2016)
• Violence against women
• Post conflict development Humphreys, Sierra, and Van der Windt (2019)

Remark:

Discrimination is very widespread; experimental approaches well suited to documenting it; growing evidence showing discrimination is affected by institutional and informational context as well as social contact.

Political inequality is poorly understood and a focus on micro political processes is not sufficient to address it.

Possible themes here for capstone course in either economics or political science

• Vaccine attitudes (Solı́s Arce et al. 2021)
• Attitudes to vaccine sharing (Geissler et al. 2022)
• Vaccine encouragements (Klüver et al. 2021)
• Economic pain (Egger et al. 2021,)
• Tradeoffs between liberal rights and welfare Hartmann et al. (2023),