1. Measuring the effect of site-specific soil input recommendations on choice
Christopher G. Magombaa
with Aurélie P. Haroub bMcGill University; aSokoine University of Agriculture
Nyambi Amuri, Malgosia Madajewicz, Hope Michelson, Cheryl Palm, Johnson Semoka, Kevin Tschirhart
Evidence to Action Conference
May 24-25, 2017
Nairobi, Kenya
Poor soil quality and the associated low crop productivity is linked to the pervasive rates of poverty and malnutrition ensnaring much of Africa (Minten and Barrett 2008, Sanchez and Swaminathan 2005).
Low agricultural productivity in much of sub-Saharan Africa. For example, Tanzania: 1.3 T/ha v. US: 5.9 T/ha
Cereal yields – kg/ha of harvested land (includes wheat, rice, maize, barley, oats, rye, millet, sorghum, buckwheat and mixed grains
SSA farmers: 9 kg/ha
Latin America: 73 kg/ha
Asia: kg/Ha
REASONS FOR LOW INPUT USE:
Weak infrastructure – limited availability, quality of fertilizer is poor
Lack of information - heterogeneous soil attributes make different levels of fertilizer profitably, recommended quantity and type of fertilizer are not adapted regionally according to area’s soil type
More efficient with complementary practices
Credit constrained
Risk - rain-fed agriculture, price volatility

2. Motivation Reasons for low input use
Weak infrastructure (Gregory and Bumb 2006, Jayne et al. 2003)
More efficient with complementary practices (Snapp et al. 2010)
Risk (Dercon and Christiaensen 2011, Morris et al. 2007)
Quality concerns (Michelson et al.)
Costs - credit constrained (Duflo et al. 2011)
Lack of information (Foltz et al. 2011, Marenya and Barrett 2009, Duflo et al. 2008)
Weak infrastructure – limited availability, quality of fertilizer is poor
Lack of information - heterogeneous soil attributes make different levels of fertilizer profitably, recommended quantity and type of fertilizer are not adapted regionally according to area’s soil type
More efficient with complementary practices
Credit constrained
Risk - rain-fed agriculture, price volatility

3. Motivation
How do plot-specific fertilizer recommendations affect farmers’ agricultural input decisions and yields?
How do these input decisions differ when farmers’ decisions are constrained by access to cash/credit.
1 - Relative to information received via traditional extension services

4. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

5. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

6. SoilDoc A portable on-farm soil testing kit
Measure biological, chemical & physical conditions
Recommendations provided in near real time
Current results correlate highly with lab methods
Lab-in-a-box developed by Dr. Ray Weil, University of Maryland
Soil fertility depletion due to nutrient mining and erosion
Information on soil properties is crucial for managing and maintaining soil health (ability of the soil to function properly for intended use)
Limited access to soil testing service among small scale farmers – hinders quality recommendations, farmers decision on how to manage soil, and limit increased productivity
After years of research – developped
Sig interest – in Zambia, Tanzania and Nigeria
Soil pH, Electrical conductivity, NO3-N, Exchangeable K+, Available P, Available S, and Active C

7. Background - literature
Fabregas et al. (ongoing)
Xavier Giné & Carolina Corral et al. (ongoing)
Tjernstorm et al. (ongoing)
Kremer et al:Fabregas et al. (2014) study how much farmers are willing to pay for results of nearby experimental fertilizer and yield plots. They first show the strong correlation between soil qualities between plots geographically close to each other. Using different elicitation techniques, they subsequently examine how much farmers are willing to pay for soil information. They find that farmers are willing to pay for such information, with a mean willingness to pay ranging between $ $2.60, depending on the method and level of information presented to the farmer.

8. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

9. Experiment The treatment has two components: 3 treatment groups
Plot-specific soil management recommendations based on SoilDoc for the 2014 – 15 growing season.
A voucher worth $40 that can be redeemed for agricultural inputs.
3 treatment groups
Treatment group 1: each farmer receives soil management recommendations.
Treatment group 2: each farmer receives soil management recommendations and a voucher.
Treatment group 3: each farmer receives a voucher.
Is designed to assess whether soil information alone can improve input use and yields.
… and whether relaxing the credit constraint increases the impact of information.

