State Council

1
State Council Ministries (NBS, MOA, MOH, CAS,
LGPR) Provinces Autonomous Cities (Prefectures)
Counties (2400) Towns/Townships Administrativ
e Village (Natural Village) Households
2

• Chinas Statistical Information System
• Administrative Reporting System (ARS) for
  collecting/aggregating data in each ministry. No
  established practice of data exchange between
  ministries therefore, assembling cross-cutting
  data sets is going to be problematic.
• Top-down planning system where leadership (State
  Council) deals with central ministries, the
  latter with their counterpart provincial
  government agencies, the latter with the
  counties, then towns/townships, administrative
  villages, etc. Hence, central leadership need
  province-level info mainly, provinces need
  county-level info, counties need townhip info,
  etc.
• NBS and a few ministries (e.g. MOA, MOH) have
  employed survey sampling methods alongside their
  ARSs. Capability and acceptability of survey
  sampling methods not yet well established.

3

• Major Databases/Sources
• NBS - Censuses (1996 CA, 2000 CPH)
• - County Database of 128 indicators 20
  published annually
• - Sample Surveys by RSO, USO, ESO, Population
  Dept.
• For poverty monitoring and FIVIMS, important
  surveys include RHS (857 sample counties, 67,000
  households) NPS (of 592 nationally designated
  poor counties results for OLGPR exclusively).

4

• MOA - County database of 140 indicators
• - Fixed Rural Observation Villages Survey, a
  panel survey run annually since 1984 with 300
  villages and 21,000 households.
• CAS- a state of the art electronic database
  system with
• 500 gigabytes of information.
• MOH- also maintains a country database, and
  conducts
• health and nutrition surveys mostly in the
  Center for
• Preventive Medicine.

5

• Emerging Need For Small Area Statistics
• PA program focused on (592) counties until now.
  Results remarkable. However, questions have begun
  to arise
• Designated poor counties not chosen objectively.
• Half of remaining poor are in 2412-592 counties.
• Remaining poor concentrated in poverty traps
  that are townships and villages, many in
  Western region.
• Need to focus also on human (non-income) poverty.
• Urban poverty needs to be given more attention
  now,
• Efficiency in targeting to reach the remaining
  poor require that PA programs go lower, to
  townships villages.

6

• OLGPR has plans to change to a new PA policy
  gradually in next 10 years.
• Main focus is on townships villages, with
  counties exercising overall administration.
  County is basic unit of planning township is not
  a basic unit of planning in terms of action
  plan, use village level
• Provinces will rank/identify poor townships
  villages CG and provincial funds will be
  allocated directly to them. Need for
  consistency in provinces methods .
• The CG will still identify poor counties (not
  necessarily the same 592) because some ministries
  donors may still want to target their
  interventions at some counties.

7
OLGPR will need help in having the needed county,
township and village data, and in developing
methodologies for identifying/ranking poor
areas. There are some initiatives in this
direction from ADB, WFP, UNDP, WB and of course
FIVIMS/FAO.
8

• FIVIMS Information Requirements
• Crosscutting e.g. agriculture (area, production
  per capita,),agro-climate (precipitation, soil
  type, ), primary education (enrolment ratios,
  gender balance,), nutrition (energy intake,
  stunting, underweight, ), health (mortality and
  morbidity rates, ). Importance of sharing,
  networking between sources, users, and
  stakeholders.
• In addition to levels, trends and variabilities
  (time series) for some indicators are needed for
  analysis and assessment of insecurity and
  vulnerability.
• Disaggregated to small areas/domains. In China,
  uniform geo-codes available down to counties only.

9

• Sources and Methods of Indicators/Statistics
  Production
• Direct from primary source. Eg. Tabulate ag
  census for villages, townships, counties,
  provinces, regions. Then there will be no need to
  reprocess every time someone needs info.
  Estimates from reporting systems, see e.g. NBS
  and MOA county indicators in Annexes 1 5.
  Estimates from surveys, e.g. see Annex 2 from
  RHS.
• 2. Combining sources to improve direct estimates,
  as in ratio- and regression-type estimators.

10
Suppose y is direct estimate from RHS sample, x
is corresponding estimate using data from
auxiliary source (ag census or reporting system,
and X is known total from auxiliary source. Ratio
estimate yr (y/x)X Regression estimate
yreg y b(X-x) Efficiency is gained if
correlation between the variables exceeds
(1/2)CVx/CVy). The techniques allow
combining/reconciling of survey and reporting
systems, an important problem in transition
countries. Example cultivated land area
11
3. Combining sources to produce new (small area)
statistics where there are none currently
available. There are two situations the area
was not sampled (e.g the 2412 857 non-sample
counties in RHS), or the questionnaire does not
support computation of the statistic (per capita
kcalorie consumption from the ag
census). Potential solution Small area
estimation techniques. y bo b1x1 b2x2
bpxp The xs should be available from both
the sample survey and the auxiliary source such
as the Agric. Census, or NBS, or MOA county
indicators.
12

• Example 1 Estimate for each of the 2412 counties
  the proportion of the population with consumption
  lt y 2100 kcalories/day (that is the proportion
  or number of undernourished).
• Estimate y for each of the 857 RHS sample
  counties.
• Choose explanatory variables present in both RHS
  and auxiliary source run regression using RHS
  data (both y and xs) test for goodness of fit
  and search for good fit iteratively.
• Plug the xs from auxiliary source to estimate y
  for the 2412 857 counties.
• For sensitivity/ assessment, compare for the 857
  RHS counties direct estimates with the regression
  estimates using the xs from the auxiliary
  source.
• Note County in above example can be replaced by
  township or village.

