March 2019

MYANMAR LIVING
CONDITIONS
SURVEY 2017
REPORT

02 TECHNICAL REPORT
 March 2019

MYANMAR LIVING
CONDITIONS
SURVEY 2017
REPORT

02 TECHNICAL REPORT

This report has been possible because of the generous financial support of the governments of Australia, Denmark,
Finland, Ireland, Korea, Sweden and the United Kingdom.

KWPF
KOREA-WORLD BANK
PARTNERSHIP FACILITY
 Acknowledgements
I would like to express my deepest gratitude to His Excellency Union Minister U Soe Win, Ministry of
Planning and Finance, for his valuable support, guidance and encouragement though the process of
undertaking the Myanmar Living Conditions Survey.

This is an exciting time for the development of statistics and the statistical system in Myanmar. A
new statistics law has been formulated which strengthens the relationship of statistics producers
and stakeholders. In addition, a National Strategy for the Development of Statistics (NSDS) has been
formulated, setting a clear path and concrete milestones for developing quality and accurate official
statistics in Myanmar.

Myanmar’s official statistics need to be credible in order to fulfill their important task to describe
the socio economic conditions in our country. They must be based on clear, publicly stated operating
principles and methodologies. These conditions guarantee the quality of statistics and improve the
trust between data providers and data users.

I would like to thank all those who have worked hard and with a firm commitment to undertake
the MLCS 2017. The financial and technical support of the UNDP and the World Bank is greatly
appreciated. I also express my gratitude to all the 13,730 households who supported the survey in the
time-consuming data collection exercise. They have provided much-needed information that will be
widely used by a variety of data users.

U San Myint Ministry of Planning and Finance

Office No.32, Nay Pyi Taw, Myanmar
Director General Phone : +95 67 406325, +95 67 406328
Central Statistical Organization Fax : + 95 67 406151
Email : [email protected]
Ministry of Planning and Finance [email protected]
Website : https://www.csostat.gov.mm
http://www.mmsis.gov.mm
Contents
Acknowledgements

Acronyms 1

Field Enumeration Summary 1

01 Introduction 3

1.1 Introduction 4
1.2 Timing of Key Events 7

02 Survey Design 9

2.1 Introduction to the Sample 10

2.2 Selection of Sample EAs 11
2.3 Final Design of MLCS 2017 Sample 13
2.4 Selection of quarterly Sub-samples 15
2.5 Segmenting EAs containing a large number of households 16
2.6 Listing and households selection 18
2.7 Questionnaire design and content 19

03 Field Organization 25

3.1 Data collection staff 26

3.2 Data processing 28
3.3 The survey teams 29
3.4 Training 31
3.5 Replacement of EAs that could not be enumerated 33
3.6 Use of interpreters during interviews 37
3.7 Replacing survey staff 38

04 Quality Assurance 41

4.1 Field Trips 42

4.2 QA reports generated from the in-field data entry program 45
05 Weighting and sampling error calculation 47

5.1 Weighting procedures 48

5.2 Weighting to take account of non-response 50
5.3 Final Response Rate 51
5.4 Survey Estimates from MLCS 2017 data 51
5.5 Calculation of Sampling Errors 52

Annexes 55

Annex A: Sample size determination for MLCS 2017 56

Annex B: MLCS 2017: Listing Form 67

Annex C: Non-enumerated and non-visited EAs and reason 68

Annex D: CSO and non CSO people used in listing and 70

survey teams

Annex E: QA results from in-field data entry program 71

Tables
Table 2.1: Distribution of Private Households by Region and Urban/Rural Stratum. 10
Preliminary Results of 2014 Myanmar Census of Population and Housing

Table 2.2: Distribution of All Enumerated EAs by Size (Number of Private Households) 11
and Urban/Rural Stratum, 2014 Census. Preliminary Results of 2014 Myanmar
Census of Population and Housing

Table 2.3: Distribution of Myanmar Master Sample PSUs by Region and 12

Urban/Rural Stratum

Table 2.4: MLCS 2017 Final Sample Design by State/Region, Urban and Rural Stratum 13

Table 2.5: Issued and actual quarter differences by cluster 16

Table 2.6: Segmented EAs 17

Table 2.7: EAs with more than four non-interviews after using all four replacement 18
households

Table 2.8: Overview of the MLCS 2017 Household Questionnaire and Length of 21
Interviewing Time

Table 2.9: Overview of the MLCS 2017 Community Questionnaire 22

Table 3.1: Summary of field staff tasks and training periods 26

Table 3.2: Summary of Data Processing staff based in Nay Pyi Taw 29

Table 3.3: Language in which interview was conducted 38

Table 4.1: Quality Assurance Visits 43

Table 5.1: Completed Household Interviews for MLCS 2017 by State/Region 50

Maps
Map 3.1: Gender distribution of the supervisors, enumerators, in-field data entry 30
operators and listers by team

Map 3.2: Replaced and not visited EAs by State/Region and Quarter 34

Map 3.3: Areas of Myanmar not enumerated in MLCS 2017 36

Figures
Figure 3.1: Number of applicants for interviewing and listing jobs by State/Region 27

Figure 3.2: Level of experience before and after training by aspect of the survey process and role 32

Figure 3.3: Percentage of EAs within each State/Region where interpreters were used 37
Acronyms
CAFÉ Computer Assisted Field-based data Entry (CAFE)

CSO Central Statistical Organization

DG Director General

DP Development Partner

DoP Department of Population

EA Enumeration Area

GoM Government of Myanmar

HIES Household Income and Expenditure Survey

IHLCA Integrated Household Living Conditions Assessment

MLCS Myanmar Living Conditions Survey (2017)

MPLCS Myanmar Poverty and Living Conditions Survey (2015)

NGO Non-Governmental Organization

NSDS National Strategy for the Development of Statistics

PAPI Paper and Pencil Interviewing

PPS Probability Proportional to Size

PSU Primary Sampling Unit

QA Quality Assurance

UNDP United Nations Development Programme

WB World Bank

Field Enumeration Summary

Number of EAs enumerated 1,145
Total number of households enumerated 13,730
Number of Replaced EAs 33
Number of EAs not visited at all 7
Number of refusing households 99
Number of household non-contacts after 3 calls 802
Number of derelict/empty dwellings 37
Number of EAs in which interpreter used 1
99
Number of EAs in which Quality Assurance visits took place 65

1 Not necessarily all of the households in the EA needed interpretation, but at least one did.
2
 01.
INTRODUCTION

3
 1.1 Introduction
This report is one in a series of products that will be produced from the 2017 Myanmar Living
Conditions Survey (MLCS). The objective of this report is to outline the survey procedures for
readers who would like to know the MLCS 2017 activities in detail. The report outlines the work
undertaken at all stages of the survey from questionnaire development and testing through to
processing the data files.

The report focuses on various stages of the survey work, recruiting and hiring data collection and
processing staff, training, questionnaire design, sample design and data processing procedures.

This report will be complimented by analytical products. Key indicators report has been published.
Poverty profile report and a detailed socio-economic report will subsequently feature detailed
analysis of living conditions in Myanmar based on MLCS 2017.

The MLCS 2017 is a comprehensive study of how people in Myanmar live. It was carried out by the
Central Statistical Organization (CSO) with technical and financial support from the United Nations
Development Programme and the World Bank. It collects data on the occupations of people, how
much income they earn, and how they use this to meet the food, housing, health, education and
other needs of their families. The data collected can be used to formulate responsive policies for the
future development of the country.

The Myanmar Living Conditions Survey has the following objectives:

• To provide updated estimates of poverty and living conditions at the national, urban/rural and
state/region level;
• To inform national data needs and selected SDG targets;
• To construct consumption weights for the national and regional Consumer Price Index (CPI)
baskets; and
• To estimate private consumption expenditure for the System of National Accounts.

The MLCS builds off earlier household expenditure and living conditions surveys conducted in
Myanmar, in particular the Integrated Household Living Conditions Assessment (2004/5 and
2009/10), the Household Income and Expenditure Survey (conducted 5 times, every 6 years between
1989 and 2012) and the Myanmar Poverty and Living Conditions Survey (2015). The Myanmar Living

4
Conditions Survey brings all these previous household surveys together into a single survey, and
provides one comprehensive source of living conditions information.

The MLCS 2017 is representative of the Union Territory, its states and regions and urban/rural
areas. It was enumerated in all the districts and 296 of the 330 townships of Myanmar. In total
13,730 households participated in the survey. The survey was a representative sample for Myanmar
of 1,145 enumeration areas.2 The sample was based on the 2014 Population and Housing Census
(Census) frame. Sampling weights were used to make estimates representative of the population
and the sample provides statistics for the fourteen states and regions and Nay Pyi Taw Council of
Myanmar.

The survey was conducted continuously over a 12-month period from late December 2016 to
November 2017. Interviewing began in the winter season (December to February) continued
throughout the dry season (March to May) and the monsoon season (June to October), ending in
the winter season of 2017.

Throughout the development of the questionnaire there was a challenge to keep questions the same
as MPLCS, to enable trend indicators to be created, or to alter/improve questions to take account
of the changing circumstances in Myanmar. To get advice on this important task, wide ranging
discussions were held with representatives from many Ministries, Development Partners, NGOs and
academics working in Myanmar. In the Data User Consultation meetings the large groups were
broken down into smaller working groups to discuss the following sections of the questionnaire:

• Household composition and demographics

• Education, literacy and training
• Health
• Housing
• Household Consumption Expenditure
• Household Durables
• Labour and Employment
• Agricultural activities
• Non-farm businesses
• Finance
• Shocks and coping strategies
• Migration & Remittances
• Other Income
• Community Questionnaire

In terms of a Steering Committee, the MLCS 2017 utilized the Central Committee for Data Accuracy
and Quality of Statistics. The remit of this committee is to develop evidence-based policy and
planning. The meetings are held to enable effective cooperation for accurate and quality statistics.
The Vice President (II) is the Patron. The Union Minister of the Ministry of Planning and Finance
Chairs the committee and the DG of CSO is the Secretary. The Union Ministers of the following
Ministries are members:

2 Outreach activities took place over the 12 months of data collection but it was not possible to interview in Northern Rakhine
and the Wa Self-Administered Area. These exclusions are fully documented in the forthcoming MLCS 2017 Survey Content and
Quality Report and can be seen in the maps presented within this report.

5
 • Ministry of Foreign Affairs
• Ministry of Home Affairs
• Ministry of Border Affairs
• Ministry of Information
• Ministry of Religious Affairs and Culture
• Ministry of Agriculture, Livestock and Irrigation
• Ministry of Transport and Telecommunication
• Ministry of Natural Resources and Environmental Conservation
• Ministry of Electricity and Energy
• Ministry of Labour, Immigration and Population
• Ministry of Industry
• Ministry of Commerce
• Ministry of Education
• Ministry of Health and Sports
• Ministry of Planning and Finance
• Ministry of Construction
• Ministry of Social Welfare, Relief and Resettlement
• Ministry of Hotels and Tourism
• Union Attorney General's Office
• Economic Related Ministries in all State and Regional Governments

The NSDS cluster on Survey Coordination and Statistical Standard Cluster is the Technical
Committee for the MLCS 2017. Selected Donor Partners were additionally invited to some of the
NSDS meetings, depending on the topics under discussion.

The sample of 13,730 households is nationwide, covering all districts and 296 townships of
Myanmar. The fieldwork period was 12-months from December 14 2016 to December 13 2017 to
ensure interviews took place in all seasons and captured the situation experienced by households at
different times of the year.

The interview was undertaken by Paper and Pencil Interviewing (PAPI) with data entered into a
laptop in the field (Computer Assisted field-based data entry, pioneered by the Living Standards
Measurement Survey (LSMS) team of World Bank). The data entry system, implemented in CSPro
software, enables data entry to first take place in the field to allow immediate automated consistency
checks. Data is then transmitted via mobile internet to a central database. Second data entry takes
place at CSO to rule out entry errors.

