1. Development Research Group
Living Standards Measurement Study – Integrated Surveys on Agriculture: Main Features, Challenges and Next Steps
Gero Carletto
Development Research Group
The World Bank
February 27th, 2012

2. Outline The Living Standards Measurement Study The LSMS-ISA project
Main features
Progress to date
Methodological validation/research
Challenges
Next steps

3. The LSMS Flagship initiative in DECRG since 1980 Evolution …
Poverty monitoring and measurement: the “McNamara anecdote”
Technical assistance, capacity building
Back to the “roots”: “wholesale” research
Focus on agriculture: LSMS-ISA

4. The LSMS-ISA project Started in early 2009
Tanzania pilot in mid-2008
Funding from the Bill and Melinda Gates Foundation, with additional funding from USAID, DFID and a number of other sources, including governments
Working on four fronts:
Household survey data collection
Methodological validation/research/tool development
Capacity Building
Dissemination

5. The team Gero Carletto Luc Christiaensen (from Nov 2011)
Kinnon Scott (on DAIS until Sept. 2012)
Kathleen Beegle (WDR until June 2012)
Diane Steele (LSMS Database coordinator)
Talip Kilic
Kristen Himelein (sampling)
Mimi Oseni (50% with AFTAR)
Alberto Zezza (50% with LDIA project)
Siobhan Murray (GIS; part-time)
Raka Banerjee (ETC; project coordinator)
Jon Kastelic (formerly in Malawi; data processing)
Prospere Backiny-Yetna (in Mali; formerly in Niger)
Bjorn Campenhout (in Uganda, 50% with IFPRI)
Colin Williams (formerly in Nigeria, now on STC)

6. Main Features 6+ year program (2009-2015)
Panel household surveys with emphasis on agriculture in 7 Sub-Saharan African countries
Sample: 3-5,000 households
Population-based frame (Global Strategy)
Representative at national- and few sub-national levels
Tracking
Movers
Subsample of individual split-offs

7. Main Features (cont’d)
Multi-faceted “integration”
Multi-topic survey instrument
Farm plus nonfarm, consumption, nutrition, inter alia
Build on existing/planned surveys
country ownership
Sustainability
… but large trade-offs
Link to major initiatives
National Strategy for the Development of Statistics (NSDS)
Global Strategy to Improve Agricultural and Rural Statistics
Improved links to other data sources
GIS, censuses, surveys, etc.

8. Main Features (cont’d)
Gender-disaggregated data
Use of technology
GPS for households and plots (area)
Concurrent field-based data entry
Computer Assisted Personal Interviews (CAPI)
Open data access policy
Micro-data publicly available within 12 months of data collection
GPS data dissemination

9. Main Features (cont’d)
Inter-institutional partnerships
In-country
Government (NSO, MoA, other line ministries)
WB country offices
Other development partners
Thematic
GPS and crop production: FAO (GS Action Plan)
Livestock: WB-ARD/ILRI/FAO
Food Security: WFP
Empowerment: IFAD
Agricultural Income: FAO/RIGA
Climate Change: WB/ENV
Fishery: WorldFish Centre
Analysis: IFPRI/Harvest Choice, IFAD
CGIAR: technology adoption

10. Surveys Tanzania National Panel Survey Uganda National Panel Survey
Malawi Integrated Panel Household Survey
Nigeria General Household Survey Panel:
Niger Enquête National sur les Conditions de Vie Des Ménages
Ethiopia Rural Socio-Economic Survey
Mali Integrated Agricultural Survey

11. Survey Schedule Tanzania Uganda Malawi Nigeria Mali Ethiopia Niger
2008/09
2010/11
2012/13
Uganda
2009/10
2011/12
2013/14
Malawi
Nigeria
Mali
2014/15
Ethiopia
Niger

12. Data release Tanzania Uganda Malawi Nigeria Niger Ethiopia Mali 2010
2011
2012
2013
2014
2015
Tanzania
  X
  X
  X
Uganda
  X X
  X
Malawi
  X
Nigeria
  O
  O
Niger
  X X
Ethiopia  X
Mali
  X
  X

13. Project outputs Documented public-access microdata
Sourcebooks, best practice guidelines
Tools
Computer Assisted Personal Interviews (CAPI) software
ADePT-Agriculture and ADePT-Livestock
Comparative Living Standards Project (CLSP)
Research and analysis for policy and programming
Methodological validation
Research pillar of Action Plan of Global Strategy
DFID grant

14. Experiments on Survey Methodology
Identification process
Alignment with Global Strategy
Field experience
Consultation
Peer review
Research topics
Land area measurement
Soil fertility
Water resources
Labor input in agriculture
Continuous/extended harvest crops
Non-Standard Units
Livestock
Post-harvest losses

15. Land Area Measurement Why is it important?
Fundamental component of agricultural statistics
Priority #1 of Global Strategy
Current methods inaccurate or impractical
What are the different methods available?
Traversing (compass and rope method)
P2/A
Self-reporting
GPS