10. Population of interest
The population from which we select villages and farmers for the experiment consist of:
Villages in the Morogoro district in Tanzania in which farmers grow maize.
Farmers who planned to grow maize in the 2016 rainy season on at least one plot on which they grew maize in the 2014 rainy season.
Farmers who in 2014 had not participated in the Tanzanian government voucher program during the preceding 3 years.

12. Selection of treatment and control
20 randomly selected villages receive the 3 treatments.
In each treated village, we randomly select 10 farmers for each of the 3 treatment groups and 10 farmers for the control group.
Control group consists of 2 groups:
Farmers in treated villages.
Farmers in 30 randomly selected villages in which no one receives treatment: 10 randomly selected farmers in each village.

13. Sample 200 households receive each of 3 treatments
200 control households are in treated villages and 300 in non-treated villages
Total sample:
600 treated households control households = 1100 households

14. Intervention Part I: soil management recommendations
Soil scientists collected soil samples from all households’ plot.
Which plot?
Main maize plot which was cultivated during the rainy season and on which maize will be cultivated in the rainy season.
A farmer’s main plot is defined as the plot a farmer considers to be most important to their household’s livelihood (in terms of food security and income generation).
In January 2016, households received their assigned treatments from extension agents: either soil management recommendations or a voucher or both.
Control households received information for the 2015 – 16 growing season.

16. Intervention II: vouchers

18. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

19. Project summary - time line
8/2014:
Baseline survey
8/2015:
Follow-up baseline survey
8/2016:
Endline survey
10/2014 – 11/2014:
Soil sampling
11/2014 – 7/2015:
Soil analysis
11/2015 – 1/2016:
Intervention

20. Data
Baseline survey: demographics, household and dwelling characteristics, all agricultural inputs and yields, knowledge about soil and farming practices, credit, savings and assets, plot map and details
Soil sampling analysis: Soil pH, Electrical conductivity, NO3-N, Exchangeable K+, Available P, Available S, and Active C
Follow-up baseline: yields/harvest, quantities of maize sold, stored and consumed, expectations about returns from fertilizers
Endline survey: demographics, household and dwelling characteristics, all agricultural inputs and yields, knowledge about soil and farming practices, credit, savings and assets, plot map and details.

21. Attrition Baseline Soil Analysis Follow up Endline Control Voucher
Baseline
Soil Analysis
Follow up
Endline
Control
190
181
147
179
Voucher
198
188
155
187
Recommendations
191
138
177
Recommendations + voucher
203
157
Control village
268
252
244
251
Total farmers
1,050
1,007
841
984

22. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

23. Outline Background Experimental design Data Preliminary results
Preliminary conclusions
Future work

24. Fertilizer recommendations
Control
Voucher
Recommend.
Recommend. & voucher
Control villages
Total
N only limited 6 12 5 35
NP limited 2 1
NK limited
NS limited
101
117
111
113
196
638
NPK limited
NPS limited 51 50 43 48 22
214
NKS limited 10 9 19 56
NPKS limited 11 18
179
187
188
251
1,001

26. Recommended cost per 0.5 acre
Nutrient limitations
Treatment
Cost breakdown
Total cost N
25kg Urea + 2x5kg Urea
35, x11,000
57,000 NS
25kg SA + 25kg Urea
30, ,000
65,000 NP
25kg DAP + 25kg Urea
45, ,000
80,000
NPS
40kg Minjingu Mazao + 25kg Urea
50, ,000
85,000 NK
25kg MOP + 25kg Urea + 2x5kg Urea
42, , x11,000
99,000
NKS
25kg SOP + 25kg Urea + 2x5kg Urea
50, , x11,000
107,000
NPK
25kg DAP + 25kg MOP + 25kg Urea
45, , ,000
122,000
NPKS
50kg Yara Mila Tobacco + 25kg Urea
110, ,000
145,000

27. Fertilizer purchases 2016 Voucher Recommend. Recommend. & voucher
Voucher
Recommend.
Recommend. & voucher
Control
Control village
Urea & SA (D) 7 99
Urea & DAP (D) 1
Only urea (D) 54 88 n
198
191
203
190
268