6
 1. 2. Timing of Key Events

Event Date Section of the report

Questionnaire development May 2016 – November 2016 3

Sample design and selection of EAs May 2016 – October 2016 3

1st Data User Consultation Meeting 14 June 2016 1

Pretest End of June 2016 3

1st Pilot July 2016 3

2nd Data User Consultation Meeting 17 August 2016 1

2nd pilot September 2016 3

Data entry system programming Sept – November 2016 3

Listers training October 2016 3

Main training (Supervisors, Enumerators and Data Entry

November 2016 3
Operators)

Data collection began 14 December 2016 3

Data entry operator and checker trainings

January 2017 3
(CSO Head Quarter)

1st Security meeting 27 February 2017 3

2 Security meeting
nd
12 June 2017 3

3rd Security meeting 17 October 2017 3

Data collection ended 13 December 2017 3

3rd Data User Consultation 29 January 2018 1

Weights finalized 30 January 2018 5

7
8
02.
SURVEY DESIGN

9
 2.1. Introduction to the Sample
A nationally representative sample of households is interviewed each quarter in order to represent
seasonality in all the survey indicators. The main geographic domains of analysis for the MLCS 2017
are the Union Territory and 14 states or regions of Myanmar. The survey results are also tabulated
for the urban and rural domains at the national level.

The sample primary sampling units (PSUs) for this sample are the enumeration areas (EAs) defined
for the 2014 Myanmar Population and Housing Census. In September 2014, the Department of
Population of the Ministry of Immigration and Population selected a representative Master Sample
of 4,000 EAs from the 2014 Census frame for the Myanmar household survey program. The sample
clusters for the MLCS 2017 are selected from the Master Sample3.

A stratified multi-stage sample design is used for the MLCS 2017. The Master Sample was stratified
by state/region, urban and rural areas. The classification of the EAs in the 2014 Myanmar Census of
Population and Housing frame by urban and rural stratum was based on the administrative structure
of the hierarchical geographic areas in Myanmar; all EAs in administrative areas defined as wards
are considered urban, and all EAs in village tracks are classified as rural. The distribution of the
households in the 2014 Myanmar Census of Population and Housing frame by region, urban and
rural stratum is shown in Table 2.1, based on the preliminary Census data.

Table 2.1

Distribution of Private Households by Region and Urban/Rural Stratum, Preliminary Results of 2014 Myanmar Census
of Population and Housing

State/Region Urban Rural Total % by Region % Urban

Kachin 91,907 174,945 266,852 2.5% 34.4%

Kayah 13,730 41,162 54,892 0.5% 25.0%

Kayin 63,951 237,614 301,565 2.8% 21.2%

Chin 19,022 71,475 90,497 0.8% 21.0%

Sagaing 179,736 909,647 1,089,383 10.1% 16.5%

Tanintharyi 63,170 207,729 270,899 2.5% 23.3%

Bago 234,228 891,334 1,125,562 10.4% 20.8%

Magway 128,363 777,702 906,065 8.4% 14.2%

Mandalay 406,173 898,061 1,304,234 12.1% 31.1%

Mon 111,929 303,825 415,754 3.8% 26.9%

Rakhine 74,120 493,336 567,456 5.2% 13.1%

Yangon 1,051,226 499,273 1,550,499 14.3% 67.8%

Shan 265,943 875,137 1,141,080 10.5% 23.3%

Ayeyarwady 198,294 1,286,144 1,484,438 13.7% 13.4%

Nay Pyi Taw 68,639 178,654 247,293 2.3% 27.8%

Total 2,970,431 7,846,038 10,816,469 100.0% 27.5%

3 The methodology for the master sample design is described in the report on “Recommendations for Developing Master
Sample for National Household Surveys in Myanmar: Sampling and Estimation Methodology” (Megill, October 2014).

10
 2.2. Selection of Sample EAs
The primary sampling units (PSUs) for the master sample are the census enumeration areas (EAs),
with an average of about 135 households each (140 for urban EAs and 133 for rural EAs). In order to
further examine the variability in the EAs by size, Table 2.2 shows the distribution of the frame of all
enumerated EAs by size category and urban/rural stratum. More than half of the EAs are within the
range of 100 to 149 households.

The EAs in the sampling frame are stratified by state/region, urban and rural strata. Within each
stratum the EAs are ordered geographically by district, township, ward or village tract and EA code,
in order to provide additional implicit stratification and ensure that the sample is geographically
representative.

Table 2.2

Distribution of All Enumerated EAs by Size (Number of Private Households) and Urban/Rural Stratum, 2014 Census.
Preliminary Results of 2014 Myanmar Census of Population and Housing

Number of EAs

EA Size Urban Rural Total

0 household 52 102 154

1-9 households 89 194 283

10-19 households 104 201 305

20-49 households 341 999 1,340

50-99 households 1,614 6,993 8,607

100-149 households 11,883 33,064 44,947

150-199 households 6,349 15,003 21,352

200-249 households 977 1,643 2,620

250-299 households 231 308 539

300+ households 133 277 410

Total 21,773 58,784 80,557

Mean 140 133 135

Std. Dev. 140.2 133.3 135.2

A Master Sample of 4,000 enumeration areas (EAs) was selected from the Census 2014 frame. The
Master Sample EAs within each stratum were selected systematically with probability proportional
to size (PPS), where the measure of size is based on the number of households in the Census frame.
The Master Sample EAs are divided into four nationally-representative replicates of 1,000 sample
EAs each. The distribution of the sample EAs in the master sample by state/region, urban and rural
stratum is shown in Table 2.3.

The sample EAs for the MLCS 2017 were selected as a subsample of the full Master Sample (all
replicates), as described later in the section on Sample Selection Procedures. Within each sample
EA selected for the MLCS 2017, the frame is updated with a new listing of households that is used at
the last sampling stage for selecting a sample of 12 households in each sample EA.

11
 The units of analysis for the MLCS 2017 are the individual households and persons who are usual
residents of the households.

Table 2.3

Distribution of Myanmar Master Sample PSUs by Region and Urban/Rural Stratum

Master Sample EAs

State/Region Urban Rural Total

Kachin 80 96 176

Kayah 40 60 100

Kayin 52 136 188

Chin 40 64 104

Sagaing 76 280 356

Tanintharyi 56 124 180

Bago 100 264 364

Magway 60 264 324

Mandalay 160 232 392

Mon 80 140 220

Rakhine 44 216 260

Yangon 272 128 400

Shan 112 252 364

Ayeyarwady 72 328 400

Nay Pyi Taw 64 108 172

Total 1,308 2,692 4,000

In order to determine the final design of the sample, the MPLCS data was tabulated to examine
the sampling errors, confidence intervals and design effects for key estimates from that data. This
analysis can be seen in detail in Annex A. Table 2.4 shows the final allocation of EAs and households
by strata.

12
 2.3. Final Design of MLCS 2017 Sample

Table 2.4

MLCS 2017 Final Sample Design by State/Region, Urban and Rural Stratum

Total Urban Rural

State/Region Sample Sample Sample Sample Sample Sample
clusters households clusters households clusters households

Kachin 72 864 36 432 36 432

Kayah 72 864 28 336 44 528

Kayin 72 864 24 288 48 576

Chin 72 864 24 288 48 576

Sagaing 72 864 20 240 52 624

Tanintharyi 72 864 28 336 44 528

Bago 72 864 24 288 48 576

Magway 72 864 20 240 52 624

Mandalay 72 864 36 432 36 432

Mon 72 864 32 384 40 480

Rakhine 72 864 16 192 56 672

Yangon 96 1,152 64 768 32 384

Shan 96 1,152 36 432 60 720

Ayeyarwady 96 1,152 24 288 72 864

Nay Pyi Taw 72 864 32 384 40 480

Total 1,152 13,824 444 5,328 708 8,496

The samples EAs for MLCS 2017 were selected from all replicates (1, 2, 3 & 4) of the Master Sample
EAs. Since the Master Sample EAs were selected with PPS within each stratum, the subsamples of
EAs for the MLCS 2017 were selected from the Master Sample with equal probability within each
region, urban/rural stratum. This ensured that the first stage probabilities for the MLCS 2017 sample
EAs were PPS within each stratum, thus improving the efficiency of the sample design by reducing
the variability of the household weights within each stratum.

In order to maintain the implicit stratification of the Master Sample for the MLCS 2017, the Master
Sample EAs were sorted in the same order in which they were selected from the census sampling
frame, that is, by stratum (region, urban/rural), district, township, ward or village tract code, and EA
code. Then the subsamples of EAs for the MLCS 2017 within each stratum were selected from the
master sample using systematic random sampling with equal probability. The sampling procedure
involved the following steps:

13
 (1) All the master sample EAs in each stratum were kept in the same order in which they had
been selected. Assigned a serial number from 1 to nh to the Master Sample EAs in the stratum.

(2) To obtain the sampling interval for selecting the subsample of n’h EAs within a stratum (Ih),
divided the number of Master Sample EAs in the stratum (nh) by the number of EAs in the
subsample for the MLCS 2017 (n’h ) found in Table 2.4 and keep at least two decimal places:

(Ih) = nh /n’h

(3) Selected a random number (Rh) with at least 2 decimal places, between 0.01 and Ih. The EAs
selected in the subsample for the MLCS 2017 were identified by the following selection
numbers:

Shi = Rh + [Ih x (i-1)], rounded up,

where i = 1, 2, 3,..., n’h

The i-th selected EA is the one with a serial number equal to Shi.

A SPSS Complex Samples application was developed to facilitate the selection of a subsample of
EAs by stratum for the MLCS 2017 from the Master Sample based on the methodology specified
above. The number of sample EAs that were selected from the Master Sample in each stratum
are shown in Table 2.4. The selection of sample EAs for the MLCS 2017 was conducted by the
Department of Population, with participation by CSO staff.

14
 2.4. Selection of quarterly Sub-samples
The sample for MLCS 2017 has a nationally representative subsample of EAs in each quarter. Given
the systematic selection of EAs within each stratum, the subsample of EAs for each quarter was
selected from the full sample systematically with equal probability. Sequential numbers from 1 to
4 were assigned to all the sample EAs within each stratum, in the same order in which they were
selected. This sampling procedure ensures that each systematic subsample within a stratum is
geographically representative.

Diagram 2.1

Each quarter of MLCS 2017 sample is nationally representative of Myanmar

Myanmar Myanmar

Quarter 1 Quarter 2

Quarter 4 Quarter 3

Myanmar Myanmar

Care was taken to try not to change the quarter in which EA was assigned to and by the end of
fieldwork only 26 (2%) of EAs had not been enumerated in their issued quarter. This was mainly due
to flare-ups in the security situation and the need to remain flexible. Table 2.5 shows the 26 EAs with
the issued quarter number and then the actual quarter in which the interviews took place.

15
Table 2.5

Issued and actual quarter differences by cluster

Cluster Issued Quarter Actual Quarter State/Region Urban/Rural

1 16 4 3 Kachin Rural

2 67 2 1 Kachin Urban

3 74 2 1 Kayah Urban

4 98 4 3 Kayah Rural

5 100 3 2 Kayah Rural

6 146 2 1 Kayin Urban

7 169 3 2 Kayin Rural

8 177 2 3 Kayin Rural

9 211 3 4 Kayin Urban

10 276 4 3 Chin Rural

11 280 4 3 Chin Rural

12 400 4 3 Tanintharyi Rural

13 431 3 4 Tanintharyi Rural

14 439 2 3 Bago Rural

15 492 3 2 Bago Rural

16 722 2 1 Rakhine Urban

17 775 3 4 Rakhine Rural

18 778 4 3 Rakhine Rural

19 782 2 1 Rakhine Urban

20 784 2 1 Rakhine Rural

21 890 2 1 Shan Urban

22 936 2 1 Shan Urban

23 941 2 1 Shan Rural

24 945 1 2 Shan Rural

25 946 2 1 Shan Urban

26 965 2 4 Shan Rural

2.5. Segmenting EAs containing a large number of

households
There were some large EAs in the sample, with more than 300 households, where it would have
taken a considerable amount of time to complete a listing. In these cases the EA was subdivided into
smaller segments, and one segment was selected to be listed.

