1

• A new Approach to Improving Sample Design for
  Crop Forecast and Post Harvest Estimates in
  Zambia
• A Contributed Paper Session Presented at the
  International Conference on Agricultural
  Statistics (ICAS III), Westin Regina Hotel,
  Cancun, Q.R. Mexico
• 2nd-4th November 2004
• By
• John Kalumbi
• Deputy Director Agriculture Environment
  Division
• Central Statistical Office (CSO)
• Lusaka, Zambia

2
I. INTRODUCTION AND INSTITUTIONAL ARRANGEMENT

• Production of Agricultural Statistics
• Responsibility of the CSO (Agric. Environment
  Division)
• Collection of data is done in conjunction with
  the Ministry of Agriculture and Cooperatives
  (MACO) through the Early Warning and Data Base
  Management Unit.
• Agriculture Environment Division
• Agriculture Branch
• Small Medium Scale Farm Section
• Large Scale Farm Section

3

• Environment Branch
• Land degradation
• Air pollution
• Water Sanitation
• Forestry
• Wildlife
• Fisheries
• Evolution of Agricultural Surveys in Zambia
• The 1970/71 Census of Agriculture marked the
  beginning of annual agricultural surveys in
  Zambia.
• Crop Forecast Survey (CFS)
• Post Harvest Survey (PHS)
• Area Measurement and Crop-Cutting Survey (AMCC)
  has since been discontinued

4
II. METHODOLOGY USED

• Survey sampling for both CFS and PHS for Small
  Medium Scale agricultural holdings
• Complete enumeration for Large-Scale Agric.
  Holdings
• Sample Design
• Cartographic census field operation defines
  Census Supervisory Areas (CSAs), which are
  further subdivided into Standard Enumeration
  Areas (SEAs)
• An SEA is the unit of data collection point at
  district level

5
Previous CFS/PHS Sample Design

• Based on 1990 census data
• Only rural districts were covered
• A stratified three stage sample design was used
• CSAs Primary Sampling Units (PSUs) selected at
  first stage with PPS
• SEAs selected at second stage, with one SEA
  selected with PPS within each sample CSA
• Measure of size in both cases is number
  households as listed in the 1990 census of
  population and Housing.
• A total sample of 405 SEAs was drawn out of 9,000
  rural SEAs
• No provision of replacement of both sample SEAs
  and sample households.
• Sample households were selected at third stage
  from the listing stratified by farm size category
  using Linear Systematic Sampling (LSS) procedure
  with random start.

6
New CFS/PHS Sample Design

• Based on 2000 census data
• A stratified two-stage sample design is now being
  used
• SEAs are the Primary Sampling Units (PSUs)
  selected at first stage with PPS
• Measure of size is number of agricultural
  households as listed in the 2000 census of
  population and housing
• A total sample of 410 SEAs were drawn out of the
  12,788 agricultural SEAs as defined for the 2000
  Zambia Census
• Sample households are selected at the second
  stage from the listing stratified by farm size
  category (land area, number of animals, special
  crops criteria)

7
Other Features of New CFS /PHS Sample Design

• Includes a small fraction (5) of urban SEAs,
  i.e, urban SEAs in which 70 percent or more
  households are agricultural according to the 2000
  Zambia census data (about 586 urban SEAs in
  total)
• Has provision for the replacement of any sample
  SEAs that may not be enumerated for one reason or
  another in order to maintain the effective sample
  size.
• Eight (8) crops were identified to receive
  special treatment at second sampling stage in
  order to improve the precision of survey
  estimates of crop area and production (i.e.,
  these are crops with limited geographical
  concentration whose CVs previously were
  relatively high due to the smaller number of
  observations sorghum, rice, cotton, burley
  tobacco, Virginia tobacco, sunflower, soybeans,
  and paprika)

8
III. STRATIFICATION OF HOUSEHOLDS

• Previous sample design
• Included households listed in sample SEAs by two
  farm size categories
• Category A 0 4.99 hectares
• Category B 5.0 19.99 hectares
• New sample design
• Stratification of households listed in sample
  SEAs is based on
• Farm size (land area) category
• Number of livestock or poultry raised criteria
• Growing of special crops criteria

9
Stratification of households by farm size (land
area)

• Category A 0 1.99 hectares
• Category B 2.00 4.99 hectares
• Category C 5.00 19.99 hectares
• In both the previous and new sample design,
  households with farm size category of 20 or more
  hectares were/are included in the special frame
  for large-scale farmers and are enumerated
  separately.

10
Stratification of households by number of
livestock or poultry raised criteria

• Based on the number of livestock or poultry
  raised, households are added to category C if the
  following minimum number are being raised
• Cattle 50
• Pigs 20
• Goats 30
• Poultry 50

11
Stratification of households by growing of
special crops criteria

• Households may be added to category B and C based
  on the special crops treatment criteria
• If the sample SEA only has 1 or 2 households with
  any of these special crops, these households will
  be asigned to category C ( in case they do not
  qualify based on land area and animals)
• If the sample SEA has only 3 to 5 households with
  any of these special crops, such households are
  assigned to category B (if they were previously
  assigned to category A based on land area and
  livestock)

12
IV. SAMPLE HOUSEHOLDS SELECTION IN SAMPLE SEAs

• Total sample size for each SEA 20 households
• Distribution of sample households to each
  category
• Category C 10
• Category B 5
• Category A 5
• If there are shortfall in category C, the
  difference from 20 will be allocated equally to
  categories B and A or category B will have one
  more sample households than category A, if the
  difference cannot be allocated equally.
• Where there is no household in category C and
  less than 10 in category B, the difference from
  twenty will be met from category A.
• If all listed households fall in category A, then
  all 20 sample households will come from category
  A.
• For each category in the sample SEA, selection of
  sample households is done using Linear Systematic
  Sampling (LSS) Procedure with random start.

13
V. ESTIMATION

• Estimating for the total involves the weighing of
  variables before additions are made at district,
  provincial and national levels. The basic
  sampling weight is obtained as inverse of the
  probabilities of selecting the SEAs and
  households by category size.
• The formula for estimating the survey estimates
  of a total is as follows
• Y ? ? ? ? W?shi Yshij
• h i s j
• Where
• Yshij value of variable y for the j-th
  sample household in category s within the i-th
  sample SEA in district h
• W?shi basic weight for the sample households
  in category s within the i-th sample SEA in
  district h.

14
END OF PRESENTATION