Spatial impacts and sustainability of farm

biogas diffusion in ItalyOriana Gava, Fabio Bartolini and Gianluca Brunori

150th EAAE Seminar‘The spatial dimension in analysing the linkages between agriculture, rural development and the environment’

Department of Agriculture, Food and Environment – University of Pisa (Italy)

Background

Innovating for Sustainable Growth: A Bioeconomy for Europe (EC, 2012)•Environment, resource use, food supply, energy supply; sectors:

1. Agriculture and forestry2. Fisheries and aquaculture3. Bio-based industries4. Food chain

•Biogas is the most consolidated “modern” bioenergy•Disagreed spatial sustainability•Mainly ex ante analyses•Few data for ex post analyses

Rationale

Italian state’s incentives (2009)

• € 0.28 / kWh electricity plugged into the national grid, flat for 15 years• Rated power < 1000 kWh• Auto-produced feedstock ≤ 51%• Purchased feedstock within 70 km

Outcomes (2010)

• Diffusion of farm plantse.g. Carrosio, 2013 – Energy policy;

Chinese et al., 2014 – Energy policy

• Impact on the demand for land and agricultural labour in plants’ neighbourhood => spillover effects

Producing agroenergy belongs to farming activities (2006)

Aim of the study

Estimating the footprint of agricultural biogas

diffusion in Italy

Spatial propensity score analysis

why?

•for drawing causal inference about spatial effects in observational studies

•for strenghtening ex post impact evaluations

Methodology

Idea Expected outcome = Yi Formal expression

Treating i-th observation: T=1 Average treatment effect on the treated

ATT = Yi(T=1)

Comaparing treated with the treated were it not treated: T=0

Average (non)treatment effect on the on the counterfactual

TT = Yi(T=0)

Comparing treated with a measurable proxy of the counterfactual

Average treatment effect on untreated observations comparable to treated ones

TT = Yi(T=0)

Adjusting for potential outcome of no treatment

Average treatment effect ATE = ATT – TT

Potential outcomes model

Neyman (1923) [1990] – Statistical Science 5 (4)Rubin (1974) – Journal of Educational Psychology 76Rubin …

Propensity score method

• Pre-treatment difference variables (x): treated and control systematically diverge

• Propensity score = p(x) is f(x) [logit] that associates each i to its relative probability (Pr) to be among the treated (T=1)

p(x) ≡ Pr(T=1|x)

Rosenbaum & Rubin (1983) – Biometrika 70(1)

Spatial analysis

Potential impact of i on the neighbouring municipality j

Yi = Wyj

•yj = outcome in j-th municipality•Yi =outcome in j-th municipality generated by i, if j shares a border with i

•W = spatial contiguity matrix made of i weightswi,j= 1 if j shares a border with i

wi,j= 0 if i,j share no borders

Anselin, 1988 – Spatial Econometrics: Methods and Models – Kluwer Academic Publishers

Treatments

Treatments: T = 1 Controls: T = 0

T1 = municipality hosts ≥ 1 farm biogas plant

municipality hosts 0 farm biogas plants

T2 = municipality is under T1 AND the plant is fed with livestock waste and dedicated crops only

municipality is under T1 AND the plant is fed with any feedstock

Sustainability indicators

Estimating Y in terms of:

1. Hired labour [# working days]

2. Household labour [# working days]

3. Utilised agricultural area [ha UAA]

4. Number of farms [#]

• Livestock intensity index [LSU / ha]

society

economy

environment

Data

Italian Statistical Institute (ISTAT)• Census of Agriculture 2000 and

2010• Population Census 2001 and

2011

Reserach Centre on Animal Production (CRPA)• Biomass-to-energy census 2010

Number of biogas plantsFeedstockRated power

operating biogas plants in 2010

operating plants fed with livestock waste

operating plants fed with dedicated crops

T2

Category Variable Obs Mean Std.Dev. Min Max

Treatment Municipality with at least one biogas plants 7159 0.02 0.16 - 1.00

Municipality with at least one biogas plant using biomass from livestock or dedicated crops

7159 0.02 0.15 - 1.00

Outcome Difference in hired labour in neighbouring municipalities [# working days] 7159 371.47 25,220 -378,248 322,337.00

Difference in household labour in neighbouring municipalities [# working days] 7159 -58,356 75,284 -779,094 165,042.00

Difference in UAA in neighbouring municipalities [ha] 7159 -325.31 1,986.44 -15,855 23,645.18