To do this the listers first made a sketch map and conducted a quick count of the number of
housing units in order to subdivide the EA into segments of approximately equal size (about 100
households each). It was important that the segments had well-recognized boundaries such as

16
 roads, paths, streams, etc., so it was not always possible to define segments with equal size. Each
segment was labelled. The listers then telephoned CSO to receive a random number and then the
selected segment was based on the number given. In total only 12 of 1,145 completed EAs involved
segmentation. These can be seen in Table 2.6.

Table 2.6

Segmented EAs

Number of
Cluster State/Region District Urban/Rural Actual Quarter
segments

1 20 Kachin Myitkyina Rural 2 1

2 24 Kachin Myitkyina Urban 2 2

3 523 Magway Magway Rural 2 3

4 736 Rakhine Sittwe Rural 10* 2

5 814 Yangon Yangon (North) Urban 2 2

6 819 Yangon Yangon (North) Urban 2 2

7 828 Yangon Yangon (East) Urban 2 1

8 832 Yangon Yangon (East) Urban 2 1

9 849 Yangon Yangon (East) Urban 3 4

10 885 Yangon Yangon (West) Urban 3 4

11 938 Shan Lashio Rural 5 1

12 961 Shan Kyukme Rural 2 3

* This EA had more than 1,000 households

17
 2.6. Listing and household selection
A sample of 12 households was selected systematically from the Listing Form for each sample EA. All
the households in occupied housing units were assigned a Serial Number from 1 to the total number
of households listed. To select the 12 households the lister used a Household Selection Table that
specified the 12 serial numbers to be selected based on the total number of households listed. For
example (see below) if there were 19 households in the EA then the Lister ticked in column 8 of the
Listing Form (Annex B) the households numbered 1,4,5,6,8,10,11,13,14,16,18,19 in Column 7 (the Serial
Number).

HOUSEHOLD SELECTION TABLE - EXAMPLE

Number of Selected Households Replacements

households
listed in the EA 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

16 1 2 3 5 6 7 9 10 11 13 14 15 4 8 12 16

17 2 4 5 6 8 9 10 12 13 14 16 17 3 7 11 15

18 1 2 3 5 7 8 10 11 12 14 16 17 4 9 13 18

19 1 4 5 6 8 10 11 13 14 16 18 19 3 7 12 17

The Household Selection Table (generated using an Excel spreadsheet) also identified an additional
four sample households that the supervisor used for replacing households that could not be
interviewed after several attempts. For each replacement the reason for the original non-interview
was entered on a single front page of a questionnaire so that the weights could be adjusted correctly
later.

The sample household replacement procedures were controlled as part of the Quality Assurance
procedures to try and avoid selection bias in the field. There were only 8 EAs in which there were
more than four non-interviews in the sample EA (all urban) and this is why the sample size is 13,730,
rather than 13,740 (1,145 EAs x 12 interviews per EA).

Table 2.7

EAs with more than four non-interviews after using all four replacement households

Number of interviews Number of interviews not

Cluster State/Region
achieved achieved

1 827 11 1 Yangon

2 842 11 1 Yangon

3 877 9 3 Yangon

4 878 11 1 Yangon

5 881 11 1 Yangon

6 884 11 1 Yangon

7 885 11 1 Yangon

8 588 11 1 Mandalay

18
2.7. Questionnaire design and content
The guiding principles when developing the questionnaire were fourfold:

1. Subject breadth: A multi-topic questionnaire should be produced with the aim of getting
a better understanding of the correlates of poverty in its many dimensions. In this light the
questionnaire should fit information needs coming from sectors (line and core ministries).
2. Comparability: The questionnaires should, as much as possible, produce comparable core
indicators to those from HIES, IHLCA and MPLCS. Deviations from comparability should be
clearly noted from the outset.
3. SDGs: Selected indicators should be collected to form the baseline for UN Sustainable
Development Goals (SDG).
4. Community focus: There should be a strong community questionnaire to capture prices,
service delivery, facilities and infrastructure available in each EA.

With these aims in mind an extensive review of indicators and objectives of the questionnaire was
undertaken. This was conducted through a consultative process with all relevant stakeholders.
Three large and inclusive Data User Consultation meetings were held.

1st meeting on June 14 2016 had the aim of introducing the basic principles of the survey and
discussing the broad contents.

2nd meeting on August 17 2016 discussed each section in detail by breaking down into small
working groups.

3rd meeting on January 29 2018 discussed the content of the analytical reports.

The MLCS 2017 was also presented during the statistical standards and coordination working group
meetings.

Although the questionnaire is the measuring instrument upon which the success of the survey
operation depends, its development and testing are the least scientifically rigorous component
of the survey process. Despite valuable research on question form and response mode issues
conducted by many investigators (e.g. Cantril, 1944; Payne, 1951; Sudman and Bradburn, 19824), the
creation of a survey questionnaire remains largely an art, based primarily on past experience with
only a few “common sense” principles as guidance.

The MLCS 2017 questionnaire went through several developmental stages. The initial version of the
questionnaire was reviewed by the MLCS team and other advisors to determine whether it would
obtain the data required for the survey, trying to gauge whether enumerators would be able to
handle the questions with ease. A pretest and two pilots were conducted to determine whether the
individual questions and the questionnaire as a whole worked as intended. During the pretesting and
piloting fieldwork Rating Forms5 recorded instances of problems with individual questions relating
to the following four issues:

4 Cantril, H., ed. (1944). Gauging Public Opinion. Princeton University Press, Princeton, New Jersey, U.S.A. Payne, S.L. (1951).
The Art of Asking Questions. Princeton University Press, Princeton, New Jersey, U.S.A. Sudman, S., and Bradburn, N.M. (1982).
Asking Questions. A Practical Guide to Questionnaire Design. Jossey-Bass, San Francisco, California, U.S.A.
5 Rating Forms based on recommendations from Cannell, C.F., Lawson, S.A., and Hausser, D.L. (1975). A Technique for
Evaluating Interviewer Performance. Survey Research Center, University of Michigan, Ann Arbor, U.S.A.

19
 • Questions were difficult to ask. Enumerators had problems reading a question because it
had a complex sentence structure or because it contained tongue twisters or words that
were difficult to pronounce.

• Comprehension problems occurred because of the use of vocabulary that was too difficult
for the respondent or because the question did not specify clearly what information was
needed.

• A lack of a common understanding occurred when terms or concepts used in a question

were understood differently by different respondents, or the question was not interpreted
as intended.

• Difficulty in cognitive processing of information. This difficulty arose when the respondent
was unable or unwilling to exert the level of effort needed to provide an adequate answer.
Sometimes the information is simply inaccessible to the respondent. Often, the information
is accessible, but the effort needed to retrieve and process it is greater than the respondent
was willing to make.

In addition to checking individual questions, the pretest and pilots also assessed the questionnaire
as a whole. Did the questionnaire flow smoothly from one topic to the next? Did the skip patterns
work as intended? Did the juxtaposition of certain questions cause problems? Was it too long?

By the end of the consultation and testing process the MLCS 2017 household questionnaire
contained 13 modules with a total of 294 questions. Table 2.8 outlines the main modules of the
questionnaire. The full questionnaire can be seen in Appendix 1.

20
Table 2.8

Overview of the MLCS 2017 Household Questionnaire and Length of Interviewing Time.

Average time taken

Module Number of questions
(minutes) to administer

1 Household member roster 27 10

2 Education, Literacy, Numeracy and training 24 9

3 Health 16 7

4 Housing 30 6

5a Food Consumption in the Last 7 Days 9 38

5b Food Consumed away from Home 5 1

5c Non food consumption expenditure in last 30 days 3 8

5d Non food consumption expenditure in last 6 months 6 9

6 Household Durables 5 10

7 Labour and Employment 60 11

8a Parcel Roster 23 2

8b Harvest and agricultural labour 9 6

8c Inputs 2 4

8d Livestock 22 3

8e Aquaculture and fishing 12 1

8f Agric and fishing machinery and equipment 3 2

9 Ownership of non-farm businesses 20 3

10 Finance 4 4

11 Shocks and coping strategies 2 3

12 Money senders in last 12 months 10 2

13 Other income 2 1

TOTAL 294 140 minutes

It was agreed from the outset that in order to maintain the quality of the data collected the household
questionnaire should be no more than three hours of interviewing per household.

In addition to the household questionnaire, a community questionnaire consisting of 114 questions

over twelve modules was administered in each survey cluster. The community questionnaire
covered, among other things, information on availability of infrastructure, government services,
communal resources and local prices. Table 2.9 provides an overview of the questionnaire which
can be seen in Appendix 2.

21
Table 2.9

Overview of the MLCS 2017 Community Questionnaire.

Module Number of questions

1 Key informant list 5

2 Village Profile 23

3 Electricity 8

4 Access to facilities 8

5 Economic Activity 4

6 Health 4

7. Communal Organizations 2

8. Programmes 2

9 Agricultural Inputs 4

10 Shocks 1

11 Schools 39

12 Prices 14

TOTAL 114

* Time was not recorded in the community questionnaire.

The Field Manual (Appendix 3) provides a more comprehensive description of each survey section
and how they were enumerated.

22
23
24
03. FIELD
ORGANIZATION

25
 3.1. Data collection staff
One aim of the MLCS 2017 project was to establish a cadre of professional enumerators and
supervisors who would focus exclusively on survey preparation and implementation for the duration
of the project. A second aim was that the structure of the teams should strengthen the state/region
level long term involvement in survey operations.

The structure for the MLCS data collection consisted of two teams. One team consisted of two
listers and the second of one supervisor, one data entry operator and three enumerators.

Diagram 3.1

Structure of MLCS data collection teams

1 Supervisor 1 Data entry person

2 Listers

3 Enumerators

Table 3.1

Summary of field staff tasks and training periods

Number in
Role in the survey Main task When trained
each team
Lister Go into the EA one month before 2 October 2016 for one week
the interviewing team and update
the household numbers based on
the Census 2014 maps

Supervisor (CSO staff) Organize logistics for their team. 1 July 2016 for two weeks in Mandalay
Monitor all activities September 2016 for two weeks in
Taunggyi
November 2016 for four weeks in
Nay Pyi Taw

Enumerator Interview households 3 November 2016 for four weeks in

Nay Pyi Taw

In-field data entry operator Enter data into laptop in the field to 1 November 2016 for four weeks in
ensure logic and completeness of Nay Pyi Taw
each household questionnaire

26
 Previous CSO surveys had involved shorter enumeration periods with large number of enumerators
and field supervisors. These were government staffs who were temporarily assigned away from
their normal jobs to conduct temporary field work. Instead, MLCS 2017 was a longer term survey (12
months) with teams of full time enumerators, supervisors and data entry personnel. The supervisors
were recruited from CSO state/region offices and dedicated to the survey for the whole year,
becoming survey implementation experts for CSO to draw on for future surveys. While enumerators,
in-field data entry operators and listers were recruited externally.

The recruitment of enumerators, in-field data entry operators and listers was undertaken in two
stages:

1. A request for curriculum vitae’s

2. An interview for those passed the first phase of selection.

All selected candidates were required to be educated to university level and be fluent in the ethnic
language relevant to the State/Region in which they would work. The State/Region coordinators
conducted interviews. The CSO Board of Directors and Survey Department staff then reviewed the
short list to make the final selection. In total, for the 140 posts, 2,415 applications were received.
This made the first phase of selecting curriculums a time consuming process.

Figure 3.1 shows the large number of applicants came from Yangon, Sagaing, Ayeyarwady and
Kachin. The pool of candidates was lowest in Kayah, Kayin and Chin. This was mainly due to the
need for language skills in Myanmar and the local language.

