Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National...

44
Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department of Agriculture Xian Zude Rural Survey Organization Chinese National Bureau of Statistics

Transcript of Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National...

Page 1: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Improvements Of Sample Design For Rural

Statistical Surveys In China

Michael SteinerNational Agricultural Statistics Service

United States Department of Agriculture

Xian ZudeRural Survey Organization

Chinese National Bureau of Statistics

Page 2: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Chinese Census of Agriculture

• First National Agricultural Census --- 1996

• Census Questionnaires – 38 sections, 687 data items

• Data collection – January 1997

• Data collected for approximately 214,000,000 rural households

• Approximately 7,000,000 interviewers were utilized for data collection

Page 3: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Use of Chinese Census Data

• Provides a wealth of crop, livestock and rural household statistics.

• Provides estimates of small administrative units.

• Provides estimates of rare commodities.

• Provides data for a sampling frame.

Page 4: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Chinese Agriculture Statistics

• National Bureau of Statistics (NBS) of PR China

• Food and Agricultural Statistics Centre (FASC) of NBS --- Chinese Census of Agriculture

• Rural Survey Organization (RSO) --- Agricultural & Rural Statistics for China – established in 1984

• Since 1996 --- NBS and the National Agricultural Statistics Service (NASS) of the U.S. Department of Agriculture --- cooperative agreement.

Page 5: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Chinese Survey Program

• Household Survey – farmers income and expenditures – since 1954

• Crop Yield Survey – since 1963

• 857 Sample Counties Were Selected – 1984 • Three types of surveys implemented: * Farmers Household Survey * Crop Yield Survey * Socio-economic Survey

Page 6: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

RSO Plan For Survey Expansion

• RSO decided to expand survey program in 1999.

• Expanded survey program to cover: * Crop Area for Major Crops * All Major Types of Livestock (Inventory and Slaughter) * Agricultural Prices and Costs * Poverty Measurement

• Complete Reporting System

Page 7: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Guangdong Province Pilot Survey Work

• Project Involving Guangdong Bureau of Statistics, NBS-RSO, and USDA-NASS

Guangdong Province designated as site for pilot survey work.

• Guangdong Province: * 21 Prefectures * 122 Counties * 23,870 Villages

Page 8: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Objectives

1) Effectively utilize data from the Census of Agriculture

2) To select samples using multiple variables which isnecessary to support the proposed expanded surveyprogram

3) To integrate the statistical needs for different levels ofgovernment

Page 9: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

1) Effectively utilize data from the Census of Agriculture

A) Vast amounts of data available from the Census

* Census questionnaire 38 sections 687 items

B) Data available for many administrative areas

* 31 Provinces, Municipalities and Autonomous Regions * 550+ Prefectures and Cities at Prefecture Level * 2,500+ Counties and Cities at County Level * 43,000 Township and Towns * 740,000+ Administrative Villages * 214,000,000 Households

Page 10: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

2) To select samples using multiple variables which isnecessary to support the proposed expanded surveyprogram

* Land in farm * Crops * Livestock * Aquaculture * Labor

Page 11: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

3) To integrate the statistical needs for differentlevels of government

* National * Provincial * Prefecture * County

Separate samples are now used for each administrative level

Samples are not additive

Page 12: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Census Data Useful For

Analysis

Sampling Alternatives

Page 13: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

0

500

1000

1500

2000

Tho

usan

ds

Number of Persons in Rural Households, by County, 1996Guangdong Province

In Sample Not Sample Mean Median

Page 14: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

0

100

200

300

400

500

600

700

800

900

1000

Tho

usan

ds

Cultivated Land, by County, 1996Guangdong Province

In Sample Not Sample Mean Median

Page 15: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

0

200

400

600

800

1000

1200

Tho

usan

ds

Grain Crops: Area Planted, by County, 1996

In Sample Not Sample Mean Median

Guangdong Province

Page 16: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

0

50

100

150

200

250

300

Tho

usan

ds

Vegetables: Area Planted, by County, 1996

Guangdong Province

In Sample Not Sample Mean Median

Page 17: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

0

100

200

300

400

500

600

700

Tho

usan

ds

Hog Inventory: Number of Head on Hand, December 31, 1996, by CountyGuangdong Province

In Sample Not Sample Mean Median

Page 18: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Alternative 1

• Select samples from villages within all counties in a province.

