Is no net loss possible?
Evaluating policy instruments for reducing
deforestation with a growing economy
Megan C Evans1, Grace Chiu2, Andrew K Macintosh3, Philip Gibbons1
1 Fenner School of Environment and Society, The Australian National University, Canberra
2 Research School of Finance, Actuarial Studies and Applied Statistics,
The Australian National University, Canberra
3 ANU Centre for Climate Law and Policy, The Australian National University, Canberra
Symposium: Conservation trade-offs in a resource-limited world
ICCB-ECCB 2015, August 2-6 2015, Montpellier - France
@megcevans
Trading off the environment
Image: Ben Phalan
Map: Lorenzo Catterino
Graph: Martina di Fonzo
Map: Roman Carrasco
Deforestation
• 2.3 million km2 forest lost 2000-2012
Hansen et al. 2013 Science
• 170 million ha could be lost 2010-2030
• Eastern Australia a global deforestation hotspot 1980-2000 (Lepers et al.
2005) and over next two decades (WWF 2015)
WWF International, 2015.
Living Forests Report: Saving Forests at Risk.
Australian policy context
Native
Vegetation
Regulation
2005 (NSW)
Native Vegetation Retention (NVR)
controls under the Planning and
Environment Act 1987 (Vic)
Vegetation
Management Act 1999
Native Vegetation Act
1991 (SA)
1987: Soil and Land
Conservation Act
(SALCA) 1945
Vegetation
Management and
Other Legislation Bill
2004
“Land clearing in Queensland triples after
policy ping pong” Maron et al. 2015 The Conversation
What was the
policy impact?
Evaluating deforestation policy instruments
• Extensive work on drivers of deforestation
• Many cross-country, sometimes spatial, rarely temporal
FAO. State of the World’s Forests 2012.
Ewers et al 2008 Env Cons
Kuznet’s curve
plot
Laurence et al. 2002 J BiogeographyRudel et al. 2005 Global Env Change
Evaluating deforestation policy instruments
• Few rigorous evaluations of policy impact
• C.f protected area effectiveness
FAO. State of the World’s Forests 2012.
Evaluating deforestation policy instruments
Miteva et al. 2012
Oxford Review of Economic Policy
Evaluating deforestation policy instruments
• Few rigorous evaluations of policy impact
• C.f protected area effectiveness
• Very few evaluate impact of policy timing
• Arima et al. 2014 Land Use Policy
FAO. State of the World’s Forests 2012.
Evaluating deforestation policy instruments
FAO. State of the World’s Forests 2012.
Arima et al. 2014 Land Use Policy
Evaluating deforestation policy instruments
Australian case study
• Developed, wealthy nation with
strong governance and high
institutional capacity – we would
expect policies to be effective
• Forest transitions theory; Kuznet’s curve
hypothesis
• Many candidate policies for
evaluation
• High resolution spatial data now
available
This study
1. Spatially quantify forest loss across Australia from
1972-2011
2. Develop a credible model of deforestation at
national/regional scale
3. Evaluate the impact of deforestation policy instruments
after removing macroeconomic and climatic effects
Deforestation imagery
Source: Australian Greenhouse Emissions Information System (AGEIS)
• Deforestation events
attributed to human
intervention
• Woody vegetation only
(“Kyoto forest”)
• 20 epochs, 1972-2011
• 19 tiles
• ~ 380 tiles total
Deforestation in Australia: Theory of change
Deforestation
Sources of deforestation
Agents and their land use decisions
Underlying (ultimate) causes
Macroeconomic variables
and policy instruments
Immediate (proximate) causes
Factors that influence agent decisions
(institutions, infrastructure, markets, technology)
Angelsen & Kaimowitz 1999
World Bank Res Obs
Deforestation in Australia: Theory of change
Deforestation
Sources of deforestation
Agents and their land use decisions
Immediate (proximate) causes
Factors that influence agent decisions
(institutions, infrastructure, markets, technology)
Angelsen & Kaimowitz 1999
World Bank Res Obs
Terms of
trade
Farmer
terms of
trade
Real
GDP
Policy
instruments
Deforestation in Australia: Theory of change
Deforestation
Sources of deforestation
Agents and their land use decisions
Angelsen & Kaimowitz 1999
World Bank Res Obs
Terms of
trade
Farmer
terms of
trade
Commodity
prices
Real
GDP
Rainfall
Forest
remaining
Policy
instruments
Deforestation in Australia: Theory of change
Deforestation
Angelsen & Kaimowitz 1999
World Bank Res Obs
Terms of