Figure 3.1

Number of applicants for interviewing and listing jobs by State/Region

400 356 359

350 321
300
248
250 218
187
200
145 133
150 120
90
100 65 54 53
39 27
50
0
i
hi
n ah yin in in
g ry go ay ay on ne go
n an ad
y w
c y
Ka Ch a a Ba w al M hi Sh ita
Ka Ka ag th ag an
d k
Ya
n rw y
S nin M Ra ya yp
Ta
M e Na
Ay

27
 3.2. Data processing
A further aim of the MLCS 2017 project was to strengthen CSO’s data processing systems and
facilities. Computer Assisted Personal Interviewing (CAPI) could have been an option for MLCS
2017. However an evaluation by the MLCS team concluded that time constraints made it too risky
to move ahead on that basis.

With the objective of using best practices and getting quality data, MLCS 2017 data was entered
twice:

1. In the field data entry was undertaken to check for enumeration errors (logic, completeness,
etc.) ensuring timely quality assurance and correction of data.
2. In the office data entry was done to correct data entry errors.

MLCS implemented a system (using a CSPro 6.3 application) of blind verification where differences
between the first and second data entry are captured in a report and these reports are given to a
third person to resolve.

First data entry in the field

A laptop with an external number pad was configured
for the in-field data entry. The laptop was configured to
access the internet using 3G technology (via smartphones).
Questionnaires were entered while the team was still in the
EA. The team worked on error reports to fix problems on the
spot. Once the data had passed all error checks it was sent to
CSO HQ.

Second data entry at CSO

A data checker looked at the error reports

generated by differences between the first and
second data entry error and resolved issues
by referring to the questionnaire. The correct
values were entered into the system by the
Once the paper questionnaires had been
second data entry operator.
delivered to CSO HQ the data was entered
again.

28
Table 3.2

Summary of Data Processing staff based in Nay Pyi Taw

Role in the survey Main task When trained

10 Data entry operators Enter data into a computer at CSO February 2017 for two weeks in Nay Pyi
Taw
3 Data checkers Check error reports generated by double
data entry process and resolve errors.

3.3. The survey teams

A total of 153 people worked to collect and process MLCS 2017 data (125 were hired and 28
were CSO staff). More detail by State/Region can be seen in Annex D. Twenty interviewing teams
were created to collect the data. Each State/Region had one traveling team (one supervisor, three
enumerators and one data entry operator). Five States/Regions had two teams - Ayeyarwady,
Chin, Rakhine, Sagaing and Shan. The reason to have two teams in some States/Regions was a
combination of:

• Analysis on MPLCS (see Annex A) revealing large Design Effects showing high clustering
of poverty in some State/Regions.
• Difficult transport links between EAs requiring more time to travel within a State/Region.

The gender distribution among the different roles can be seen in the map on the next page.

29
Map 3.1

Gender distribution of the supervisors, enumerators, in-field data entry operators and listers by team

KACHIN

SAGAING 1

SAGAING 2

SHAN NORTH

CHIN NORTH
MANDALAY SHAN EAST & SOUTH

CHIN SOUTH

MAGWAY

NAY PYI TAW

RAKHINE 1

KAYAH
BAGO
RAKHINE 2

KAYIN
AYEYARWADY 1

AYEYARWADY 2
MON
YANGON

TANINTHARYI
KEY: Red male Yellow female
Supervisor

Enumerator

Data Entry Operator

Lister

30
 3.4. Training
The Supervisors and Enumerators played a critical role in the success of MLCS 2017. They were
required to perform multiple tasks with a high level of accuracy, including:

• physically locate the sampled household

• contact the household and explain the purpose of the MLCS
• enumerate household members
• motivate the respondent to participate
• ask questions in the required manner
• put the respondent at ease
• accurately record respondent’s answers

Enumerators can influence responses through their personal attributes and their behaviours,
otherwise known as an “interviewer effect” and one important aim of the training was to unify
approaches to the questionnaire to ensure the interviewer effect was minimized. The various
training events have been identified earlier in this report. Training took place over a period of six
months. CSO Survey Department ran and organized the training events and was supported by
external trainers6.

Written tests and practical work during the listing training

All training involved a lot of practical work in the field. Selection of staff was based on the results
of written tests.

An additional 38 reserve people were trained (twenty enumerators, ten data entry operators
and eight listers). Over the 12 month data collection some of these reserve staff were used (see
Section 3.7).

6 Daw Htar Htar Ei, Ms Valerie Evans, Ms Rachel Smith-Govoni, Dr Reena Badiani Magnusson, Daw Mar Mar Thwin supported
by the World Bank. Mr Felix Schmieding, U Sa Si Thu Htike San, Mr Harold Coulombe, Mr Juste Nitiema, Dr Mya Mya Thet and Ms
Dilrukshi Fonseka supported by UNDP.

31
 At the end of the MLCS main training (2nd December 2016) an assessment was conducted to examine
self-assessed level of experience before and after the MLCS training in the areas of:

• Survey interviewing/data collection

• Survey methodology and design
• Survey administration (for Supervisors only).

On the questionnaire the respondents were asked to rate their level of experience from a scale (1 no
experience to 5 very experienced). The results in Figure 3.2 show that in all aspects the supervisors,
enumerators and data entry operators believed that their skill level had improved.

Figure 3.2

Level of experience before and after training by aspect of the survey process and role

5.0
Supervisor before
4.4

3.8 Supervisor after

4.0 3.8

3.3 3.2 Enumerator before

3.0
3.0 2.7
2.6 Enumerator after
2.5
2.3
2.0 2.0
2.0 Data entry operator
1.6
1.4 before
Data entry operator
1.0 after

0.0
Survey interviewing Survey methodology Survey
and design administration

It is pleasing to see that after a total of two months training the Supervisors rated their skills in
relation to interviewing very highly (4.4 out of 5). Enumerators rated their skill level at 3.3 after their
month of training. Considering this was, for many of them, their first time at such a complex, long
survey this was a good achievement. After 12 months in the field on MLCS it is expected that this
cadre of enumerators are now highly experienced and will be a valuable source for future surveys
in Myanmar.

32
3.5. Replacement of EAs that could not be enumerated
In order to be considered nationally representative, the MLCS aimed to reach all hard to reach
areas and population that were included in the sample – and to do so with consistent listing and
enumeration. There were two issues that were taken into account when assessing access to areas:

1. Security context: was it safe for listers, enumerators and respondents to conduct an
interview?
2. Could accurate data be collected?

To enable this assessment, at the beginning of each quarter, Enumeration Areas were placed into
one of four categories:

1. EA with no foreseen issues;

2. EA where teams would run into serious security issues that are unlikely to be resolved in
the duration of the project;
3. EA where teams may proceed with dialogue and further information gathering;
4. EA on permanent hold until a strategy is devised that allows for quality data to be collected.

This categorization was designated by two teams:

1. CSO interviewing staff and Regional Coordinators who gathered information on the ground.
2. A three person external team who used a variety of other sources to evaluate the latest
situation.

Field teams were informed about these assessments, and individual strategies are taken to deal with
them. Field teams were informed that their safety was the priority and asked to hold-still and check
back with local authorities and the headquarters team in case they had any reservations about
entering an area.

Annex C shows the EAs that could not be enumerated because of security or access problems in
detail. By the end of data collection 33 EAs had been replaced and 7 were not visited at all. The
distribution by State/Region and quarter can be seen on Map 3.2. In order to minimize bias in the
sample, the replacement of sample EAs was only considered in extreme cases after attempts had
been made to reach and enumerate the sample EA.

In order to maintain the geographical representativeness of the sample, the replacement EA for a
sample EA that could not be enumerated was selected from other Master Sample EAs in the same
stratum (state/region, urban/rural) and within the same township when possible. In these cases
all of the Master Sample EAs from the same stratum and township were identified, and one of
these EAs selected at random with equal probability (since the Master Sample EAs had already been
selected with PPS).

If an entire township could not be accessed because of security problems, a neighboring township
was identified, and a random EA selected from the Master Sample EAs for that township. Since the
Master Sample EAs were selected using PPS, the measure of size (number of households in the
frame) for the replacement EA was used for calculating the weights, as specified in Section 5.

33
Map 3.2

Replaced and not visited EAs by State/Region and Quarter

Kachin

Shan
Rakhine:
Maungdaw,
Buthidaung & Naypyitaw
Taungpyoletwea
sub-township

Central Rakhine Kayah

Bago
Rakhine: 19 replaced EAs in total.
15 from original sample, 4 from
supplemental sample – see below
for more detail
Kayin

KEY
Quarter 1

Quarter 2 Tanintharyi

Quarter 3

Quarter 4

33 EAs Replaced

7 EAs not visited at all

(Wa self-administered area)

34
In terms of the distribution of the households in the areas that were not covered by the survey,
Map 3.2 clearly shows that the worst affected areas were three townships in the Northern part of
Rakhine: Maungdaw, Buthidaung and Taungpyoletwea (Sub-Tsp).

In the original sample selection no areas of Myanmar were excluded from the selection process and
as a result these three townships were included in the sample. Despite a processes of review and
plans to conduct interviews in these townships throughout the 12 month period it was not possible
to enumerate any sample EAs in these three townships, so they are not represented in the MLCS
2017 data.

The final enumerated sample for Rakhine consists of 72 EAs, the same as the original sample
allocation, because of a supplemental sample that was selected to compensate for the missing
Northern Rakhine sample EAs. Although the supplemental sample maintained the effective sample
size, it does not reduce the bias. The total number of households in the frame for these three
townships is 92,615, and the total number of households in the frame for all of Rakhine State is
553,991, so the percentage of households in the three townships is 16.7% of the state total. This is a
significant part of the population of Rakhine.

In Quarter 4 it was not possible to interview in four largely Muslim EAs in Sittwe and Mrauk-U in
Central Rakhine. In quarters 1, 2 and 3 interviews had successfully been undertaken with Muslims
in these areas. However tensions increased greatly following attacks on police posts on August 25th
2017 and the subsequent exodus of the Muslim population from northern Rakhine. The four central
Rakhine EAs were replaced. The final weighted estimate of the total number of households in these
townships from the MLCS 2017 data was slightly higher that the corresponding frame, so the sample
in these areas should be fairly representative.

35
Map 3.3

Areas of Myanmar not enumerated in MLCS 2017

No interviews in the Wa
Self-Administered Area. The
7 EAs were not replaced as
Shan had already had a larger
proportion of EAs due to its
large size (the largest State/
Region in area in Myanmar).

No interviews in Buthidaung,
Maungdaw townships at
Northern Rakhine. All EAs
replaced with EAs in other
parts of Rakhine

36
 3.6. Use of interpreters during interviews
There were 99 (out of 1,145) EAs in which the survey teams could not administer the questionnaire
in Myanmar language. In such areas, CSO hired local interpreters, usually recruited at a town
closest to the enumeration areas. Efforts were made to ensure that the interpreters had sufficient
understanding of the questionnaire and spoke both the local language and dialect and Myanmar
language comfortably. The local interpreters were given training by field supervisors on interpreting
the questionnaire and were provided explanations of the protocols and procedures for enumerating
households.

Chin had the highest rate of interpreters being used with 68% of the EAs requiring interpretation
(556 households). It was not always the case that all 12 of the selected households in the EA required
interpretation.

Figure 3.3

Percentage of EAs within each State/Region where interpreters were used

68

29
19
15
4 4 4 3

Chin Shan Kachin Kayah Kayin Tanintharyi Mon Sagaing

Most interviews were conducted in the Myanmar language (87.1%).

37
Table 3.3

Language in which interview was conducted

Language Percentage

Myanmar 87.1

Kachin 0.2

Kayin 1.0

Chin 0.6

Mon 0.4

Rakhine 3.8

Shan 2.9

Dawei 2.4

Other7 1.5

Total 100.0

3.7. Replacing survey staff

Due to personal reasons and a motorbike accident only one in field data entry-operators and one
lister could not complete the assigned EAs during the fieldwork duration. The data entry person was
replaced with a reserve person that had been trained in Nay Pyi Taw in November. The replacement
lister was also a reserve person that had been trained at the listers training in October.