Difference in farm number in neighbouring municipalities [#] 7159 -530.93 772.89 -7,974 1,050.00

Difference in livestock density index in neighbouring municipalities [LSU/ha] 7159 -8.98 64.88 -2,100 37.37

Municipality localisation

Municipality located in the mountains 7159 0.53 0.50 - 1.00

Municipality located on the coast 7159 0.08 0.27 - 1.00

Change in population density from 2000 to 2010 7159 16.34 56.24 -967.01 1,082.32

Urban 7159 0.03 0.16 - 1.00

Intermediate urban 7159 0.01 0.11 - 1.00

Urban belt 7159 0.43 0.49 - 1.00

Intermediate 7159 0.31 0.46 - 1.00

Periphery 7159 0.19 0.39 - 1.00

Remote area 7159 0.04 0.19 - 1.00

Central 7159 0.47 0.50 - 1.00

Marginal area 7159 0.53 0.50 - 1.00

Farm features

Average farm size [ha] 7157 0.34 0.24 - 1.00

Farms with arable land [#] 7157 94.57 163.44 - 2,186.00

Livestock size in 2000 [#] 7159 1,210.91 3,046.73 - 81,528.46

Mechanisation intensity [# tractors] 7157 276.35 441.58 1.00 6,458.00

UAA [ha] 7157 1,264.94 1,856.05 - 27,776.93

UAA rented [ha] 7157 99.73 281.13 - 7,106.20

Y = 2010 - 2000Treatmen

t ATEStandard

Error z P>zConfidence Interval

95%# working days from hired labour / year

T1 1268.79 212.21 5.98 0 852.84 1684.73T2 931.19 343.53 2.71 0.007 257.88 1604.51

# working days from household labour /

year

T1 -542.86 1740.52 -0.31 0.755 -3954.24 2868.50

T2 -1177.11 1235.36 -0.95 0.341 -3598.39 1244.16

ha UAA / year

T1 39.57 69.35 0.57 0.568 -96.34 175.50T2 17.40 89.25 0.19 0.845 -157.54 192.34

# farms / year

T1 -47.27 16.69 -2.83 0.005 -79.99 -14.55T2 -46.58 16.60 -2.81 0.005 -79.13 -14.04

LSU / ha / year

T1 0.52 0.23 2.19 0.028 0.05 0.98T2 0.46 0.255881 1.83 0.068 -0.03 0.96

Results – Average effects of treatments T1, T2

Y = 2010 - 2000Treatmen

t ATEStandard

Error z P>zConfidence Interval

95%

# working days from hired labour / year

T1 1268.79 212.21 5.98 0 852.84 1684.73T2 931.19 343.53 2.71 0.007 257.88 1604.51

# working days from household labour /

year

T1 -542.86 1740.52 -0.31 0.755 -3954.24 2868.50

T2 -1177.11 1235.36 -0.95 0.341 -3598.39 1244.16

Average effects of treatments T1, T2

Y = 2010 - 2000Treatmen

t ATEStandard

Error z P>zConfidence Interval

95%

ha UAA / year

T1 39.57 69.35 0.57 0.568 -96.34 175.50T2 17.40 89.25 0.19 0.845 -157.54 192.34

# farms / year

T1 -47.27 16.69 -2.83 0.005 -79.99 -14.55T2 -46.58 16.60 -2.81 0.005 -79.13 -14.04

Average effects of treatments T1, T2

Y = 2010 - 2000Treatmen

t ATEStandard

Error z P>zConfidence Interval

95%

LSU / ha / year

T1 0.52 0.23 2.19 0.028 0.05 0.98T2 0.46 0.255881 1.83 0.068 -0.03 0.96

Average effects of treatments T1, T2

Discussion\1

• Biogas plant distribution is spatially uneven• Biogas diffusion has a marked impact on rural economy• Sustainability trade-offs:

+positive effects on income and job availability in rural areas- increased environmental pressure: agricultural intensification

and marginalisation of small farms• Legal constraints affect feedstock sourcing area and transport

costs

Discussion\2

What is missing How we suggest to improveNon-rural drivers of rural areaViability: sector mobility, off-farm income, availability of infrastructures

Supplementing the dataset

Spatial modulation of the treatment Generalised propensity score and dose-response model

Further research

Thank you!

oriana.gava@for.unipi.it

This research was supported the by the EU 7th Framework Program, grant No. 609448

IMPRESA – The Impact of Research on EU Agriculture http://www.impresa-project.eu/about.html