Long-Term, Heterogeneous Treatment Effects from Non-Pecuniary Environmental Programs: A Large-Scale Field Experiment

1
Long-Term, Heterogeneous Treatment Effects from
Non-Pecuniary Environmental Programs A
Large-Scale Field Experiment

• Paul J. Ferraro
• Department of Economics
• Georgia State University
• Juan Jose Miranda
• Department of Economics
• Georgia State University

2

3
Information Campaigns

1. Info on behavioral and technological
   modifications
2. Request users to voluntarily change their
   behavior for the public good
3. Provide social comparisons to induce individuals
   to conform to a social norm.

4
Field Experiment

• Water system in Metro Atlanta with Ferraro,
  implemented targeted, residential information
  campaign as randomized experiment.
• Three treatments sent via first class mail in
  May 2007.
• Ferraro and Price focus on short-term, mean
  treatment effects in summer 2007.

5
Ferraro and Miranda

• Longer-term impacts (2008, 2009).
• Heterogeneous Treatment Effects
• Relevant for policy and practice
• Relevant for understanding mechanisms

6
Treatment 1 Tip Sheet
Information on behavioral changes and
technologies that can reduce water consumption
7
Treatment 2 Weak Social Norms (standard message)
Tip sheet Civic Duty Language
8
Treatment 3 Strong Social Norms
Tip sheet Letter with Civic Duty Language
Social Comparison
9
Social Comparison
Your own total consumption June to October 2006
52,000 gallons   Your neighbors average
(median) consumption June to October
2006 35,000 gallons   You consumed more water
than 73 of your Cobb County neighbors.
10
Reduction in Summer 07 Consumption 95 CI Number of Homes
Control -- 71,643
Treat 1 (Tip) - 0.7 -1.7, 0.4 11,675
Treat 2 (Weak Norms) - 2.7 -3.7, -1.8 11,675
Treat 3 (Strong Norms) - 4.8 -5.7, -3.9 11,676
11
Persistence of Treatment Effects
Summer 2008 Summer 2009
Treat 1 (Tip) 0 0
Treat 2 (Weak Norms) 0 0
Treat 3 (Strong Norms) -2.5 (plt0.01) -1.3 (plt0.05)
12
Heterogeneous Treatment Effect
Nonparametric Tests (Crump et al. 2008) Test
Null of Zero Conditional Average Treatment Effects
Cannot reject that Treatment 1 effect is zero for
all subgroups.
13
Heterogeneous Treatment Effect
Nonparametric Tests (Crump et al. 2008) Test
Null of Constant Conditional Average Treatment
Effects
Reject null that Treat 2 (plt0.10) and Treat 3
(plt0.01) have constant treatment effects
14
Treatment 1 (tips)
Treatment 2 (weak norm)
Treatment 3 (strong norm)
15
Subgroup Analysis
Household Wealth (market value of house), age of
home, ownership
Neighborhood characteristics Education, race,
ownership
16
Treatment 3 (2007) by Subgroups

• More responsive wealthy households, owners and
  households in neighborhoods with more white and
  educated householders and fewer renters.
• No difference conditional on year in which home
  was built.

17
Nonpecuniary vs Pecuniary

• Strong social norm message
• 2007 effect equivalent to average price increase
  of 12-15.
• Immediately detectable one month later.
• High-income households
• Most responsive to norms
• Least responsive to price (Mansur Olmstead
  2007)
• Effect declines over time.

18
Mechanisms

• Evidence consistent with behavioral changes with
  recurring costs rather than one-time fixed-cost
  investments. Mostly from outdoor use.
• Evidence consistent with interpretation that
  social comparison operates through social norms
  rather than private efficiency signals to a
  boundedly rational agent.

19
Choosing Among Non-experimental Estimators
Design-replication study Form a
non-experimental comparison group from households
in neighboring counties. Then use
best-practice econometric methods (regression,
matching, panel data, regression discontinuity)
to estimate treatment effect. Contrast
non-experimental estimates to experimental
estimates.