7 Mainly Shan hill tribal languages and Chinese.

38
39
40
04.QUALITY
ASSURANCE

41
 4.1 Field Trips
Quality Assurance on MLCS 2017 took place through two main mechanisms:

• Field trips
• Reports generated from the data entry program

By the end of fieldwork all 20 teams had been visited by CSO, UNDP or World Bank staff. The actual
visits can be seen in Table 4.1.

During these visits the Quality Assurer used the visit to discuss if there were any specific problems
or questions. They asked the data entry operator to show them the status report on the laptop
and also checked that data entry had been completed before the team left the EA. In addition, the
following specific items were checked:

Listing

• Could the 12 selected households ticked in Column 8 of the Listing Form be found on the map,
including any new buildings.
• Was Column 7 (serial number) numbered in sequence?
• Did the final number written in Column 7 match the number written in “Total Number of
Households in the EA” on the Front Page of the Listing Form?
• Was the ticking in Column 8 for the 12 households correct according to the Household Selection
Table?
• Asked the village head/elder/representative that met the listing team whether the listers had
walked around every street of the EA to update the map.
• Asked a village head/elder/representative whether there were cases of multiple households
residing in one building to check that they were listed as separate households on the Listing
Form.

Interviewing

• Went to one or two selected households to ensure that the household interviewed was actually
the household selected on the Listing Form, i.e. to ensure they didn’t go to an easier/nearer
household instead.
• Checked 1-2 household questionnaires from each enumerator, watching out for inconsistencies
and paying attention to supervisor corrections.

42
 • Checked supervisor conversions in Section 5A.
• Checked in the Community Questionnaire whether the market price section had the correct
fixed/non-fixed codes and whether the non-fixed units corresponded with those used in the
household questionnaire.

Table 4.1

Quality Assurance Visits

State/Region Supervisor Date of visit Quality Assurers EAs visited

1 Kachin U Hla Myo Than 24.2.2017 Daw Win Pa Pa Zaw 010101115001
Daw Khin Sett Yi 010101124016
010102104002
010155701005
10-14. 11. 2017 DG Dr Wah Wah Maung 010101102003
Daw Khin Sett Yi 010101116007
2 Kayah U Thet Naing Win 2-3.3.2017 U Sa Si Thu Htike San 020101101004
020101101012
3.5.2017 DG Dr Wah Wah Maung 020101111003
Daw Win Pa Pa Zaw
25-11-2017 Daw War War Myint 020202706001
Daw Thet Htar Nwe 020101101014
3 Kayin U Khun Tue Naung 6-9.2.2017 DG Dr Wah Wah Maung 030301103001
Lay Daw Phyu Pyar Tun 030301105023
Ms Rachel Smith-Govoni 030101773003
Mr Rivandra Royono 030101106003
120308119008
4 Chin (North) U Lwin Naing 25-29. 11. 2017 Daw Khin Sett Yi 040101103005
Ms Rachel Smith-Govoni 040201101001
040101105003
Chin (South) U Than Myint 26-27.1.2017 Daw Hsu Hnin Wai 040304732001
Ms Rachel Smith-Govoni 040304101004
5 Sagaing Daw Khaing Thandar 21.2.2017 Mr Felix Schmieding 120102708020
Thein
U Tin Tun Aung 26.7.17 Daw Ohn Mar Myint 050301125008
Hlaing Ms Rachel Smith-Govoni 050301102002
6 Tanintharyi U Win Zaw Oo 17.1.2017 Mr Felix Schmieding 060201703012
U Sa Si Thu Htike San
Ms Hyeran Kim
11-18.11. 2017 Daw Ohn Mar Myint 060202715010
Daw Phyu Pyar Tun 060202703009
7 Bago Daw Mya Mya Thin 15-17.2.2017 Daw War War Myint 070101123008
Daw Thet Htar Nwe 070105302001
Mr Rivandra Royono 070106735002
070104743008
26.11.2017 Dr Reena Badiani Magnusson 070305726009
Mr Clarence Tsimpo Nkengne
8 Magway Daw Khin Phway 23.2.2017 Visit by UNDP Country Director 080501102004
Phway Zaw
22.5.2017 Daw Khin Sett Yi 08013108001
Ms Diane Steele 08010375003
Ms Rachel Smith-Govoni

43
 State/Region Supervisor Date of visit Quality Assurers EAs visited
9 Mandalay Daw Ei Ei Min 24.5.2017 Daw Khin Sett Yi 09102114002
26.5.2017 Ms Diane Steele 09201105010
Ms Rachel Smith-Govoni 09701725001
10 Mon U Oo Thant 14-17.2.2017 DG Dr Wah Wah Maung 100201719004
Daw Thet Htar Nwe 100105103002
Daw Ohmar Soe 100202103005
100204702009
23.10.2017 Dr Reena Badiani Magnusson 100104201002

11 Rakhine Daw Nilar Thein 17.1.2017 Daw Khin Sett Yi 110101130002

Ms Rachel Smith-Govoni
18-19.1.2017 110201764004

Daw Kyi Kyi Than 26.5.2017 Mr Felix Schmieding 110202715004

U Sa Si Thu Htike San
12 Yangon Daw Yadaner Win 16.12.2016 Daw Khin Sett Yi 120302718002
Aung Daw Hsu Hnin Wai
Mr Juste Nitiema
5.1.2017 Mr Felix Schmieding 120102102011
U Sa Si Thu Htike San
23.1.2017 Daw Htar Htar Ei 120301112002

4.2.2017 Ms Rachel Smith-Govoni 120205104007

Daw Htar Htar Ei 120406119014
2.5.2017 DG Dr Wah Wah Maung 120103102005
Daw Win Pa Pa Zaw
14.2.2017 Ms Rachel Smith-Govoni 120308119008
Daw Htar Htar Ei
21.2.2017 Mr Felix Schmieding 120108708020
U Sa Si Thu Htike San
4.4.2017 Daw Htar Htar Ei 120204114040

1.5.2017 Daw Htar Htar Ei 120214106013

Ms Rachel Smith-Govoni 120203104012
6.5.2017 Daw Htar Htar Ei 120410105005

13 Shan (North) U Win Zaw Htay 6.5.2017 DG Dr Wah Wah Maung 130103710007
Daw Win Pa Pa Zaw
3.2 2017 Daw Win Pa Pa Zaw 130401107013
Daw Khin Sett Yi 130401112022
Shan (South) U Thet Aung 17.12.2016 Ms Rachel Smith-Govoni 130102720007

14 Ayeyarwady Daw Maw Maw Khin 18-22.9.17 Daw Khin Sett Yi 140402109001
Daw Than Than Soe 140302107001
140303739005
U Soe Htet Paing 13-14.2.2017 Mr Felix Schmieding 140201772003
U Sa Si Thu Htike San
15 Nay Pyi Taw Daw Maw Maw Nyein 13.1.2017 Daw Thet Htar Nwe 150102708021
Ms Rachel Smith-Govoni
5.5.2017 Daw Khin Sett Yi 150202703008
Ms Rachel Smith-Govoni

44
4.2. QA reports generated from the in-field data entry
program
Within the CSPro program a number of tables
were produced at the enumerator level and
on a weekly basis (see Annex 5). Tables were
provided to the MLCS central team on a weekly
basis to examine if issues were arising. Before
each QA visit the Quality Assurers reviewed
the most recent tables for the team they were
visiting.

The main tables produced were:

• Number of interviews completed each week: Enabling the team to see if any team or
enumerator were particularly slow and to investigate the cause for this.
• Average Household Size: Compared to other data sources, such as the Census 2014, LFS etc.
and examined if particular enumerator were recording fewer people in the household and
whether this could be explained.
• Average number of items consumed in the last 7 days: Taken from Section 5A this counts the
number of different types of food consumed over one week. The number of items may change
by season as some products become available and others are no longer available. The number
of items may also differ by region or in remote areas where fewer items are available. As Section
5A is time consuming (an average of 38 minutes) examining this data helped identify whether
enumerator fatigue had become an issue.
• Average number of items purchased in the last 7 days: Taken from Sections 5C & 5D this
counts the number of different types of products purchased and again can be examined to see
whether fatigue over the 12 months of the survey is an issue.
• Average number of durable goods owned: Taken from Section 6 this can be compared to other
data sources, MPLCS, DHS etc.
• Average time taken to complete interviews: The average time taken to complete an interview
was two hours and twenty minutes. This data could be examined to see if any enumerators
were completing interviews particularly quickly.

The reports were examined every week and occasionally action had to be taken as a consequence
of seeing the results. For example in February 2017 the QA report showed one Ayeyawaddy team
was late delivering questionnaires. A QA trip was undertaken to examine the delay and identified
a misunderstanding about the purpose of in-field data entry. This was rectified by the team and
the problem didn’t arise again. In April the QA report identified an enumerator in Naypyitaw whose
interviews averaged one hour 23 minutes in length (almost one hour faster than the overall average
time). A specific QA visit was carried out to examine his interviewing style and give the enumerator
advice how to improve his work. In July the QA report showed a noticeable reduction in the number
of food items consumed for one of the Shan teams. The Supervisor was contacted and made aware
of the issue and told to follow the work of his enumerators. The work improved again after this
intervention.

45
46
05.
WEIGHTING AND
SAMPLING
ERROR
CALCULATION

47
 5.1 Weighting procedures
In order for the MLCS 2017 sample estimates to be representative of the population, it was
necessary to multiply the data by a sampling weight, or expansion factor. The basic weight for each
sample household is equal to the inverse of its probability of selection (calculated by multiplying the
probabilities at each sampling stage).

The sampling probabilities at each stage of selection are maintained in an Excel spreadsheet
with information from the sampling frame for each sample EA so that the corresponding overall
probability and corresponding weight can be calculated.

Based on the Myanmar Master Sample design and the sampling procedures for the MLCS 2017, the
overall probability of selection for the MLCS 2017 sample households can be expressed as follows:

where:

phi = probability of selection for the sample households in the i-th sample EA in stratum h
for the MLCS 2017

nh = number of sample EAs selected in stratum h for the Master Sample, specified in Table
2.3

Mh = total number of households in the 2014 Census frame for stratum h

Mhi = total number of households in the 2014 Census frame for the i-th sample EA in stratum
h

n'h = number of MLCS 2017 sample EAs selected from the Master Sample for stratum h,
specified in Table 2.3

pShi = probability of selection for the selected segment in large sample EA that is subdivided;
this probability is equal to 1 for all EAs that are not segmented

mhi = number of sample households selected in the i-th sample EA in stratum h (12)

M'hi = total number of households listed in the i-th sample EA in stratum h

The different components of this probability of selection correspond to the individual sampling
stages. The probability of selecting a segment in a large EA (pShi) depends on the type of selection
procedure that is used. If the sample segment is selected with PPS, this probability would be
calculated as follows:

48
 where:

M''hij = total number of housing units or households from the quick count for the j-th selected
segment in the i-th sample EA in stratum h

M''hi = total number of housing units or households from the quick count for the i-th sample
EA in stratum h

If one segment is selected randomly with equal probability, this probability would be calculated as
follows:

where:

Shi = total number of segments in the i-th large sample EA in stratum h

The basic sampling weight, or expansion factor for the MLCS 2017 sample households, is calculated
as the inverse of this probability of selection. Based on the previous expression for the probability,
the weight can be simplified as follows:

where:

Whi = basic weight for the MLCS 2017 sample households in the i-th sample EA in stratum h

49
 5.2. Weighting to take account of non-response
For the MLCS 2017 the sample households which could not be interviewed were replaced with one
of the four replacement households selected from the Household Selection Table. It is important
to adjust the sample household weights to take into account the non-interview households in each
sample EA. Since the weights are calculated at the level of the sample EA, it would be advantageous
to adjust the weights at this level. The final weight (W'hi ) for the sample households in the i-th
sample EA in stratum h can be expressed as follows:

where:

m'hi = total number of valid (occupied) sample households selected in the i-th sample EA in
stratum h for the MLCS 2017

m''hi = total number of sample households with completed MLCS 2017 interviews in the i-th
sample EA in stratum h, including replacements

If each non-interview household is replaced within a sample EA, the weight adjustment factor would
be equal to 1.