trade
Farmer
terms of
trade
Commodity
prices
Real
GDP
Policy
instruments
Rainfall
AgricultureMining Urban
Forest
remaining
Deforestation in Australia: Theory of change
Deforestation
Angelsen & Kaimowitz 1999
World Bank Res Obs
Terms of
trade
Farmer
terms of
trade
Commodity
prices
Real
GDP
Policy
instruments
Rainfall
Agriculture
Forest
remaining
Spatial analysis
Annual rainfall
0.01deg
Land use
Local Government Areas (LGAs)
n=564
Human induced deforestation
1972-2011
0.0002deg (25m)
Analysed using R
raster package
Spatial analysis
Annual rainfall
0.01deg
Land use
Local Government Areas (LGAs)
n=564
Human induced deforestation
1972-2011
0.0002deg (25m)
Analysed using ARGGH
raster package
Statistical approach
𝑦𝑡𝑠 = 𝑓𝜃 𝒙 𝐸𝑡 , 𝒙 𝐶𝑡𝑠 + 𝜀𝑡𝑠
𝑦𝑡𝑠where is deforestation at time t in LGA s
relative to amount of primary forest remaining
Macroeconomic variables at time t (national scale)
GDP, ToT, FTOT, CPI, inflation, agric commodity prices
Climatic variables at time t and LGA s
annual rainfall, rainfall in driest month
Crude statistical approach
Piecewise linear regression
• aka segmented or broken-stick
regression, interrupted time series
• Used to evaluate impact of policy
timing when difficult to isolate a
control (counterfactual)
• pharmaceutical health , medicine,
ecotoxicology, political science
Crude statistical approach
Piecewise linear regression
• aka segmented or broken-stick
regression, interrupted time series
• Used to evaluate impact of policy
timing when difficult to isolate a
control (counterfactual)
• pharmaceutical health , medicine,
ecotoxicology, political science
Crude statistical approach
Piecewise linear regression
• aka segmented or broken-stick
regression, interrupted time series
• Used to evaluate impact of policy
timing when difficult to isolate a
control (counterfactual)
• pharmaceutical health , medicine,
ecotoxicology, political science
Improved statistical approach
Bent-cable regression model
• Generalisation of piecewise regression
Chiu et al. 2006 J. Amer. Stat. Assoc.
• More flexible model for describing
system change
• Greater capacity to capture potential
effect of regulation
Improved statistical approach
Bent-cable regression model
• Generalisation of piecewise regression
Chiu et al. 2006 J. Amer. Stat. Assoc.
• More flexible model for describing
system change
• Greater capacity to capture potential
effect of regulation
Improved statistical approach
Bent-cable regression model
Khan et al. 2012 Int J Statistics and Probability
• Generalisation of piecewise regression
Chiu et al. 2006 J. Amer. Stat. Assoc.
• More flexible model for describing
system change
• Greater capacity to capture potential
effect of regulation
Results from exploratory data analysis
Total deforestation (National)
Results from exploratory data analysis
Total deforestation (by State)
Evaluation challenges
• Unable to isolate control or synthetic control
• Data resolution (spatial and temporal)
• Few data points relative to number of interventions; some missing
years
• Data aggregated at State level too sparse for detecting policy
impact
• Try to overcome with
i) spatially disaggregated data points (n=564 LGAs)
ii) bent-cable model
(still might not work!)
Deforestation over time (n = 156 LGAs)
Queenslandonly
Crude linear model residuals, temporally detrended
Queenslandonly
2000: Control of
remnant vege
2006: End of
broadscale
clearing
2009:Regrowth
protections
Crude linear model residuals, temporally detrended
Queenslandonly
Crude linear model residuals, temporally detrended
Chiu et al. 2006 J. Amer. Stat. Assoc.
Next steps
• Develop holistic model (similar to Khan et al. 2009; 2012)
and apply to LGA data
• “holistic” in that all n=564 LGAs are modelled collectively to
maximize statistical power of a single model
• Obtain most recent data to try and detect recent
relaxation of laws in Queensland and NSW
Concluding comments
• Evaluation is hard!
• Detection of policy impacts can still be difficult
with fancy methods and high resolution data
• Answering “what” is the impact won’t necessarily
tell you “why”
 need for qualitative data & clear theory of change
Thank-you
P.S – come to SCB
Oceania 2016!
@megcevans
mcevansresearch.wordpress.com

Is no net loss possible? Evaluating policy instruments for reducing deforestation with a growing economy

  • 1.