16. Impact of GPS on IR
Old controversy in development economics, with recent reprise
Friends: Barrett (1996), Benjamin and Brandt (2002), … Binswanger, Pingali, …
Foes: Bhalla and Roy (1988); Benjamin (1995); Collier and Dercon (2009), …\
Binswanger et al (1995); Eastwood et al. (2010).
Possible explanations
Factor market imperfections
Omitted variables (land quality)
Measurement errors
Low correlation bet/w measurements (Udry and Goldstein, 1999)
Recent findings:
Lamb (2003)
Factor market imperfections and land quality differences
Barrett et al (2010)
Lab soil test. Marginal impact of land quality. Measurement error?
De Groote and Traore (2005)
Systematic bias in reporting (small over-report)
Objective: Test robustness of IR to land measurement error

17. Systematic bias in reporting

18. Systematic bias in reporting
Acres
Land Deciles (GPS)

19. Yields and farmsize

20. The models Plot level: Discrepancy in measurement
HH head characteristics
Plot size, squared
Control for rounding
Involvement in disputes, Slope, PSU dummies
Household level: Farm profits/acre
Farm size
HH char’s (age, edu, gender)
Inputs:
Family labor (interacted with land)
Hired labor
Variable inputs
Land quality (soil, slope, irrigation)
PSU dummies

21. Main findings Household level: IR strengthened by GPS measurement
Plot level: Small farmers over-report
Household level: IR strengthened by GPS measurement
Area Self-Reported
Area GPS
Log Land Size
-0.62***
-0.83***
[0.000]

22. Smallholders are more efficient however you measure it

23. Experiments on Survey Methodology
Research topics
Land area measurement
Soil fertility
Water resources
Labor input in agriculture
Continuous/extended harvest crops
Non-Standard Units
Livestock
Post-harvest losses

24. Soil Fertility Why is it important?
Fundamental driver of productivity in Africa – remains a key unobserved variable for analysis
What are the different methods available?
Conventional soil analysis (CSA)
Farmers’ subjective evaluations
Spectral soil analysis (SSA)

25. Water Resources Why is it important?
Agriculture in Africa is predominantly rainfed – water is a key input to production
Large discrepancies across data sources
Low resolution/idiosyncratic risk
Lack of access to data
What are the different methods available?
Satellite imagery
Weather stations
Self-reporting
Community rain gauges

26. Labor Inputs Why is it important?
Labor inputs are fundamental to labor productivity measurement
Very poorly measured
What are the different methods available?
Recall (6 month or 12 month, by activity, by demographics)
Computer-assisted telephone interviews (CATI)
Labor input diaries

27. Experiments on Survey Methodology
Research topics
Land area measurement
Soil fertility
Water resources
Labor input in agriculture
Continuous/extended harvest crops
Non-Standard Units
Livestock
Post-harvest losses

28. Extended Harvest Crops
Why is it important?
Continuously / extended harvest crops are major staple crops in many African countries
Inaccuracy of recall
May extend across seasons
What are the different methods available?
Recall (6 month or 12 month)
Crop cards (with crop card monitors)
Ongoing work with the Uganda Bureau of Statistics
CATI (data collection / supervision)

29. Diary vs. recall (Uganda)
Compare recall and diary methods for crop production estimates (and consumption from own production) in Uganda
Lack of gold standard
“well-administered” diary?
Crop cutting?
Does it vary by crop type?
Extended-harvest crops
Cassava
Banana

30. Frequencies of crops reported
Diary vs. recall (Uganda) (cont’d)
Frequencies of crops reported
Recall
Diary %
Coffee
31.9
21.7
Maize
76.8
73.8
Cassava
58.8
82.1
Tomatoes
3.4
22.2
Avocado
2.8
25.8

31. Consumption plus sales
Diary vs. recall (Uganda) (cont’d)
Output values (usd)
Recall
Diary
Consumption plus sales
Ratios
(1)
(2)
(3)
(2)/(1)
(3)/(1)
Cash crops
48.6
27.2
57.1
0.56
1.18
Food crops seasonal
135.4
297.4
164.5
2.20
1.21
Food crops continuous
178.1
275.6
270.3
1.55
1.52
Fruit & Vegetables
10.8
14.9
43.4
1.37
4.00
Total
372.9
615.1
535.3
1.65
1.44

32. Experiments on Survey Methodology
Research topics
Land area measurement
Soil fertility
Water resources
Labor input in agriculture
Continuous/extended harvest crops
Non-Standard Units
Livestock
Post-harvest losses

33. Non-Standard Units

34. Weight in kgs of a 50 kg sack
Non-standard units (cont’d)
Weight in kgs of a 50 kg sack
Maize
50.0
Cassava
41.7
Sweet potato
36.2
Irish potato
42.6
Groundnut
44.2
Ground bean
43.2
Rice
56.2
Finger millet
50.5
Sorghum
49.6
Pearl millet
Bean
77.6
Soyabean
53.1
Pigeonpea
57.1

35. Non-standard units (cont’d)
State of crop …
Maize: A 50 kg size sac weighs only 29.0 kilograms when it is filled with fresh maize. This translates into 14.8 kgs of maize grains.
Cassava: A 50 kg size sac weighs 37.8 kilograms when it is filled with dried cassava. To attain these 37.8 kgs of dried cassava, one would have had to start with 108 kgs of fresh cassava.
Rice: A 50 kg size sac weighs 38.5 kgs when filled with rice which has not been husked. These 38.5 kgs translate into 24.3 kgs of grain when husked.

