Accountability and Information 1

Final Project Essentials: Fixed Effects

Author

Affiliation

Jeremy Springman

University of Pennsylvania

Logistics

Assignments

• Today
  • Dunning et al. (2019)
  • Data Assignment 1 (due by midnight)
• Tuesday (3/28)
  • Foreign aid readings
  • Github tutorial (maybe)
• April 9
  • Final project proposal

Agenda

1. Dunning et al. (2019)
2. FINAL questions about Data Assignment 1
3. Final Project Essentials: Fixed Effects

Dunning et al. (2019)

Barriers hindering accumulation

What is accumulation?

• Building knowledge across studies

What makes it difficult?

• Limited replication
• Heterogeneity of design and measurement
• Publication bias

Barriers hindering accumulation

Facilitating accumulation

What solutions are discussed by the authors?

• Pre-registration
• Harmonizing theory, measurement, and estimation
• Publication of null results

Facilitating accumulation

Models of political accountability

Accountability requires that voters:

• Observe credible signal of performance
  • Attribution
  • Benchmarking
• Update their beliefs
• Have credible alternatives

Models of political accountability

What does the literature say?

• Theory is mixed
  • Partisan and sectarian attachments are strong
  • Voters may be reluctant to update their beliefs
• Experimental results are mixed
  • Demobilization
  • Limited recall
  • Ephemeral effects

Research Design

Intervention

• Information on political performance
  • Legislative behavior
  • Spending irregularities
  • Budget allocation
  • Candidate experience

Research Design

Accountability requires that voters:

• Observe credible signal of performance
  • Attribution
  • Benchmarking
• Update their beliefs
  • Good news: $Q>P\mid 1(P=Q,Q>med(Q))$
• Have credible alternatives

Ecological Validity

• How is information disseminated?
• What is the real world activity being tested?

Describing Information

• good news group is above the line
• correlation is positive but weak

Problems hindering accumulation

Research Design

Core hypotheses:

1. Good news increases voter support for incumbents
2. Bad news decreases voter support for incumbents
3. Effect of information will increase with gap between Q and P
4. Strongest for nonpartisan and non-coethnic voters

Findings

Why the null results?

• Source credibility?
  • No…
• Lack of retention
• Lack of updating on politician performance
• Lack of updating about politician quality
• Lack of coordination (private treatment)

Statistical Power

Null results were not “foregone conclusion”?

• 80% power
• Change the vote of 5/100 voters
• Change turnout of 4/100 voters

Findings

Findings

Findings

Policy Implications

• Much of the work funded by donors is probably not having an impact on accountability
• Targeting of information provision needs to be rethought
  • Public dissemination and coordination

Data Assignment Questions

Final Project Essentials

Unit Fixed Effects

What is a unit fixed effect?

• Simple linear models make comparisons both across and within units (when there are multiple measures)
• Fixed effects ‘absorb’ variation across units; now coefficients estimate only the variation within units
• This ‘controls’ for time-invariant heterogeneity

What are they used for?

• Reducing concerns about confounding
• Difference-in-differences

Unit Fixed Effects

Normal model

$$Y_{i}t=\alpha +{\beta }_1{X}_{it}+{ϵ}_{it}$$ Unit fixed Effects model

$$Y_{i}t={\gamma }_i+{\beta }_1{X}_{it}+{ϵ}_{it}$$

Example: Unit FEs with one binary predictor

• $\alpha$: $E[Y_i]$ when $X_{i1}=0$
• ${\gamma }_i$: $E[Y_i]$ for every unit when $X_{i1}=0$

Unit Fixed Effects

Unit fixed Effects model

$$Y_{i}t={\gamma }_i+{\beta }_1{X}_{it}+{ϵ}_{it}$$

**How does the interpretation of ${\beta }_1$ change?

• ${\gamma }_i$: $E[Y_i]$ when $X_{it}=0$
• ${\beta }_1$ Change in $E[Y_i]$ when $X_{it}=1$
• Each ${\gamma }_i$ will absorb everything that doesn’t vary over time in that unit

Unit Fixed Effects

Unit Fixed Effects

Attaching package: 'ggdag'

The following object is masked from 'package:stats':

    filter