Page 19: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Alternative 2

• Select villages within a NEW sample of counties.

• Continue the practice of having samples only in selected counties (not all counties).

• Select a new sample of counties and replace the old sample of counties.

Page 20: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Alternative 3

• Select villages with the current sample of counties.

Page 21: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Using Census Data to Analyze Sampling Strategies

Alternative 1 — Villages within all counties in theprovince.

Alternative 2 — Villages within a new sample ofcounties.

Alternative 3 — Villages within counties currentlyutilized in Rural Household Survey.

Yk,j is the estimate for the jthth replicate and Yk k is the known population total for thesimulated survey response for item.

RMSEk 1000

j 1(Yk,j Yk)2

1000

Page 22: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Comparison of Sample Design Strategies

Ratio of RMSEs (root mean square errors)

Stages of Sampling

Total Area Sown

Total Area for Grain

Total Area for Rice

All Counties, Villages

1 1 1

New Counties, Villages

1.8 3.6 3.1

Old Counties, Villages

5.4 5.6 5.4

Page 23: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Methods of Sampling

Page 24: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Stratified Sampling

***Census Data Available For Stratification***

Disadvantages:

Becomes difficult to create efficient stratified

design when number of variables of interest

increases.

Page 25: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Possible Commodities For Sample Selection

• Land in Grain• Wheat• Rice• Corn• Tuber Crops• Rapeseed• Peanuts• Vegetables

• Orchard Area• Pond Area

• Cattle• Sheep• Hogs• Poultry

Page 26: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Stratified Design ProblemTwo Commodities – Three Size Groupings

ItemsRice Area

(Large)

Rice Area

(Medium)

Rice Area

(Small)

Hog Inventory – Large

L L L M L S

Hog Inventory -Medium

M L M M M S

Hog Inventory

- SmallS L S M S S

Page 27: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

MPPS

Methods of Sampling

Page 28: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

MPPS Sampling

Multivariate Probability Proportional to Size:

• Probability proportional to size sample design

in which the measure of size is determined

by more than one variable.

Page 29: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Probability Proportional to Size (PPS) Sampling

• A sample is said to be chosen with probabilityproportional to size if the probability of selectionfor each unit in the population is proportional tosome measure of the size of the unit.

Page 30: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Determining Probability ofSelection for MPPS Design

• Probability = Max (PPS1, PPS2, ...,PPSK)(for 1 to k commodities)

• Sample Weight = 1 / Probability

Page 31: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Determining Probabilities ofSelection for MPPS

• Probability =

Total for State 1( Farm control 1

Max ,..., Farm control K

Total for State K)n1 * nK *

(for 1 to K commodities)

Page 32: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

1. Select a GENERAL sample with n=5 for cropland and n=5 for capacity

Record Control Control Relative Relative Probabilit

yProbabilit

y Max Prob Random In ExpNumber Data Data Measure Measure of of of Number Sample Factor

Cropland Capacity Col 2 / Col 3 / Selection Selection Selection = 1 1 / col 8(000) 10000 10000 Col 4 *n Col 5 *n Max(6,7)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)1 2000 1200 0.200 0.120 1.000 0.600 1.000 0.1603 1 1.0002 1800 800 0.180 0.080 0.900 0.400 0.900 0.98563 1500 700 0.150 0.070 0.750 0.350 0.750 0.2247 1 1.3334 1400 600 0.140 0.060 0.700 0.300 0.700 0.4889 1 1.4295 800 700 0.080 0.070 0.400 0.350 0.400 0.0972 1 2.5006 600 300 0.060 0.030 0.300 0.150 0.300 0.86417 500 0.050 0.000 0.250 0.000 0.250 0.72998 400 500 0.040 0.050 0.200 0.250 0.250 0.98749 300 0.030 0.000 0.150 0.000 0.150 0.1318 1 6.667