36. Land area by zone (Nigeria)
Non-standard units (cont’d)
Land area by zone (Nigeria)
Zone Specific Conversion Factors into Hectares
Zone
Conversion Factor
Heaps
Ridges
Stands 1 2 3 4 5 6

37. Challenges Integration comes with trade-offs
Data preparation and data release
Time consuming and not enough “control”
Resident advisor in bet/w years
Dissemination of GPS information
protocol
Sample
Too small? Representative at crop level? Livestock?
Linking to Ag census and large Ag surveys
CAPI
More sustainable solution
Frequency of surveys
Moving to every other year
Tracking
Still a challenge in some countries

38. Next steps More tools? More countries ? More analysis Mobile phones
Ag input/labor use, climate data; Listening to Africa
More countries ?
in Africa: Burkina Faso
beyond Africa?
More analysis
Gender and Agriculture
Agriculture and Nutrition
Myths and Facts in African Agriculture

40. The End

41. Is women’s control of income important for child nutrition?
>>
Dependent Variable: Z-Score of Height-for-Age
Definitions of Woman’s Share of Household Income V1
Assumption 100 to Head V2
Assumption 50/50 Split V3
Assumption
a la HH
V4 Preferred
Child: Male
-0.130
-0.129
-0.147
-0.186**
Woman's Share of Household Income x Male Child
-0.735
-0.070
-0.008
0.155***
Observations
2,522 R2
0.711
0.710
note: *** p<0.01, ** p<0.05, * p<0.1

42. The tale of the missing plot measurements
Concurrent Data Entry
The tale of the missing plot measurements
High initial rates of missing gps data in months 1 & 2
Percentage of Plots coming from the field without measurements over the 12 months of field work. Background on Malawi Plot Measurement protocol;

43. Concurrent Date Entry (cont’d)
The tale of the missing plot measurements
Intervention - High rate of missing data observed and new instructions to field disseminated.
In Month 2 we find that the frequency of missing measurement is too high. And adjust the protocol for plot measurement.

44. Concurrent Data Entry >>
The tale of the missing plot measurements
Substantial decrease in missing data. Because of revisit of households in month 4-6, part of the missing data was now captured.
Icing on the cake. Part of the households surveyed were going to again be surveyed three months later. Thus, under the new protocol originally missing plot measurement were captured, decreasing the overall percentage of missing plot measurements and contributing to more accurate yield calculations. At the end of the survey under 1 percent of plots were going unmeasured.

45. CAPI
“Comparative Assessment of Software Programs for the Development of Computer-Assisted Personal Interview (CAPI) Applications”, with the University of Maryland IRIS Center
Implementation in LSMS-ISA partner countries relying on readily available commercial software products
Plans to create a public, freely available CAPI software package
>>

46. GPS: What we are collecting
Household location
with current devices
coordinate precision is
within 5-10m (atmospheric
interference, vegetation
canopy, buildings..
may effect precision )
Just a reminder of the GPS data that we are collecting with the LSMS-ISA surveys.
With mapping grade precision a particular dwelling is easily identified using survey GPS coordinates

47. GPS: What we are collecting
Plot area
Plot location
Plot outline

48. Dissemination Strategy
NOT proposing to disseminate actual household GPS coordinates
ARE proposing to disseminate modified EA center-points, offset to prevent identification of communities AND a set of geovariables generated using the true locations
true household GPS locations maintained in the Statistics Office, to be used for continuation of the panel or at the discretion of the NSO
Acceptable risk of disclosure?

49. Dissemination: geovariables
Having georeferenced hh and plot locations enables integration with other datasets, making available large range of additional variables. 3 types of geovariables we would like to disseminate with the survey dataset are:
Distance – mapping showing distance to major market – darker areas more remote
lots of geophysical characteristics that would be relevant, particularly looking at variations in ag productivity, this just a subset
3. Time series – as we mentioned yesterday in shock module, information on timing of rains and rainfall anomalies can be useful in interpreting production statistics and other survey data. Where time series are available we can use use rainfall or vegetation response measures to enhance augment what is capture in the survey instruments
Distance
HH to Plot
HH to Market
HH to Major Road
Environmental
Climatology
Landscape typology
Soil
Elevation
Terrain
Time Series
Rainfall
Vegetation Indices
>>