13

• Example 2. Use household models (similar to
  World Bank approach in its Poverty Mapping
  Project).
• Suppose y is household per capita income,
  available from sample households in a survey
  only, but not in census.
• Choose xs available from both survey and census
  the timing of both sources should be as close as
  possible.
• Run a regression of y (or log y) on xs using the
  household level survey data. Test goodness of fit
  and run iteratively.
• Calculate predicted ys household by household,
  using the census. It is not recommended that
  these the used directly for targeting households
  instead,
• Count the number of households in the village
  whose predicted ys fall below a threshold. Or
  you can have an estimated income distn. for
  villages, townships, counties.

14

• ADB WPI for Villages. Uses scores for 8
  indicators.
• Livelihood poverty grain output/person/year,
  cash income per person per year, of bad quality
  houses
• Infrastructure poverty of hhs with access to
  potable water, of natural villages with
  reliable electricity, of natural villages with
  all-weather road to town.
• Human resource poverty of women with long-term
  health problems, of eligible children not in
  school
• The higher the index, the poorer the village.
• Where to get the indicators, short of collecting
  them directly?

15
Participatory poverty surveys These are not meant
to replace , but can only supplement, statistical
sample surveys like RHS. At some point in the
assessment and monitoring process you will need
statistics that can be subjected scientific
inference procedures. The results of surveys like
RHS possess important properties like
replicability, which pps do not in general.
16
The aims in inviting CAS to show and discuss
their GIS were simply to inform at this point
that state-of-the-art capability exists in China,
hence no need for external professional
assistance a tool/platform exists in which to
integrate and share databases, hence no need to
reinvent elsewhere CAS geophysical,
environmental, etc. data get integrated with the
predominantly socio-economic data from NBS, MOA,
etc. Unfortunately, this led to much of
yesterdays time being consumed on targeting, how
to do it, and what indicators are more
appropriate for it that the indicators should be
identified first. All these are important, and
OLGPR will need the results of analysts
researches on this matters in the coming years.
These belong in FIVIMS medium-to-long-term
objectives. But these researches will need data,
so we should get back this morning to the
short-term objective , which is to deliver BABY
FIVIMS. We could fuzz about the baby after it is
born.
17

• THE WAY AHEAD
• Start small, simple.
• How? Prototype in one province, a small province
  not far from Beijing, with good statistical
  infrastructure (Ningxia?) Why? To maximize
  probability of successful birthing. This is a
  little more ambitious than Mr. Xus proposal
  yesterday (two counties).
• How? Establish network, but keep it within key
  focal points initially NBS, MOA, OLGPR, CAS,
  (MOH?). Why? Maximize probability of successful
  prototype.
• How? Priorities are to get agreements to share
  databases, implement, and make integrated
  database available in a user-friendly DBS
  platform. Why? To maximize probability . So
  analysts will have data to work with and pursue a
  research agenda to help OLGPR among others in
  evolving methodologies for the new PA policy.
  Research agenda in second phase of TCP?

18

• How? Dont get bogged down by issues like what
  data/indicators to include/exclude. Initially,
  the more the better. Culling and expanding (e.g.
  derived small area estimates) should happen
  incrementally and gradually from results of
  comparative studies of data quality and which
  indicators are analytically more appropriate, for
  example.
• How? Individual persons in the focal points
  prime movers or leaders need to be identified
  to get FIVIMS (or any project for that matter)
  off the ground and keep it going.
• How? Technical assistance during the formative
  years. FAO to play lead role in
  soliciting/providing assistance Why? To support
  pilot project in one province, a research agenda,
  and to provide incentives to focal points and
  prime movers.

19

• VAM/WFP/IFAD.
• This is a Province and county GIS.
• Database land use, crop yield, food production,
  husbandry production, population, labor force and
  some geographic and climatic information.
• Has township and village data in WFP/IFAD project
  areas.
• Statistical capabilities PCA, CA, ranking,
  z-scores, that can be used for vulnerability
  mapping and analysis subject to data
  availability.
• WFP/China will be phased out in 2005.

20
World Bank is currently discussing with NBS re
poverty mapping in one pilot province. Will
apply usual WB methodology based on combining
sample survey and census data using a household
model to produce poverty indicators at village
level.
21

• Ms. Wang Pingping new data collection is not
  recommended data collection capacity is very
  weak below county small area for counties and
  townships ok, but difficult for villages.
• When a theoretical and field expert in data
  collection speaks, we should listen and take
  heed. But yesterdays discussion covered so much
  ground that there may be risk of losing sight of
  the immediate goal(s).
• Immediate goal is to make the existing
  FIVIMS-related data available to more agencies
  and analysts, in particular the agri census and
  RHS, but also other databases in NBS, MOA, MOH,
  CAS, etc.
• Personal view collect new data if and only if
  there is consensus that the present level of
  knowledge needs augmenting or updating.
• Which leads to another important issue How often
  do you do the targeting?(as opposed to assessment
  and monitoring?) How often should poverty, food
  insecurity and vulnerability information be
  updated? At what levels of disaggregation? This
  is important review and decision needed at some
  point, but not here now or perhaps not even next
  year.

22
UNDP has agreed in principle to provide a
technical assistance grant to OLGPR for capacity
building in poverty data collection and analysis,
monitoring and evaluation poverty targeting
methodology and to develop a poverty database
system initially at county level, with future
plans to expand it to townships and
villages. Highlights need/importance of
coordination among donors FIVIMS network members
and focal points.
23
Last, but not least, a prototype FIVIMS that
successfully and convincingly demonstrates its
value added to improving the efficiency of the
countrys poverty, food insecurity and
vulnerability alleviation programs is a necessary
condition for the Government to put in place the
mechanisms (high level coordinating body) and the
resources required to support the development of
a national CHINA FIVIMS.