By the end of the data collection 13,730 households were successfully interviewed from the sample size
of 13,824 original households. The completed interviews included 348 replacement households. The
distribution of the final sample households interviewed by replacement status is presented in Table 5.1

Table 5.1

Completed Household Interviews for MLCS 2017 by State/Region

Number of actual Number of actual interviews

Total expected interviews
interviews from Original EAs from Replacement EAs

Kachin 864 840 24

Kayah 864 840 24

Kayin 864 852 12

Chin 864 864 -

Sagaing 864 864 -

Tanintharyi 864 852 12

Bago 864 840 24

Magway 864 864 -

Mandalay 864 863 -

Mon 864 864 -

Rakhine 864 684 180

Yangon 1,152 1143 -

Shan 1,152 1020 48

Ayeyarwady 1,152 1152 -

Nay Pyi Taw 864 840 24

Total 13,824 13,382 348

50
5.3. Final Response Rate
The final household response divides the number of actual interviews achieved (13,730) by the total
number of attempts to interview (14,668) producing a final household response rate of 93.6%.

5.4. Survey Estimate from MLCS 2017 data

The most common estimates calculated from the MLCS 2017 data are in the form of totals and
ratios. The survey estimate of a total can be expressed as follows:

where:

L= number of strata (state/region, urban/rural) for the domain

yhij = = value of variable y for the j-th sample household in the i-th sample EA in stratum h

The survey estimate of a ratio is defined as follows:

where Ŷ and X are estimates of totals for variables y and x, respectively, calculated as specified
previously.

In the case of a stratified multi-stage sample design, means and proportions are special types of
ratios. In the case of the mean, the variable x, in the denominator of the ratio, is defined to equal 1
for each element so that the denominator is the sum of the weights. For a proportion, the variable
x in the denominator is also defined to equal 1 for all elements; the variable y in the numerator is
binomial and is defined to equal either 0 or 1, depending on the absence or presence, respectively,
of a specified characteristic for the element.

51
 5.5. Calculation of Sampling Errors
Tables with calculated sampling errors and confidence intervals for all survey estimates in the MLCS
2017 Report can be found on the CSO website.

The standard error, or square root of the variance, is used to measure the sampling error, although it
may also include a small variable part of the non-sampling error. The variance estimator should take
into account the different aspects of the sample design, such as the stratification and clustering.
Stata (using a linearized Taylor series variance estimator) has been used to calculate the variances
for survey data from stratified the multi-stage sample designs of MLCS 2017.

For each estimate the output tables show the standard error, coefficient of variation (CV), 95
percent confidence interval, the design effect (DEFF) and the number of observations. The design
effect is defined as the ratio of the variance of an estimate from a complex (stratified, multi-stage)
sample to the variance of a simple random sample of the same size. It is a relative measure of the
sampling efficiency. Most of the design effects are greater than 1 given the clustering effects in the
sample design.

The variance estimator of a total can be expressed as follows:

where:

The variance estimator of a ratio can be expressed as follows:

where:

V(Ŷ) and V(X) are calculated according to the formula for the variance of a total.

52
53
54
ANNEXES

55
 Annex A: Sample size determination for MLCS 2017
It was useful to tabulate the sampling errors, confidence intervals and design effects for key
estimates from MPLCS in order to examine the sample size requirements for the MLCS 2017 and
study the efficiency of the sample design.

The design effect is defined as the ratio of the variance of an estimate from the actual sample design
and the corresponding variance from a simple random sample of the same size; it is a measure of the
relative efficiency of the sample design, which mostly depends on the clustering effect.

The previous national household surveys in Myanmar that included information related to household
expenditure and the measurements of poverty were the 2005 and 2010 Integrated Household
Living Conditions Assessment Survey (IHLCA), and the 2015 Myanmar Poverty and Living Conditions
Survey (MPLCS). One limitation of the IHLCA sample design is that no recent Census of Population
and Housing was available for constructing the sampling frame at that time. The primary sampling
units (PSUs) selected at the first stage were townships, resulting in an extremely clustered sample.
Two townships were selected per district, so there were a total of 124 sample PSUs. There were
three additional stages of selection, but much of the sampling errors will come from the between-
PSU variance component, resulting in very high design effects, indicating that the sample design is
very inefficient. The data from the 2010 IHLCA were used for calculating the sampling errors, which
confirmed that the design effects were very high, so these survey data could not be used to estimate
the level of precision that can be expected from the MLCS 2017. On the other hand, the 2015 MPLCS
sample design was based on the new Master Sample selected from the 2014 Myanmar Census of
Population and Housing frame, which is also being used for selecting the sample EAs for the MLCS
2017. Therefore the 2015 MPLCS data were used for tabulating sampling errors and design effects
for key indicators in order to simulate the level of precision that can be expected from alternative
sample designs for the MLCS 2017.

In order to calculate the sampling errors for estimates from the 2015 MPLCS data, it is first necessary
to understand the sample design for this survey. Therefore a brief description of the sample design
is presented here. A more detailed description of the sample design and estimation procedures for
the 2015 MPLCS is included in the report on “Sample Design and Estimation Procedures for the
2015 Myanmar Poverty and Living Conditions Survey” (Megill, September 2015).

The geographic domains of analysis for the MPLCS were five agro-ecological zones that were
combinations of states/regions, as defined in Table A1; the state of Yangon is considered an individual
domain. The sampling strata were defined as the urban and rural areas of each state/ region, the
same as the stratification of the Master Sample. A total of 60 sample EAs were allocated to each zone
except for Yangon, where 64 sample EAs were selected. With a sample of 12 households selected
per sample EA, this resulted in a sample size of 768 households for Yangon and 720 households for
each of the remaining zones, for an overall sample size of 3,648 households.

56
Table A1

Distribution of Sample EAs and Sample Households for the MPLCS by Region, Urban and Rural Stratum

Urban Rural Total

Zone State
Sample Sample Sample Sample Sample Sample

Kachin 4 48 5 60 9 108
Hills Kayah 2 24 2 24 4 48
and
Kayin 3 36 7 84 10 120
Mountains
Chin 2 24 2 24 4 48

Shan 8 96 25 300 33 396

Sagaing 4 48 14 168 18 216

Magway 3 36 12 144 15 180

Dry zone
Mandalay 8 96 14 168 22 264

Nay Pyi Taw 2 24 3 36 5 60

Bago 5 60 17 204 22 264

Delta Mon 3 36 6 72 9 108

Ayeyarwady 5 60 24 288 29 348

Tanintharyi 6 72 14 168 20 240

Coastal
Rakhine 7 84 33 396 40 480

Yangon Yangon 45 540 19 228 64 768

Total 107 1,284 197 2,364 304 3,648

In allocating the sample EAs within each zone to the urban strata, the proportion of urban households
in each zone was first multiplied by a factor of 1.3 for all zones except Yangon in order to increase
the overall number of sample urban households compared to a strictly proportional allocation.
Since the Yangon zone is predominantly urban, the sample allocation by stratum for this zone was
based on a simple proportional allocation. At the national level, this resulted in a total sample of 107
urban EAs and 197 rural EAs, with a corresponding overall sample size of 1,284 urban households
and 2,364 rural households. Within the urban and rural stratum of each zone the sample EAs were
then allocated proportionately to the corresponding strata in each state/region. The final allocation
of the sample EAs and households for the 2015 MPLCS by state/region, urban and rural strata is
presented in Table 2.4.

For calculating the sampling errors it is important that the variance estimator take into account
the effects of the stratification and clustering in the sample design. The Complex Samples module
of SPSS was used for tabulating the standard errors, 95% confidence intervals and design effects
for selected indicators based on the stratified multi-stage sample design. It uses a linearized Taylor
series variance estimator, which is described later in the section on "Calculation of Sampling Errors”.

The measures of precision were tabulated for the estimates of the average per capita expenditures
and the poverty rate at the national level, for the urban and rural strata and by state/region, from
the 2015 MPLCS data. The results for the estimates of average per capita expenditure by domain are
presented for illustrative purposes. It should be pointed out that since the sample within each zone
was allocated to the state/regions proportionately to their size, the smallest states/regions within
each zone will have a relatively small sample size. The estimates at the regional level cannot be

57
 considered reliable since the sample was designed for zone-level domains. However, the estimates
of the sampling errors and design effects at the regional level were still useful for conducting a
simulation study of the expected level of precision for key estimates from the MLCS 2017 data
based on different sampling alternatives, as described in the next section.

Sample Size and Allocation for MLCS 2017

The sample size for household surveys such as the MLCS 2017 is determined by the accuracy
required for the survey estimates for each domain, as well as by the logistical, timing and resource
constraints. The accuracy of the survey results depends on both the sampling error, which can
be measured through variance estimation, and the non-sampling error, which results from all
other sources of error, including response and measurement errors as well as coding, keying and
processing errors. The sampling error is inversely proportional to the square root of the sample
size. On the other hand, the non-sampling error may increase with the sample size, since it is more
difficult to control the quality of a larger survey operation. It is therefore important that the overall
sample size be manageable for quality and operational control purposes. This is especially important
given the challenge of collecting accurate information on household income and expenditures, as
well as crop production.

The geographic domains of analysis for the MLCS 2017 are the 15 states or regions of Myanmar,
identified in Table 2.4. The urban and rural areas will also be domains at the national level. It is
necessary to ensure that each of these domains have a sufficient number of sample households to
obtain reliable survey estimates at the domain level. One important consideration in the allocation
of the sample by state/region is that the sample size for a particular domain does not depend on the
total population of the domain. The only exception would be if there were extremely small domains
where the overall sampling rate was greater than 5%, which is not the case for the states/regions.
Since a similar level of precision is required for each state/region, a similar number of sample EAs
and households should be selected in each. It would only be necessary to increase the sample size
slightly for any state/region which has a greater variability in socioeconomic characteristics, or a
larger design effect from the clustering of the sample, as explained later.

Each state/region is further divided into urban and rural sampling strata, corresponding to the
stratification of the master sample by state/region, urban and rural areas. This stratification will
improve the statistical efficiency of the sample design, but the MLCS 2017 results will not be
obtained at the sampling stratum level.

Another important consideration in the sample design is how many households to select in each
sample cluster (EA). For this type of socioeconomic survey it has been found in various countries
that the optimum number of sample households to be selected per cluster is generally close to
12, so it was decided to select 12 households per EA for the MLCS 2017. This provides an effective
balance between the number of sample EAs and the number of sample households per EA, and
provides a reasonable dispersion of the sample. If the number of sample households per EA were to
be increased, this would result in higher design effects and sampling errors.

In order to spread the sample evenly to the 12 months within each state/region, the total number
of sample EAs allocated to each state/region should be a multiple of 12. A nationally-representative
subsample of EAs will be assigned to each quarter, and these EAs will be evenly allocated to the
months within the quarter, in order to ensure that seasonality is effectively represented throughout
the year. Given that a similar level of precision will be required for the survey estimates from each
state/region, a similar sample size is needed for each of these domains. However, the sample

58
size may be increased some for states/regions that have a greater design effect or variability in
socioeconomic characteristics. In reviewing the sampling errors and design effects for the estimates
of average per capita expenditure from the 2015 MPLCS data, it was found that Yangon, Shan and
Ayeyarwady have the higher design effects and variability, so it was decided to allocate a slightly
larger sample for these domains. Two different sampling alternatives were compared to study the
corresponding level of precision as well as considerations for the relative costs and data quality
related to non-sampling error.