28. Distribution of fertilizer purchases

29. Cash Voucher Recommend. Recommend. + voucher Control Control village
Cash balance (D) 61
183
Cash only (D)
121 3
Cash balance (TZ Sh)
31,687
N/A
21,807
Cash only (TZ Sh)
68,000

30. Effect of interventions
Interested in quantifying the effect of giving farmers plot-specific fert recommendations on farmers’ use of fertilizer and other inputs and on their yields
The effect of the intervention (T) on maize yield & fert use for main maize plot

31. Effects on main maize plots (as defined in 2014)
Fertilizer used on maize plots, kg/acres main maize plot
Fertilizer used on other crops, kg/acres main maize plot
Maize yields, kg/acres main maize plot
Interaction: recommendations & 2016
0.00
16.765
(12.064)
(0.00)
(62.122)
Interaction: recommendations + vouchers & 2016
10.91***
90.341
(1.700)
( )
(59.488)
Interaction: vouchers & 2016
4.549***
0.161
27.603
(1.561)
(0.159)
(57.289)
Interaction: control in trt villages & 2016
0.397
(0.421)
(62.597)
Trt: recommendations (D)
13.063
(12.893)
(56.054)
Trt: recommendations + vouchers (D)
-0.01
(0.029)
(50.647)
Trt: vouchers (D)
0.427
(0.313)
(47.666)
Control in Trt Villages (D)
-0.017
-5.31
(0.025)
(57.533)
Year 2016 (D)
0.291*
(0.171)
(43.705)
Constant
0.03
***
(0.021)
(30.155) R2
0.01 N
2,022

32. Effects on main maize plots (as defined in 2016)
Fertilizer used on maize plots, kg/acres main maize plot
Fertilizer used on other crops, kg/acres main maize plot
Maize yields, kg/acres main maize plot
Interaction: recommendations & 2016
-0.53
(12.069)
(0)
(63.983)
Interaction: recommendations + vouchers & 2016
11.563***
83.384
(1.763)
( )
(60.087)
Interaction: vouchers & 2016
4.488***
0.161
12.753
(1.546)
(0.159)
(57.478)
Interaction: control in trt villages & 2016
-0.024
-57.07
(0.552)
(65.833)
Trt: recommendations (D)
13.063
(12.893)
(56.054)
Trt: recommendations + vouchers (D)
-0.01
(0.029)
(50.647)
Trt: vouchers (D)
0.427
(0.313)
(47.666)
Control in Trt Villages (D)
-0.017
-5.31
(0.025)
(57.533)
Year 2016 (D)
0.713*
(0.396)
(45.729)
Constant
0.03
***
(0.021)
(30.155) R2
0.01 N
2,022

33. Effects on entire farm
Fertilizer used on maize plots, kg/acres - entire farm
Fertilizer used on other crops, kg/acres - entire farm
Maize yields, kg/acres - entire farm
Interaction: recommendations & 2016
-0.1
13.853
(12.49)
(0.1)
(68.294)
Interaction: recommendations + vouchers & 2016
10.23*** *
(1.485)
( )
(65.271)
Interaction: vouchers & 2016
4.477***
0.061
42.409
(1.467)
(0.188)
(59.662)
Interaction: control in trt villages & 2016
-0.024
(0.552)
(76.324)
Trt: recommendations (D)
13.063
(12.893)
(65.388)
Trt: recommendations + vouchers (D)
-0.01
(0.029)
(59.803)
Trt: vouchers (D)
0.427
(0.313)
(59.162)
Control in Trt Villages (D)
-0.017
-6.002
(0.025)
(75.077)
Year 2016 (D)
0.713*
0.1
-90.17*
(0.396)
(48.465)
Constant
0.03
***
(0.021)
(39.212) R2
0.01 N
2,022

34. Monthly maize sales and purchases

35. Use of maize n Mean Std. dev. Min Max Kg consumed 922 556 624 16 6,860n
Mean
Std. dev.
Min
Max
Kg consumed
922
556
624 16
6,860
Kg sold
295
1009
1844 19
18,200
Kg stored
132
391
552 4
3,120
Kg harvested
926
931
1550

36. Expectations of yield returns

37. Expectations of fertilizer returns by good, bad, average season

38. Some Conclusions Learning effect Financial constraints are real
Information flow

39. Asanteni Sana