Measuring the impacts of conservation tillage on household income and consumption: A Syrian case

Transcription

1 Measuring the impacts of conservation tillage on household income and consumption: A Syrian case Tamer El-Shater 1, Yigezu A. Yigezu 1*, Amin Mugera 2, Colin Piggin 1, Atef Haddad 1, Yaseen Khalil 1, Stephen Loss 1 1 International Center for Agricultural Research in the Dry Areas (ICARDA) 2 University of Western Australia 6 th World Congress on Conservation Agriculture Winnipeg, Manitoba June, 2014

2 I. Background CT has environmental benefits: Conserves soil moisture Reduces wind and water erosion Enhances soil carbon sequestration CT also reduces costs: Fuel, labor, and machinery (Ribera et al., 2004). A need to evaluate its impacts: Yields Net returns Yield and income variability dry areas

3 Background cont d 9% of global cropland in 2012 under CA Often with only some of the components of the CA package CA little known in Syria until 2005 ACIAR-AusAid funded project promoted CT since 2005 In 2010/11 total area under CT reached at least 15,000 ha Adoption rate > 8%

4 II. Objectives of the study Measure impacts of conservation tillage (CT) on: Household income Consumption Among Syrian wheat farms Contribute to the literature Empirical evidences on economic benefits of CT. Provide useful information to: Policy makers, extension and development agents, researchers Syria and similar agro-climatic and production systems.

5 III. Data Farm survey conducted by ICARDA in 2011 Cluster sampling Power analysis (Cohen 1988) A sample of 820 farms: 500 random sample 320 purposive 28 villages, 17 districts, 7 governorates Only 621 wheat producers in the sample Analysis focuses only on wheat producers.

6 Characterization of sample farms Variables Average values for the entire sample of 621 (and random sample of 500) farmers Unit Adopters Non-Adopters Total Number of farmers Number 249 (15) 372 (293) 621(308) Average farming experience of household head Years Average education level of household heads Years Proportion of farmers with salinity affected soil % Average time since farmer started using CT Years Proportion of farmers who are in zone one % Proportion of farmers who hosted demonstration trials % Proportion of farmers who participated on field days % Proportion of farmers who participated in field days % and hosted trials Total area cultivated (average) Hectare Total wheat area cultivated (average) Hectare Proportion of farmers who know the CT technology % Average distance to the nearest input market (Km) km Average value of total assets(million SP) SP What is the total number of plots Number Average length of time since the farmer first heard Year about the CT technology Area under CT (Ha) 15.2 (8.6) 0 (0) 6.1(0.32)

7 Income IV. Methodology Selection bias: a major problem in impact assessment Income Income Observed of income a participant of participant (observed) in PVS ATT Income Outcome if if a participant had not participated had not in participated PVS (unobserved) (unobserved) Selection Bias bias Naïve Naive estimate impact estimate of impact Outcome Income if of a if non-participant he had Participate had in PVS participated (unobserved) ATU Observed Income income of a non-participant of Participant (observed) in PVS Treatment Time Figure: Exposition of the True Measure of Average Treatment Effect (ATT) on the Treated

8 Methodology cont d Selection bias: Factors affecting participation more important Observable/unobservable Instrumental Variables (IV) corrects for both Finding a good instrument always a challenge Propensity Score Matching (PSM) : observable Conditional independence assumption (CIA) Endogenous Switching Regression (ESR): unobservable

9 Methodology cont d PSM and ESR used in this study to measure ATT PSM Suppose Y 1i and Y 0i are outcomes with and without CT Actual impact = = Y 1i Y 0i Can t observe both Y 1i & Y 0i ATT = E (Y 1i Y 0i A i = 1) where: E= expectations operator A i =1 i th individual adopted CT Y 1i and Y 0i observed and potential outcomes respectively

10 PSM continued: Methodology cont d Propensity P(X) = P (A i = 1 X) Logit/Probit functions to estimate propensity Assuming Y 1i, Y 2i A X ( means independent) E (Y 2i A = 1, P(X)) = E(Y 2i A = 0, P(X)) Guarantees every adopter has a matching non-adopter Comparison between counterparts in the two groups Different matching algorithms (nearest neighbour, Radius caliper, Kernel bandwidth)

11 Methodology cont d Endogenous switching regression technique Suppose Zi= a vector of factors affecting farmers decisions to adopt If γzi +ui>0, farmer i chooses to adopt (regime 1, denoted Ii =1) ; If γzi +ui 0, farmer i chooses not adopt (regime 2, denoted Ii =0) Suppose X1 i and X0 i = two vectors of factors that affect farmers performance under adoption and non-adoption scenarios. Then: Farmer i's income with CT ( I i =1) is: Y1 i = β1 X1 i +ε1i and Farmer i's income without CT ( I i =0 ) is: Y0 i = β0 X0 i +ε0i ATT = Y1 Y0

12 V. Results Results from Propensity Score Matching: Table: Propensity Scores Estimates from a Logit Model Coef. Std. Err. Zone(0,1) Participated in Field Day(0,1) *** Has demonstration field(0,1) *** Both demonstrations and field days (0,1) *** Level of education (Years) ** Experience (Year) *** Total area(ha) ** Distance to the nearest input market(km) *** Do you know the ZT technology(0,1) *** Value of total assets(sp) 0 0 Constant *** Log likelihood Dependent variable: Adoption of CT (No=0,Yes=1)