Under the first sampling alternative, 6 sample EAs would be enumerated in most states/regions
each month, except for Yangon, Shan and Ayeyarwady, where 8 sample EAs would be covered each
month. With the selection of 12 sample households per EA, this sampling alternative would result
in a total sample of 72 sample EAs and 864 sample households for most states/regions; and 96
sample EAs and 1,152 sample households for Yangon, Shan and Ayeyarwady. At the national level the
total sample size for the 12-month period would be 1,152 sample EAs and 13,824 sample households.
The urban and rural strata are only considered domains of analysis at the national level. However,
the urban and rural strata are defined at the state/region level in order to improve the efficiency of
the sample design. One consideration for the allocation of the sample to the urban and rural strata
is that there is generally more variability in socioeconomic characteristics within the urban areas,
and the cost of the fieldwork is also generally lower in urban areas. For this reason a weight of 2 was
applied to the number of urban households in the proportional allocation of the sample to the urban
and rural strata of each state/region, except for Yangon, where a strictly proportional allocation
was used given that this state is predominantly urban. This sample allocation strategy is designed to
increase the proportion of urban households in the MLCS 2017 sample. In this way the proportion of
urban households in the sample was increased to 38.5%, compared to 27.5% urban households in the
2014 Census sampling frame. Since the weights applied to the survey data for sample households
will be based on the corresponding probabilities of selection, the weighted results for the urban
and rural domains will be consistent with the distribution of the frame. This first sampling strategy
results in an urban sample of 443 EAs and 5,316 households, a rural sample of 709 EAs and 8,508
households. The allocation of sample EAs and households by state/region and urban/rural stratum
under the first sampling alternative is shown in Table 7.4.

The SPSS Complex Samples module was used for calculating the sampling errors from the 2015
MPLCS data for the estimates of average per capita expenditure and the poverty rate at the national
level, urban and rural domains, and by state/region. The sampling error tables for the estimates of
average per capita expenditure by domain are shown below.

59
Table A2

Estimates, sampling errors, coefficients of variation (CVs), 95% confidence intervals, design effects and number of
observations for average per capita expenditure by domain from 2015 MPLCS data

95% confidence interval No.

Domain Estimate SE CV DEFF observations
Lower Upper (households)

Myanmar 871,088 15,241.7 0.017 841,082 901,093 2.9 3,648

Residence

Urban 1,127,467 36,556.3 0.032 1,055,500 1,199,434 2.8 1,332

Rural 767,877 13,829.4 0.018 740,651 795,102 3.1 2,316

State/Region

Kachin 822,715 82,633.6 0.100 660,038 985,393 3.7 108

Kayah 862,419 131,179.6 0.152 604,171 1,120,667 2.5 48

Kayin 748,983 65,127.7 0.087 620,768 877,197 2.1 120

Chin 423,978 6,738.1 0.016 410,713 437,243 0.04 48

Sagaing 794,379 37,530.9 0.047 720,493 868,264 3.0 216

Tanintharyi 931,848 62,629.6 0.067 808,552 1,055,144 1.6 240

Bago 756,272 36,440.5 0.048 684,532 828,011 3.5 264

Magway 863,130 31,562.6 0.037 800,994 925,266 2.0 180

Mandalay 901,817 32,837.2 0.036 837,172 966,463 2.1 264

Mon 709,485 37,785.3 0.053 635,099 783,872 2.4 108

Rakhine 737,927 46,822.7 0.063 645,749 830,105 1.3 480

Yangon 1,222,255 61,071.1 0.050 1,102,026 1,342,483 2.9 768

Shan 740,687 35,188.3 0.048 671,413 809,961 3.7 396

Ayeyarwady 859,982 47,293.8 0.055 766,877 953,088 5.4 348

Nay Pyi Taw 1,079,307 35,617.9 0.033 1,009,187 1,149,426 0.5 60

These results were used for the simulation study to estimate the expected level of precision based
on the two sampling alternatives. The estimated measures of precision from this simulation study
are shown in Tables A4 and A5.

In the case of the second sampling alternative, the sample size would be increased to enumerate
8 sample EAs each month in all states/regions except for Yangon, Shan and Ayeyarwady, where 12
sample EAs would be covered each month. This sampling alternative would result in a total sample
of 1,152 sample households for most states/regions, and 1,728 sample households for Yangon,
Shan and Ayeyarwady. The total sample size at the national level for the 12-month period would
be 1,584 sample EAs and 19,008 sample households. The allocation of the sample to the urban and
rural strata within each state/region was based on a similar strategy as that described for the first
sampling alternative. This resulted in an urban sample of 620 EAs and 7,440 households, a rural
sample of 964 EAs and 11,568 households. The allocation of sample EAs and households by state/
region and urban/rural stratum under the second sampling alternative is shown in Table A3.

60
Table A3

Sampling Alternative 2 – Allocation of Sample Clusters and Households for MLCS 2017

Total Urban Rural

State/Region Sample Sample Sample Sample Sample Sample
clusters households clusters households clusters households

Kachin 96 1,152 48 576 48 576

Kayah 96 1,152 40 480 56 672

Kayin 96 1,152 36 432 60 720

Chin 96 1,152 32 384 64 768

Sagaing 96 1,152 28 336 68 816

Tanintharyi 96 1,152 36 432 60 720

Bago 96 1,152 32 384 64 768

Magway 96 1,152 24 288 72 864

Mandalay 96 1,152 48 576 48 576

Mon 96 1,152 40 480 56 672

Rakhine 96 1,152 24 288 72 864

Yangon 144 1,728 100 1,200 44 528

Shan 144 1,728 56 672 88 1,056

Ayeyarwady 144 1,728 36 432 108 1,296

Nay Pyi Taw 96 1,152 40 480 56 672

Total 1,584 19,008 620 7,440 964 11,568

In order to determine the approximate level of precision that can be expected based on these
alternative sample designs, a simulation study was carried out using the sampling error and design
effect results from the 2015 MPLCS data. Given that the sample EAs for both the 2015 MPLCS
and the MLCS 2017 are selected from the Master Sample based on the 2014 Myanmar Census of
Population and Housing, the stratification is the same, and 12 households are selected per sample
EA for both surveys, the design effects for key indicators should be similar.

61
 The ratio of the variance (square of the standard error) for a survey estimate based on the proposed
sample design for the MLCS 2017 to the corresponding variance based on the 2015 MPLCS data can
be expressed as follows:

where:

= approximate variance (square of standard error) for a survey estimate

based on a particular sampling alternative for the MLCS 2017

= variance for estimate from the 2015 MPLCS data based on the actual
sample design

= population variance (square of standard deviation) of variable; in the case

of the estimate of a proportion (p),

= proposed number of sample households (for domain of estimation) in the

MLCS 2017 based on a particular sampling alternative

= actual number of sample households (for domain of estimation) for 2015

MPLCS

Assuming that the standard deviations of the variables and the design effects of the estimates
are similar for the two surveys, this ratio of the variances simplifies as the number of sample
households in the 2015 MPLCS to the corresponding number of sample households in the MLCS
2017 for the same domain. From this ratio of the sample sizes for the two surveys, we can calculate
the approximate standard error for an estimate that would result from the two different sampling
alternatives for the MLCS 2017 as follows:

This formula was used with the 2015 MPLCS estimates of standard errors in order to determine the
approximate level of precision for the estimates by domain based on the two sampling alternatives
described above. This study was based on the tabulated sampling errors for average per capita
expenditure and the poverty rate from the 2015 MPLCS data.

A practical measure of precision for comparing the expected results from the two sampling
alternatives is the margin of error, which is equal to half of the width of the confidence interval (that
is, the +/- value). In the case of a 95% confidence level, the margin of error is calculated as follows:

62
 In the case of an indicator which is a proportion, it is useful to compare the margins of error from the
different sampling alternatives. However, for indicators which are in the form of averages or totals,
it is more relevant to compare the relative margin of error, which is equal to the margin of error
divided by the value of the estimate, expressed as a proportion or percentage.

The results from this simulation study for the estimates of the average per capita expenditure by
state/region from the two sampling alternatives for the MLCS 2017 are presented in Table A4.

Table A4

Simulation of Expected Level of Precision for MLCS 2017 Estimates of Average per Capita Expenditure by State/
Region for Two Sampling Alternatives, Based on 2015 MPLCS Data

Sampling alternative 1 Sampling alternative 2

State/Region DEFF (MPLCS)
Sample households Relative ME Sample households Relative ME

Kachin 3.7 864 7.0 1,152 6.0

Kayah 2.5 864 7.0 1,152 6.1

Kayin 2.1 864 6.4 1,152 5.5

Chin 1.0* 864 3.5 1,152 3.0

Sagaing 3.0 864 4.6 1,152 4.0

Tanintharyi 1.6 864 6.9 1,152 6.0

Bago 3.5 864 5.2 1,152 4.5

Magway 2.0 864 3.3 1,152 2.8

Mandalay 2.1 864 3.9 1,152 3.4

Mon 2.4 864 3.7 1,152 3.2

Rakhine 1.3 864 9.3 1,152 8.0

Yangon 2.9 1,152 8.0 1,728 6.5

Shan 3.7 1,152 5.5 1,728 4.5

Ayeyarwady 5.4 1,152 5.9 1,728 4.8

Nay Pyi Taw 1.0* 864 2.4 1,152 2.1

*The estimates of the DEFF that were calculated with a value of less than 1 were rounded up to 1 in order to provide more realistic results.

Under both sampling alternatives the relative margins of error are under 10%, indicating an acceptable
level of precision. As expected, since the sample size is larger under the second sampling alternative,
the margins of error are slightly lower than those from the first sampling alternative. However, the
gain in precision from the corresponding increase in sample size is relatively small. The design effect
for Ayeyarwady is relatively high (5.4), indicating a larger clustering effect for average per capita
expenditure in this state, and a larger variability between the clusters. The design effects calculated
from the 2015 MPLCS data for two states were less than 1, which is unusual and cannot be expected
to happen again, so these design effects were rounded up to 1 for the simulation study. The sample
size for the 2015 MPLCS for some states was relatively small, so the corresponding estimates of
the design effects are less robust and more variable. It is expected that the design effects from the
MLCS 2017 data will have less extreme values compared to the 2015 MPLCS.

63
 The results of the simulation study on the expected levels of precision for estimates of the poverty
rate by state/region based on the two sampling alternatives are presented in Table 8. The margins
of error for the poverty rate by state/region are less than 8% under both sampling options, so the
level of precision is also acceptable for this indicator. The design effect for Ayeyarwady (9.8) is
even higher than for the estimate of average per capita expenditure, indicating that there is also a
relatively high level of clustering for poverty in this state.

Table A5

Simulation of Level of Precision for MLCS Estimates of Poverty Rate by State/Region for Two Sampling Alternatives,
Based on 2015 MPLCS Data

Sampling alternative 1 Sampling alternative 2

State/Region DEFF (MPLCS)
Sample households Margin of error Sample households Margin of error

Kachin 3.6 864 6.0 1,152 5.2

Kayah 2.3 864 6.1 1,152 5.3

Kayin 2.2 864 5.2 1,152 4.5

Chin 1.0* 864 2.2 1,152 1.9

Sagaing 2.2 864 3.7 1,152 3.2

Tanintharyi 2.3 864 5.5 1,152 4.7

Bago 3.3 864 4.6 1,152 4.0

Magway 2.8 864 4.0 1,152 3.5

Mandalay 2.0 864 3.5 1,152 3.0

Mon 3.2 864 4.6 1,152 4.0

Rakhine 2.1 864 6.8 1,152 5.9

Yangon 1.2 1,152 3.1 1,728 2.5

Shan 3.2 1,152 5.2 1,728 4.2

Ayeyarwady 9.8 1,152 6.9 1,728 5.6

Nay Pyi Taw 3.4 864 3.9 1,152 3.4

*The estimates of the DEFF that were calculated with a value of less than 1 were rounded up to 1 in order to provide more realistic results.