    Is no netloss possible? Evaluating policy instruments for reducing deforestation with a growing economy Megan C Evans1, Grace Chiu2, Andrew K Macintosh3, Philip Gibbons1 1 Fenner School of Environment and Society, The Australian National University, Canberra 2 Research School of Finance, Actuarial Studies and Applied Statistics, The Australian National University, Canberra 3 ANU Centre for Climate Law and Policy, The Australian National University, Canberra Symposium: Conservation trade-offs in a resource-limited world ICCB-ECCB 2015, August 2-6 2015, Montpellier - France @megcevans
  • 2.
    Trading off theenvironment Image: Ben Phalan Map: Lorenzo Catterino Graph: Martina di Fonzo Map: Roman Carrasco
  • 3.
    Deforestation • 2.3 millionkm2 forest lost 2000-2012 Hansen et al. 2013 Science
  • 4.
    • 170 millionha could be lost 2010-2030 • Eastern Australia a global deforestation hotspot 1980-2000 (Lepers et al. 2005) and over next two decades (WWF 2015) WWF International, 2015. Living Forests Report: Saving Forests at Risk.
  • 5.
    Australian policy context Native Vegetation Regulation 2005(NSW) Native Vegetation Retention (NVR) controls under the Planning and Environment Act 1987 (Vic) Vegetation Management Act 1999 Native Vegetation Act 1991 (SA) 1987: Soil and Land Conservation Act (SALCA) 1945 Vegetation Management and Other Legislation Bill 2004
  • 6.
    “Land clearing inQueensland triples after policy ping pong” Maron et al. 2015 The Conversation What was the policy impact?
  • 7.
    Evaluating deforestation policyinstruments • Extensive work on drivers of deforestation • Many cross-country, sometimes spatial, rarely temporal FAO. State of the World’s Forests 2012. Ewers et al 2008 Env Cons Kuznet’s curve plot Laurence et al. 2002 J BiogeographyRudel et al. 2005 Global Env Change
  • 8.
    Evaluating deforestation policyinstruments • Few rigorous evaluations of policy impact • C.f protected area effectiveness FAO. State of the World’s Forests 2012.
  • 9.
    Evaluating deforestation policyinstruments Miteva et al. 2012 Oxford Review of Economic Policy
  • 10.
    Evaluating deforestation policyinstruments • Few rigorous evaluations of policy impact • C.f protected area effectiveness • Very few evaluate impact of policy timing • Arima et al. 2014 Land Use Policy FAO. State of the World’s Forests 2012.
  • 11.
    Evaluating deforestation policyinstruments FAO. State of the World’s Forests 2012. Arima et al. 2014 Land Use Policy
  • 12.
    Evaluating deforestation policyinstruments Australian case study • Developed, wealthy nation with strong governance and high institutional capacity – we would expect policies to be effective • Forest transitions theory; Kuznet’s curve hypothesis • Many candidate policies for evaluation • High resolution spatial data now available
  • 13.
    This study 1. Spatiallyquantify forest loss across Australia from 1972-2011 2. Develop a credible model of deforestation at national/regional scale 3. Evaluate the impact of deforestation policy instruments after removing macroeconomic and climatic effects
  • 14.
    Deforestation imagery Source: AustralianGreenhouse Emissions Information System (AGEIS) • Deforestation events attributed to human intervention • Woody vegetation only (“Kyoto forest”) • 20 epochs, 1972-2011 • 19 tiles • ~ 380 tiles total
  • 15.
    Deforestation in Australia:Theory of change Deforestation Sources of deforestation Agents and their land use decisions Underlying (ultimate) causes Macroeconomic variables and policy instruments Immediate (proximate) causes Factors that influence agent decisions (institutions, infrastructure, markets, technology) Angelsen & Kaimowitz 1999 World Bank Res Obs
  • 16.
    Deforestation in Australia:Theory of change Deforestation Sources of deforestation Agents and their land use decisions Immediate (proximate) causes Factors that influence agent decisions (institutions, infrastructure, markets, technology) Angelsen & Kaimowitz 1999 World Bank Res Obs Terms of trade Farmer terms of trade Real GDP Policy instruments
  • 17.
    Deforestation in Australia:Theory of change Deforestation Sources of deforestation Agents and their land use decisions Angelsen & Kaimowitz 1999 World Bank Res Obs Terms of trade Farmer terms of trade Commodity prices Real GDP Rainfall Forest remaining Policy instruments
  • 18.