10 250 150 0.025 0.015 0.125 0.075 0.125 0.153011 200 100 0.020 0.010 0.100 0.050 0.100 0.295212 100 0.010 0.000 0.050 0.000 0.050 0.382913 60 100 0.006 0.010 0.030 0.050 0.050 0.228314 50 50 0.005 0.005 0.025 0.025 0.025 0.438215 30 0.003 0.000 0.015 0.000 0.015 0.657916 10 0.001 0.000 0.005 0.000 0.005 0.282517 1800 0.000 0.180 0.000 0.900 0.900 0.2366 1 1.11118 1500 0.000 0.150 0.000 0.750 0.750 0.845919 1000 0.000 0.100 0.000 0.500 0.500 0.0659 1 2.00020 500 0.000 0.050 0.000 0.250 0.250 0.9685

Total 10000 10000 1 1 5 5 7.47 7 16.040

Sampling Exercise: Multivariate Probability Proportional to Size (MPPS)

Page 33: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Multivariate Probability Proportional To Size Sampling (MPPS)

πi min 1, max n1

x3/41,i

Ni 1

x3/41,i

, ... ,nK

x 3/4k,i

Ni 1

x3/4k,i

(1)

Yk Nj 1

xk,j

ni 1

wi Yk,i

ni 1

wixk,i

(2)

V(Yk)

Nj 1

xk,j

2

ni 1

wix

k,i

2 n

i 1w 2

i e2k,i where e

k,i Yk,i x

k,i

nl 1

wl Yk,l

nl 1

wl xk,l

and wi

1πi

(3)

Page 34: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

3) To integrate the statistical needs for differentlevels of government

* National * Provincial * Prefecture * County

Separate samples are now used for each administrative level

Samples are not additive

Page 35: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample Sizes – MPPS Design

Separate samples were selected in order to accommodate different levels of government

• Part A --- Funded by Provincial Government

• Part B --- Funded by Prefecture Governments

• Part C --- Funded by County Governments

Page 36: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample Sizes

• Part A --- 1000 Villages

• Part B --- 2000 Villages

• Part C --- 3000 Villages

Page 37: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample Options

(1) Part A ---- 1000 Villages

(2) Parts A and B ---- (1000 + 2000) 3000 Villages

(3) Parts A, B and C --- (1000 + 2000 + 3000) 6000 Villages

Page 38: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample SizesSample of Villages – Three Levels Of Government

High Priority Items

Items Province Prefecture County

Grain Area 450 40 20

Vegetables 500 50 30

Hogs

Poultry

450

500

40

100

20

40

Page 39: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample SizesSample of Villages – Three Levels Of Government

Medium Priority Items

Items Province Prefecture County

Tuber Crops

250 20 8

Orchards 350 70 30

Cattle 250 25 8

Page 40: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample SizesSample of Villages – Three Levels Of Government

Specialty Items

Items Province Prefecture County

Pond Area 200 15 5

Peanuts 150 5 5

Page 41: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Sample Selection Procedure

First Stage:

• Selection of villages (Using MPPS)

Second Stage:

• Selection of Households (random stratified)

* “Large” Households

* Other Households

Page 42: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Guangdong Province Pilot Surveys

• First pilot survey conducted in 2000 --- villages selected using MPPS in select counties

• Test survey of villages conducted in three prefectures in 2001, using MPPS for village selection

• Survey of villages (MPPS sample selection) conducted in entire province in 2002.

• Beginning in 2002, Households were sampled in selected villages.

Page 43: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Guangdong Province Survey - 2003

Commodity C.V. (%)

Grain Area

Vegetable Area

Tuber Crop Area

Orchard Area

Peanut Area

3.58

4.89

6.53

7.38

6.86

Hog Inventory

Poultry Inventory

5.62

6.19

Net Income

Per Capita 4.67

Page 44: Improvements Of Sample Design For Rural Statistical Surveys In China Michael Steiner National Agricultural Statistics Service United States Department.

Future Of Agricultural Statistics In China

Expansion of MPPS Procedures - Nationwide:

• Crop Area Planted Survey.

• Livestock Survey

Agricultural Census:

• 2006