In determining the most effective sampling alternative, it is also important to consider the effects of
non-sampling error on the accuracy of the survey results. The accuracy of survey estimates depends
on the total error, which is measured by the mean square error, defined as follows:

MSE = Variance + Bias2

The variance is the square of the standard error, so it is based on the sampling variability; the variance
decreases in relation to the sample size. The bias comes mostly from systematic non-sampling
errors. As the sample size increases the non-sampling error and corresponding bias tend to increase
since it is more difficult to manage and control the quality of a larger survey operation. The following
graph shows the relationship of the sample size to both sampling and non-sampling errors:

64
 Error
Total error

MSE = Variance + Bias2

Non-sampling error Sampling error

Sample size

It can be seen in this graph that the when the sample size is increased beyond a certain point the
accuracy of the survey estimates actually begins to decrease because of the increased non-sampling
errors.

Given the relatively small reduction in the margins of error under the second sampling alternative,
and considering the effects of potentially higher non-sampling errors, the World Bank Consultants
recommended that the first sampling alternative with a total sample size of 1,152 clusters and 13,824
households be used for the MLCS 2017.

Summary

The overall sample size depends on the level of precision that is required for individual domains,
which in the case of MLCS 2017 is States and Regions. One could use the Multi Indicator Cluster
Survey (MICS) formula below for determining the sample size an individual domain, you would then
need to add up the sample size for all the domains to the national level.

where:

n = is the required sample size

4 = is a factor to achieve the 95% confidence level

r = is the predicted or anticipated prevalence (coverage rate) for the indicator being
estimated

1.1 = is a factor necessary to raise the sample size by 10% for non-response

f = is a shortened symbol for design effect (deff)

0.12r = is the margin of error to be tolerated at the 95% level of confidence, defined as 12
percent of r (12 percent thus represents the relative sampling error of r)

p = is the proportion of the total population upon which the indicator, r, is based, and

nh = is average household size.

This formula estimates the sample size for one domain, applying only to one indicator.

65
 An important consideration is that we are tabulating many different indicators from the MLCS data
and each indicator would need a different sample size. Although you could choose the largest sample
size needed for all the indicators, this would result in a very large sample size that is not affordable,
so there is a need to compromise after examining the sample size requirements for the different
indicators as well as the survey budget.

In the case of the MLCS 2017 sample design, we did not have to use a formula such as that above
based on guessing the different parameters, because we had actual data from the MPLCS 2015 and
these data provided a much more accurate estimate of the level of precision through the simulation
study described above.

66
Annex B: MLCS 2017: Listing Form

State/Region Total Number of Households in the EA

District

Township Lister Name

Ward/Village tract

EA Listing Date

1 2 3 4 5 6 7 8
Building code from Address / Number Status of Building Status of Residential Name of Head of Serial Tick if
map Description of for units building Household Number (only selected (1-
Building without if 1 in col 4 12) from
Residential .............1 Occupied .............1
physical AND 1 in col Household
Only Commercial ...2
number Unoccupied….2 END 5) Selection
END
Table
Only Institution.....3 END Destroyed….3 END
Annex C: Non-enumerated and non-visited EAs and reason

CLUSTER EACODE STATE/REGION DISTRICT TOWNSHIP VILLAGE/WARD U/R Reason for replacing

1 25 10106712001 Kachin Myitkyina Gawt Maw Hsawlaw Ward 2 SECURITY ISSUES KIA

2 65 10304718002 Kachin Bhamo Mansi Mai Bat 2 SECURITY ISSUES KIA

3 140 20202703001 Kayah Bawlakhe Hpasawng Ba Han Lawt 2 SECURITY ISSUES KNPP

4 141 20202706001 Kayah Bawlakhe Hpasawng Mawchi 2 SECURITY ISSUES KNPP

5 216 30452717001 Kayin Kawkareik KYAIK DONE (Sub-township) Khwar Hay 2 SECURITY ISSUES KNU

6 432 60353701014 Tanintharyi Kawthoung Karathuri Ma Thay 2 MAP/DATABASE PROBLEM FROM CENSUS

7 433 70101102039 Bago Bago Bago Myo Twin (East-Kha Gway) 1 FOREST – NO HOUSEHOLDS

8 500 70406777701 Bago Thayarwady Nattalin Pyin Ma Khaung 2 FOREST – NO HOUSEHOLDS

9 725 110101705010 Rakhine Sittwe Sittwe Aung Daing 2 CENTRAL RAKHINE, SECURITY ISSUE IN Q4

10 730 110102737002 Rakhine Sittwe Ponnagyun Kar Di 2 CENTRAL RAKHINE, SECURITY ISSUE IN Q4

11 2102 110101727029 Rakhine Sittwe Sittwe Bu May 2 CENTRAL RAKHINE, SECURITY ISSUE IN Q4

12 2107 110203718003 Rakhine Mrauk-U Minbya Shwe Ta Mar 2 CENTRAL RAKHINE, SECURITY ISSUE IN Q4

13 757 110301102005 Rakhine Maungdaw Maungdaw Myo Ma (South) 1 NORTHERN RAKHINE

14 758 110301708001 Rakhine Maungdaw Maungdaw Kha Maung Seik 2 NORTHERN RAKHINE

15 759 110301723001 Rakhine Maungdaw Maungdaw DonePaik 2 NORTHERN RAKHINE

16 760 110301744001 Rakhine Maungdaw Maungdaw Myaw Taung 2 NORTHERN RAKHINE

17 761 110301758004 Rakhine Maungdaw Maungdaw Ka Nyin Taw 2 NORTHERN RAKHINE

18 762 110301769005 Rakhine Maungdaw Maungdaw (Du) Chee Yar Tan 2 NORTHERN RAKHINE

19 763 110301784003 Rakhine Maungdaw Maungdaw Kyauk Pan Du 2 NORTHERN RAKHINE

20 764 110302103001 Rakhine Maungdaw Buthidaung Myo Thit 1 NORTHERN RAKHINE

21 765 110302714001 Rakhine Maungdaw Buthidaung Nga Yant Chaung 2 NORTHERN RAKHINE

22 766 110302727004 Rakhine Maungdaw Buthidaung AwRaMa(a)Yee 2 NORTHERN RAKHINE

23 767 110302739003 Rakhine Maungdaw Buthidaung Ka Kyet Bet Kan P 2 NORTHERN RAKHINE
 CLUSTER EACODE STATE/REGION DISTRICT TOWNSHIP VILLAGE/WARD U/R Reason for replacing

24 768 110302758001 Rakhine Maungdaw Buthidaung Hpa Yar Pyin Thein Tan 2 NORTHERN RAKHINE

25 769 110351703005 Rakhine Maungdaw Taungpyoletwea(Sub-Tsp) Kun Thee Pin 2 NORTHERN RAKHINE

26 2110 110301765005 Rakhine Maungdaw Maungdaw Pan Taw Pyin 2 NORTHERN RAKHINE

27 2111 110302748001 Rakhine Maungdaw Buthidaung Pyin Chaung 2 NORTHERN RAKHINE

28 925 130206707004 Shan (South) Loilen Mongkung Hkay Ong Awayt 2 SECURITY ISSUES RCSS, POPPY GROWING

29 928 130253701009 Shan Loilen Kar Li (Sub-tsp) Keng Lun 2 SECURITY ISSUES SSA, POPPY GROWING

30 929 130255703003 Shan Loilen Mongsan (Hmonesan) Hopong 2 SECURITY ISSUES SSA, POPPY GROWING
(Sub-tsp)

31 942 130404726001 Shan Lashio Tangyan Tawng Hyoe 2 NEAR WA AREA, NO ACCESS

32 978 135003701016 Shan (E) Mine Lar Nam pan War Maing Phar Kyan 2 WA AREA, NO ACCESS
(Wa D)

33 979 135601704003 Shan Wan Hong Mong Hsat Mong Yan (Tar Ku) 2 WA AREA, NO ACCESS

34 980 136002703001 Shan Minemaw Nhar Wee Waine Kaung 2 WA AREA, NO ACCESS

35 981 136006708002 Shan Minemaw Mut Right Sa Out 2 WA AREA, NO ACCESS

36 982 136011703006 Shan (N) Minemaw Kaut Maing Sa Maung Laing 2 WA AREA, NO ACCESS

37 983 136105704003 Shan Wain Kaung Pan Yan Young Sai 2 WA AREA, NO ACCESS

38 984 136206702001 Shan Mine Pauk Mine Nyin Nang Paw 2 WA AREA, NO ACCESS

39 1100 150103101013 Naypyitaw Naypyitaw(North) Oaktarathiri Oke Ta Ra Thi Ri 1 ONLY GOVERNMENT OFFICE BUILDINGS

40 1139 150203101001 Naypyitaw Naypyitaw(South) Za Bu Thi Ri Zay Ya Theik Di 1 PRESIDENTS COMPOUND

Annex D: CSO and non CSO people used in listing and survey teams

Number of issued Number of survey Number of listing Number of Number of Number of in field Number of listers Number of listers Total team
EAs teams teams supervisors - CSO enumerators data entry - CSO - non-CSO members

Kachin 72 1 1 1 3 1 1 1 7

Kayah 72 1 1 1 3 1 - 2 7

Kayin 72 1 1 1 3 1 - 2 7

Chin 72 2 2 2 6 2 - 4 14

Sagaing 72 2 2 2 6 2 1 3 14

Tanintharyi 72 1 1 1 3 1 1 1 7

Bago 72 1 1 1 3 1 1 1 7

Magway 72 1 1 1 3 1 - 2 7

Mandalay 72 1 1 1 3 1 - 2 7

Mon 72 1 1 1 3 1 1 1 7

Rakhine 72 2 2 2 6 2 - 4 14

Yangon 96 1 1 1 3 1 1 1 7

Shan 96 2 2 2 6 2 1 3 14

Ayeyarwady 96 2 2 2 6 2 1 3 14

Naypyitaw 72 1 1 1 3 1 - 2 7

Total 1152 20 20 20 60 20 8 32 140

Annex E: QA results generated from in-field data entry program

Figure 1a:

Average household size by week of fieldwork

6
Number of people

5
4
3
2
1
0
1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 20 21 22 23 24 25 27 28 29 30 31 32 33 34 35 36 37 38 40 41 42 43 44 45 46 47 48 49 50 51

Week of survey (1 to 51)

Figure 1b:

Average household size by enumerator

5
Number of people

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Enumerator ID
 Figure 2a:

Average number of food products consumed in last 7 days by week of interview

Number of food items consumed 45

40

35

30

25

20
1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 20 21 22 23 24 25 27 28 29 30 31 32 33 34 35 36 37 38 40 41 42 43 44 45 46 47 48 49 50 51

Week of survey (1 to 51)

Figure 2b:

Average number of food products consumed in last 7 days by enumerator

50
Number of food items consumed

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Enumerator ID
Figure 3a:

Average number of non food products purchased by week of interview

Number of items purchased

24
22
20
18
16
14
12
10
1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 20 21 22 23 24 25 27 28 29 30 31 32 33 34 35 36 37 38 40 41 42 43 44 45 46 47 48 49 50 51

Week of survey (1 to 51)

Figure 3b:

Average number of non food products purchased by Enumerator

35
Number of items purchased

30

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Enumerator ID
 Figure 4a:

Average number of durables owned by week of interview

12

10
Number of durables

0
1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 20 21 22 23 24 25 27 28 29 30 31 32 33 34 35 36 37 38 40 41 42 43 44 45 46 47 48 49 50 51

Week of survey (1 to 51)

Figure 4b:

Average number of durables owned by enumerator

16
14
Number of durables

12
10
8
6
4
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Enumerator ID
Figure 5a:

Average length of interview (minutes) by week of fieldwork

200
Number of minutes

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 20 21 22 23 24 25 27 28 29 30 31 32 33 34 35 36 37 38 40 41 42 43 44 45 46 47 48 49 50 51

Week of survey (1 to 51)

Figure 5b:

Average length of interview (minutes) by enumerator

190
170
Number of minutes

150
130
110
90
70
50
30
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Enumerator ID
76
77
INQUIRIES
For further information about this publication and related statistics, contact the:

Central Statistical Organization

Ministry of Planning and Finance
Office No. 32
Nay Pyi Taw
http://www.csostat.gov.mm;
www.mmsis.gov.mm