13 Matching estimators Results cont d Table. Choosing the best matching algorithm Balancing test^ Performance criteria Pseudo-R 2 Matched sample size Nearest neighbor (1) *** Nearest neighbor (2) *** Nearest neighbor (3) *** Nearest neighbor (4) *** Nearest neighbor (5) *** Radius caliper(0.01) *** Radius caliper (0.25) *** Radius caliper (0.5) *** Kernel bandwidth (0.1) *** Kernel bandwidth (0.25) *** Kernel bandwidth (0.5) *** ^Number of explanatory variables with no statistically significant mean difference between the matched groups of adopter and non-adopter households ***,**,* Significant ATT at 1%, 5% and 10% levels respectively

14 Results cont d Table: Propensity score and covariate balance test of variables Variable Sample Treated Control %bias %reduction bias t p>t PSCORE Unmatched Matched Zone Unmatched Matched Participated in Field Day Has demonstration field Unmatched Matched Unmatched Matched Both Unmatched Matched

15 Results cont d Table: Joint significance test of variables Sample Pseudo-R 2 LR chi 2 p>chi 2 Unmatched Matched

16 Table: ATT for net wheat income and consumption (Using Propensity Score Matching) Group Treatment group Results cont d Control group average treatment effect on the treated (ATT) Net wheat income (SP/year) S.E. T-stat Unmatched *** ATT *** Consumption (Kg/capita/year) Unmatched *** ATT *** 1US$=50SP

17 Table: FIML estimates of the endogenous switching regression model Independent Variables Results cont d Net income Equation for adoption (I i =1) Net income Equation for Non- Adoption (I i =0) Adoption of zero Tillage (No=0,Yes=1) Coef. Std.Er. Coef. Std.Er. Coef. Std.Er. Zone(0,1) *** *** Participated in Field Day(0,1) *** Has demonstration field(0,1) *** Both demonstrations and field days(0,1) *** Level of education (Years) *** Experience(Year) *** *** Total area(ha) ** Do you know the ZT technology(0,1) *** Distance to the nearest input market(km) *** *** *** Value of total assets(sp) Constant *** ***

18 Results cont d Table: ATT for net income and consumption (ESR) Variable Mean Std.Err. Students t Net Wheat Income (SL/year) Adopter actual (Y11) Adopter potential (Y21) ATT= Y11-Y t=17.16*** Consumption (Kg/capita/year) Adopter actual (Y11) Adopter potential (Y21) ATT= Y11-Y t=26.56***

19 VI. Conclusions Results from both the propensity score matching and switching regression suggest that even after controlling for all confounding factors, adoption of the CT technology leads to: About 9000 SYP (US$180) higher crop income per ha. 26kg (34%) higher wheat consumption per adult equivalent per year.

20 Conclusions cont d Achieving agricultural growth and development and thereby improving rural household welfare requires increased efforts to provide yield enhancing, cost saving and natural resources conserving technologies CT is one of few such technologies which can be justified on: Environmental, economic and food security grounds Sizeable impacts in transforming agriculture & enhancing sustainability

21 Conclusions cont d The policy implication of these results are: Governments should consider embracing CT as one of the priority technology packages in their national extension programs. Devise strategies for enhancing wider adoption Commitment to documenting impacts and building an impact assessment culture

22

23 A) Total Sample Type of producers Number of farmers producing wheat barley Sampling Total Wheat Barley Purposive % 84% Random % 57% Total % 68% B) Sample of Adopters Sampling Total Wheat Barley wheat barley Total Purposive % 72%100 Random % 64% Total % 71% 4.87% 11.8%

24 Governorates included in the sample District Name Number of villages included in the sample Districts included in the survey Sample size from the district: Whole sample Randomly selected Total population in sample villages Total Wheat producers Total Aleppo Al Bab Wheat produce rs Ein Al Arab Sama'an Sfiera Al-haska Kamshly Tel-Hamis Malkia Amoda Hasaka Ras-Alain Edleb Khan-Shikon Almara Hamah Slmiah Sabora Homs Ksier Deraa Alshajra Alswieda Salked Total 28 14,

25 Table. Choice of matching algorithm Matching estimators Performance criteria ATT Balancing test^ Pseudo-R 2 Matched sample size Net Income Consumption Nearest neighbor (1) *** Nearest neighbor (2) *** Nearest neighbor (3) *** Nearest neighbor (4) *** Nearest neighbor (5) *** Radius caliper(0.01) *** Radius caliper (0.25) *** Radius caliper (0.5) *** Kernel bandwidth (0.1) *** Kernel bandwidth (0.25) *** Kernel bandwidth (0.5) *** ^Number of explanatory variables with no 3 statistically 0.166significant 621 mean difference between the matched groups of adopter and non-adopter households ***,**,* Significant ATT at 1%, 5% and 10% levels respectively

26 Methodology cont d Endogenous switching regression: Assumption about error terms The error terms are assumed to have a trivariate normal distribution, with zero mean and nonsingular covariance matrix expressed as: cov e 1i, e oi, u i = 2 σ e2. σ e2u. 2 σ e1 σ e1u.. 2 σ u

27 Results cont d Table. Choosing the best matching algorithm Matching estimators Performance criteria ATT Balancing test^ Pseudo-R 2 Matched sample size Net Income Consumption Nearest neighbor (1) *** Nearest neighbor (2) *** Nearest neighbor (3) *** Nearest neighbor (4) *** Nearest neighbor (5) *** Radius caliper(0.01) *** Radius caliper (0.25) *** Radius caliper (0.5) *** Kernel bandwidth (0.1) *** Kernel bandwidth (0.25) *** Kernel bandwidth (0.5) *** ^Number of explanatory variables with no statistically significant mean difference between the matched groups of adopter and non-adopter households ***,**,* Significant ATT at 1%, 5% and 10% levels respectively

28 Results cont d Table: Joint significance test of variables Sample Pseudo-R 2 LR chi 2 p>chi 2 Unmatched Matched Table: Mean of estimated propensity scores Group Obs Mean Min Max Non-adopters Adopters Total