    Deforestation in Australia:Theory of change Deforestation Angelsen & Kaimowitz 1999 World Bank Res Obs Terms of trade Farmer terms of trade Commodity prices Real GDP Policy instruments Rainfall AgricultureMining Urban Forest remaining
  • 19.
    Deforestation in Australia:Theory of change Deforestation Angelsen & Kaimowitz 1999 World Bank Res Obs Terms of trade Farmer terms of trade Commodity prices Real GDP Policy instruments Rainfall Agriculture Forest remaining
  • 20.
    Spatial analysis Annual rainfall 0.01deg Landuse Local Government Areas (LGAs) n=564 Human induced deforestation 1972-2011 0.0002deg (25m) Analysed using R raster package
  • 21.
    Spatial analysis Annual rainfall 0.01deg Landuse Local Government Areas (LGAs) n=564 Human induced deforestation 1972-2011 0.0002deg (25m) Analysed using ARGGH raster package
  • 22.
    Statistical approach 𝑦𝑡𝑠 =𝑓𝜃 𝒙 𝐸𝑡 , 𝒙 𝐶𝑡𝑠 + 𝜀𝑡𝑠 𝑦𝑡𝑠where is deforestation at time t in LGA s relative to amount of primary forest remaining Macroeconomic variables at time t (national scale) GDP, ToT, FTOT, CPI, inflation, agric commodity prices Climatic variables at time t and LGA s annual rainfall, rainfall in driest month
  • 23.
    Crude statistical approach Piecewiselinear regression • aka segmented or broken-stick regression, interrupted time series • Used to evaluate impact of policy timing when difficult to isolate a control (counterfactual) • pharmaceutical health , medicine, ecotoxicology, political science
  • 24.
    Crude statistical approach Piecewiselinear regression • aka segmented or broken-stick regression, interrupted time series • Used to evaluate impact of policy timing when difficult to isolate a control (counterfactual) • pharmaceutical health , medicine, ecotoxicology, political science
  • 25.
    Crude statistical approach Piecewiselinear regression • aka segmented or broken-stick regression, interrupted time series • Used to evaluate impact of policy timing when difficult to isolate a control (counterfactual) • pharmaceutical health , medicine, ecotoxicology, political science
  • 26.
    Improved statistical approach Bent-cableregression model • Generalisation of piecewise regression Chiu et al. 2006 J. Amer. Stat. Assoc. • More flexible model for describing system change • Greater capacity to capture potential effect of regulation
  • 27.
    Improved statistical approach Bent-cableregression model • Generalisation of piecewise regression Chiu et al. 2006 J. Amer. Stat. Assoc. • More flexible model for describing system change • Greater capacity to capture potential effect of regulation
  • 28.
    Improved statistical approach Bent-cableregression model Khan et al. 2012 Int J Statistics and Probability • Generalisation of piecewise regression Chiu et al. 2006 J. Amer. Stat. Assoc. • More flexible model for describing system change • Greater capacity to capture potential effect of regulation
  • 29.
    Results from exploratorydata analysis Total deforestation (National)
  • 30.
    Results from exploratorydata analysis Total deforestation (by State)
  • 31.
    Evaluation challenges • Unableto isolate control or synthetic control • Data resolution (spatial and temporal) • Few data points relative to number of interventions; some missing years • Data aggregated at State level too sparse for detecting policy impact • Try to overcome with i) spatially disaggregated data points (n=564 LGAs) ii) bent-cable model (still might not work!)
  • 32.
    Deforestation over time(n = 156 LGAs) Queenslandonly
  • 33.
    Crude linear modelresiduals, temporally detrended Queenslandonly
  • 34.
    2000: Control of remnantvege 2006: End of broadscale clearing 2009:Regrowth protections Crude linear model residuals, temporally detrended Queenslandonly
  • 35.
    Crude linear modelresiduals, temporally detrended Chiu et al. 2006 J. Amer. Stat. Assoc.
  • 36.
    Next steps • Developholistic model (similar to Khan et al. 2009; 2012) and apply to LGA data • “holistic” in that all n=564 LGAs are modelled collectively to maximize statistical power of a single model • Obtain most recent data to try and detect recent relaxation of laws in Queensland and NSW
  • 37.
    Concluding comments • Evaluationis hard! • Detection of policy impacts can still be difficult with fancy methods and high resolution data • Answering “what” is the impact won’t necessarily tell you “why”  need for qualitative data & clear theory of change
  • 38.
    Thank-you P.S – cometo SCB Oceania 2016! @megcevans mcevansresearch.wordpress.com