ETHIOPIAN DEVELOPMENT RESEARCH INSTITUTE

Scaling up the Adoption of Improved Technologies in Staple Food Production:

The Case of Row Planting of TeffJoachim Vandercasteelen (LICOS, KU Leuven); Mekdim Dereje (EDRI, ESSP); Bart Minten (IFPRI, ESSP); and Alemayehu Seyoum Taffesse (IFPRI, ESSP)

Ethiopian Economics Association (EEA) and the Econometric Society 19th Annual Conference of the African Region Chapter of the Econometric Society12th International Conference on the Ethiopian EconomyJuly 16-19, 2014Addis Ababa

1

Introduction

• Agricultural growth matters– Economic development and food security– Low in Sub Saharan Africa

• Improved technology adoption crucial– Agricultural productivity to be increased– Welfare, food security and poverty implications– However, often not understood how to scale up adoption of

technological innovations

• We study effect of promotion campaigns – Transfer of knowledge and information– Awareness, teaching, training

Introduction

• Lack of evidence on potential and impact of such promotion campaigns

• Evaluation of on-farm productivity effects – Yield benefit at farm level using an experimental survey– Yield gaps

• Specific study of row planting in teff production in Ethiopia– Low teff yield– Large scale promotion campaign

BACKGROUND

Teff in Ethiopia

• Teff– Major staple food

• 2 out of 3 Ethiopians consume teff daily • Produced by 6 million farmers • In value of production/area, most important crop

– Low agricultural productivity• Limited knowledge• Constraints inherent to the crop (small seeds…)

– Row planting• High agronomic yields

• Promotion campaign– Package (fertilizer, improved seeds)– From 1,400 to 1,600,000 farmers targeted

Broadcasting

Row planting

Experimental Survey

• Rolled out in line with “pre-scale-up” phase (2013)– Scientific recommendations and definitions intensively

promoted during training days– Input provision (seed and fertilizer for free)– Selection and extension done by Development Agents (DAs)

• 2 stage randomization approach– 4 Farmer Training Centers (FTC) in 9 Woreda’s of Oromia– 10 farmers row planting/ traditional broadcasting

• Experiment – Farm and village level– Experimental plot of 300 m²– Free improved seed and fertilizer

Survey outcome

Sample

Random (19) n=537

Non-Random (17)n=341

Compliersn= 506

Non-compliersn= 31

878 Farmers 36 Villages

187 village demonstration plots

METHODOLOGY

Treatment effect

• The impact of promoting row planting– Yield from crop-cut– Reported yield after harvest

• Average Treatment effect on the Treated (ATT)

• Identification of program effect– Confounding factors– Control group (counterfactual)– Sample selection bias

Randomized Control Trial

• Expected mean yield difference = ATT– Assignment of treatment is independent of Yi and Ti

– Broadcasting and row planting farmers are statistically identical

• Balancedness of random sample – Household characteristics

• Demographics, education, assets

– Experimental plot characteristics• Plot quality, input use, production practices• But plot size

– Unobserved heterogeneity

• Management, intellectual capacities, networks

Randomized Control TrialVariable Traditional

Broadcasting

(n=173)

Row Planting

(n=333)

Mean Difference t-value

Household head

characteristics

Age (years) 43.8 -0.80 -0.72

Gender (male=1) 99.4 -2.73 -2.39**

Literacy (yes=1) 68.9 4.79 -1.12

Primary education (yes=1) 65.9 3.17 0.72

Household

characteristics

Distance to FTC (minutes) 34.2 -0.60 -0.26

Household size (members) 7.1 -0.18 -0.86

Total agricultural assets value (ln of ETB) 6.8 0.00 0.03

Total assets value (ln of ETB) 7.2 0.14 0.72

Income from other activities (yes=1) 76.3 -7.83 -0.96

Experimental

plot

Area (m²) 599.5 -161.80 -4.29***

Number of plows (number) 4.9 0.04 0.29

Number of weedings (number) 1.9 0.07 0.66

Organic input used (yes=1) 11.6 -2.25 -0.77

Inorganic fertilizer used (yes=1) 99.4 -0.32 -0.42

Manure used (yes=1) 10.0 -3.20 -1.12

Rate of Urea used (g/m²) 9.2 1.91 1.06

Rate of DAP used (g/m²) 11.7 0.19 1.04

Rate of herbicides used (100 ETB/m²) 2.0 -0.11 -0.37

Teff characteristics

in Meher 2011/2012

Teff cultivated in both seasons (yes=1) 0.7

0.01 0.21

Average teff area (ha) † 0.6 0.01 0.24

Average teff yield (ton/ha) † 0.9 0.08 1.61

Farmers’ unobserved heterogeneity (.) † 1.6 0.04 1.05

Intention To Treat

• Imperfect compliance– 6% are non-compliers

• Intention To Treat (ITT)– No spill-over effects

Matching

• Full sample of farmers• Possible selection bias because of purposive

selection• Propensity Score Matching (PSM)– Estimate Propensity Score – Yield difference between matched farmers

