Sampling Procedures
Sampling Procedures
All national surveys in the Asian Barometer Survey have a sample size of 800 to 3200
respondents (adapted to the size of the population), which allows a minimum
confidence interval of plus or minus 3 percent at 95 percent probability. The target
respondents are voting age adults, randomly sampled according to PPS method.
Taiwan Survey
(a) Population
The target population is defined as ROC citizens who are 20 and over 20 years of
age and have the right to vote, that is, the sample is of voting age and has not been
disenfranchised due to mental illness, or incarceration. The target population was
sampled according to the PPS method.
Using PPS method voters were stratified and leveled into eight divisions. Each
division was Systematically Sampled. Within each division four, six, or eight
Towns/Countries were selected, from these two Villages/Li were selected, and out of
the villages between thirteen and sixteen people were sampled. However, in the
municipalities of Taipei and Kaohsiung, only Li and individuals were sampled. The
total sample is displayed in the Table A-1 below:
1
3 73 1,729,024 11.2% 157 6 x 2 x 13 = 156
4 51 2,603,242 16.9% 237 8 x 2 x 15 = 240
5 48 1,889,900 12.3% 172 6 x 2 x 14 = 168
6 95 2,372,053 15.4% 216 8 x 2 x 14 = 224
7 (Taipei) 444* 1,914,915 12.4% 174 30 x 6 =180
8 (Kaoshiung) 463* 1,042,117 6.8% 95 16 x 6 = 96
Total 15,422,119 100.0% 1402 1416
Note: The Taiwan project does not have the authority to access the personal
information required to select individuals to survey. Instead that responsibility fell to the
Academia Sinica.
Both age and education failed the chi-square test. This means that the data for
age and education are inconsistent with the whole population. The problem is rooted
in the sampling. In order to rectify the data, Raking, a kind of weighting method in
accor dance with multiple variables, was used to render the samples gender, age and
education data consistent with the entire population. Table A-3 is the result after data
weighted which shows Chi-square test and the significance (p-value) of three
characteristics passed.
2
Education Elementary 374 413.3
(a) Population
The target population of the survey is Hong Kong people aged 20 to 75 residing
in permanent residential living quarters in built-up areas. For practical reasons, the
land population residing in temporary structures in non-built-up areas, marine
population, transients, inmates of institutions and non-Chinese speaking population
were excluded from the survey.
3
(b) Sample Design
The sampling method involved a multi-stage design, with the following three
separate stages of selection.
1. Selection of addresses
With the replicated systematic sampling method, a sample of 2,000 residential
addresses from the computerized Sub-Frame of Living Quarters maintained by the
Census and Statistics Department was selected in June 2001. In selecting the sample,
living quarters were first stratified with respect to area and type of housing. The
sample of quarters selected is of the EPSEM (equal probability of selection method)
type and is random in the statistical sense.
2. Selection of households
Where there were more than one household that has persons of age 20 to 75 or
with a group household (such as hostel) at the selected address, a random numbers
table pre-attached to each address was used to select one household or one person.
Details of the sampling results are summarized in Table B-1. After exclusion of
the unused and invalid addresses, the effective sample size is 1,651. In all, 811
successful interviews were obtained, yielding a response rate of 49.12%.
Non-Chinese 32 Incomplete 18
Non-residential 26 Others 40
4
No such address 11
where n is the number of respondents on which the percentage is based. With the
sampling error obtained, confidence intervals can be calculated. For example, a 95%
confidence interval is
If the data are not weighted such formulae give only rough estimates.
(d) Goodness-of-Fit
SPSS nonparametric Chi-square tests were conducted to compare the gender, age
and educational attainment of the sample with that of the target population as reported
in the Hong Kong 2001 population census. The results are given in Table B-2 and
Table B-3.
As shown in Table B-2, the gender and educational attainment distributions of
the sample do not differ significantly from that of the target population.
Age failed the Chi-square test. The results reveal that the younger age group
(aged between 20 and 39) is under-represented while the more mature group (aged 40
and above) is over-represented. This could be due to the fact that younger people are
more active and are less likely to stay home.
5
Sex Male 369 390.6 2.3 P = 0.129
Female 442 420.4
China Survey
Sampling Design
The mainland data comes from a survey conducted in China between March
2002 and August 2002 in cooperation with the Institute of Sociology of Chinese
Social Science Academy. The sample represents the adult population over eighteen
6
years of age residing in family households at the time of the survey, excluding those
living in the Tibetan Autonomous Region. 1 A stratified multistage area sampling
procedure with probabilities proportional to size measures (PPS) was employed to
select the sample.
The Primary Sampling Units (PSUs) employed in the sample design are counties
(xian) in rural area and cities (shi) for urban areas. In direct municipality, districts
(qu) are used as PSU. Before selection, counties were stratified by region and
geographical characteristic and cities or districts by region and size. A total of sixty-
seven cities or districts and sixty-two counties were selected as the primary sampling
units. The secondary sampling units (SSUs) were townships (xiang) and districts (qu)
or streets (jiedao). The third stage of selection was geared to administrative villages
in rural areas and neighborhood committees (juweihui) or community committee
(shequweiyuanhui) in urban areas. We selected 249 administrative villages and 247
neighborhood or community committees in the third stage of the sampling process. A
total of 496 sampling units were selected. Household were used at the fourth stage of
sampling.