RESULTS & DISCUSSION

Farm level

• RCT– Positive, but non-significant, effect of row planting on

yields

Treatment

effect

Yield from crop-cut Reported yield after harvest

control row plantingControl

row planting

Coefficient 1.096*** 0.015 1.147*** 0.116

ATT se (0.049) (0.065) (0.051) (0.075)

Observations 403 506

Coefficient 1.132*** 0.108

ITT se (0.051) (0.072)

Observations 531

Heterogeneous effectVARIABLES Yield from crop-cut harvest Reported yield after harvest

Level Interaction with treatment

Level Interaction with

treatmentRow planting (yes=1) 0.653 0.120

(0.555) (0.543)Age of household head (years) 0.002 -0.002 0.004 0.011

(0.004) (0.006) (0.004) (0.008)Primary education household head (yes=1) 0.077 -0.175 0.234** -0.178

(0.094) (0.138) (0.096) (0.169)Gender of household head (male=1) -0.774*** 0.332 -0.805*** 0.341

(0.010) (0.368) (0.084) (0.313)Farm size (ln of ha) 0.047 -0.048 0.045 -0.051

(0.038) (0.043) (0.060) (0.079)Size of the household (number of persons) 0.052 -0.042 0.048 -0.046

(0.037) (0.043) (0.032) (0.042)Experimental plot size in line with guidelines (yes=1)

-0.048 -0.208 0.204* -0.173(0.043) (0.156) (0.109) (0.157)

Constant 0.955*** 0.957***

(0.336) (0.321)Observations 403

0.045506

0.045R-squared

Farm level

• Matching– First stage: Probit model of selection– Second stage: ATT; again positive effects but not significant

Matching algorithm

Yield from crop-cut Reported yield after harvest

Observations Row-planters 456 541 Controls 293 371

Nearest Neighbor Matching (NNM) ATT 0.027 0.060 Standard error (0.070) (0.076)

Kernel Matching (KM) ATT 0.056 0.078 Standard error (0.058) (0.065)

Radius Matching (RM) ATT 0.028 0.057 Standard error (0.058) (0.065)

Village level

• Village level– Plot is managed by an extension agent directly– Stronger and significant effect

Yield from crop-cut

ATT 0.307**

Standard error (0.117)

Control 1.016***

Standard error (0.108)

observations 187

Mechanisms

• Yield gap at farm level– “Bad adoption”: plot management – Roll out of promotion campaign

• “Problematic input supply”• “Extension quality”

• Tested by interaction effects:– Seed rate different w.r.t. scientific recommendations– Receiving inputs too late– Yield benefit achieved by DA at village level

Mechanisms

Mechanisms Yield from crop cut reported yield after harvest

row planting 0.427*** 0.558*** (0.071) (0.149)

Bad adoption° 0.001 0.000

(0.004) (0.001)

Bad adoption * row planting -0.005*** -0.005***

(0.001) (0.001)

Problematic input supply -0.254*** -0.201*

(0.060) (0.101)

Problematic input supply * row planting -0.144 -0.636** (0.123) (0.235)

Extension quality° 0.020*** -0.001 (0.004) (0.010)

Extension quality * row planting 0.022** 0.040*** (0.008) (0.011)

Constant 1.064*** 1.125***

(0.052) (0.084) °=normalized with Kebele average difference

Farmers plans for next year

• Questions asked on plans after the experiment:- Farmers overall positive and still planning to continue but on

limited areas, possibly because of labor demand issues

Percentage

1. “Will you allocate some part of your teff area to row planting?” 73%

2. “Share of the total teff production land allocated to row planting next year” 19%

3. “The major reasons for not doing row planting next year” (top three):

- “Too much additional labor” 96%

- “Difficulty of doing row planting after rain” 25%

- “It does not give higher yields” 16%

CONCLUSION

Conclusion• Large scale promotion campaign

– Low yields ask for adoption of new technologies– Promising on-station results – No empirical evidence

• Promotion of row planting to increase yield in the first year– Moderate farm level effect on yields, seemingly partly explained

by some issues with respect to implementation

• Implications:– Higher effects found in recent ATA study on effect of TIRR package

(improved seed + row planting): 44% increase in yield – Might be explained by learning-by-doing? Improved

implementation afterwards? Package approach? External validity issues? Further monitoring and evaluation required that could help improve adoption processes of improved technologies

– Mechanization potential to reduce labor issues?