In the selection of PSUs, the National Statistic Yearbook (National Statistic Bureau
1999) was used as the basic source for constructing the sampling frame. The number
of family households for each county or city was taken as the measure size (MOs) in
the PPS selection process. For the successive stages of sampling, population data
were obtained either from the All China Women Association (ACWA). In 2000, the
organization organized a survey on womens status in China. For the survey, ACWA
asked local chapter of collect sample data in each province for their survey. We
bandwagoned on the ACWA survey and contacted the ACWA to collect sampling data
for us in their survey. For areas ACWA survey was not covered, we ask local ACWA
chapter to collect sampling data for us. All village and neighborhood committee
levels, lists of household registration (hukou) were obtained. The lists are used as the
sampling frame for the fourth stage of the sampling process.
The project scheduled interviews with 3,752 people, and 3,154 of the prospective
respondents contacted by interviewers answered our questions, for a response rate of
84.1 percent. The response rate for urban areas is lower than that for the rural areas.
For urban area, the response rate is 82.5 percent and the response rate for rural area is
1
To conserve resources, we decided to exclude Tibet from this study. First, many Tibetans do not
speak Chinese. Second, transportation in Tibet is extremely difficult since there is no railroad and the
highway system is not well developed. And third, it is difficult to find qualified interviews who can
work there effectively.
7
86.5 percent.
Thailand Survey
Sampling Plan:
The sampling plan for the study consisted of two-stage cluster sampling, plus
systematic sampling of the final populations. The stages are:
8
establish that the data is consistent with the distribution among the entire population. 2
Three sample characteristics are addressed: gender, age and region. Below are three
tables highlighting the Chi-square result, the degrees of freedom and the significance
(p-value) of each characteristic.
2
Comparison of means with expected results from the decennial census conducted by the National
Statistical Office produced the following comparisons.
9
NORTH 297 297 p > .99**
NORTHEAST 514 504
SOUTH 197 193
Total 1545 1545
** less than 1 % probability that there are differences between the sample distribution
and population distribution on this variable.
Only age failed the chi-square test. This means that the data for age are
inconsistent with the whole population. The problem is that the young people were
less sampled. If young persons are not available at the time, the substitution process,
going to the person on either side of the voting list, will likely produce an older
person if a disproportionate number of young people are absent from the home.
However, tests for age effects can be performed to see if there are significant
differences by age. Preliminary analysis shows only minor differences in orientations
by age.
Philippines Survey
SAMPLING DESIGN
The Philippines is divided into 4 study areas: the National Capital Region
(NCR), Balance Luzon, Visayas and Mindanao. The sample size of each study area is
300 voting-age adults (18 years and older), for a total sample size of 1,200.
10
To yield representative figures at the national level, census-based population
weights are applied to the survey data. The weight projection is computed by
dividing the projected population in the area by the sample size of the same area.
Appropriate projected factors are applied so that the original population proportion is
reflected in the data tables using this formula:
Population
Pr ojection factor
No.of int erviews
(Weight)
Japan Survey
We will be interviewing about 2,500 respondents that will be randomly sampled from
the citizens of voting population in all 47 prefectures. We will use a two-stage
stratified probability random sampling method.
The South Korea survey was conducted during the month of February 2003. Of
the 2,575 voters selected by multi-stage random sampling, it completed face-to-face
interviews with a total of 1,500, registering a response rate of 58 percent. Using the
2000 report of the Population and Housing Census of the National Statistical Office,
the voting-age population was first stratified by region (Do) and the seven large cities
on the basis of their proportionate share of the national population. The island of
11
Cheju-Do, with 1.2 percent of the total population, was excluded. Then each region or
large city was stratified by its administrative subdivisions (Dong, Eup, or Myun) on a
basis proportionate to population. Within each administrative subdivision, the primary
sampling units (ban or village) were randomly selected with six to eight households
in a ban and twelve to fifteen in a village. At the household level, the interviewers
were instructed to select for interview the person whose birthday came next.
Mongolia Survey
(a) Population
12
The following methods are used in the survey sampling:
(c ) Sampling Results
Altogether
13
Khovd 78 6.7 6.1
Khuvsgul 98 3.4 8.3
Dornod 73 6.2 5.1
Education
None 2.2% 1.0% 2.2%
Primary completed 16.6% 5.0% 15.4%
Incomplete secondary 30.1% 11.9% 29.7%
Secondary completed 24.0% 25.3% 26.5%
Post-secondary trade / 10.8%
vocational school 10.7% 20.0%
University-level, with 13.8%
diploma 13.2% 35.2%
Degree 1.6% 1.6% 1.6%
Gender
Male 49.5% 47.3% 49.3%
Female 50.5% 52.7% 50.7%
14