2011 Household Survey Sample Redesign
Project Board Meeting
1 March 2012
Agenda item 5
2011 MPS Sample Design
Purpose of the paper
The purpose of this paper is to present the proposed MPS sample design as part of the 2011
household sample redesign.
Issues for Household Surveys Sample 2011 Redesign Board discussion
1. Does the Board have any comments on the approach taken to the sample design?
2. Does the Board feel that the EPS constraint should be relaxed, noting the trade off of cost
savings against likely increases in sampling error for other estimates?
3. Does the Board have any concerns over the slight reduction in overall sample which comes
as a result of how the state accuracy targets have been set?
4. Does the Board have any concerns over the increase required to the NT sample?
Note
This paper contains the key information outlining the details of the sample design. Extended
information relating to the key design decisions is documented in the Appendicies, however it is
not necessary to read these to gain an understanding of the overall design.
2011 MPS Sample Design
1 Introduction
This paper presents the proposed MPS sample design as part of the 2011 household
sample redesign. The sample design described in this paper concerns the master sample
of areas for the private dwelling component of the MPS sample.
The key parameters specifying the sampling strategy for a multi-stage area sample are the
cluster sizes and the number of clusters to select for different areas of Australia. This paper
presents the required values of these parameters in order to satisfy output objectives
specified by Labour Branch. The parameters have been derived using an optimisation
procedure which aims to provide most cost-effective sampling strategy by trading off the
precision (variance) of estimates with the corresponding cost of obtaining these estimates.
The sample design presented incorporates the constraint that all dwellings are selected with
equal probability within each State (EPS constraint). The impact of removing this constraint
on the sample allocation and costs are noted.
2. Sample Design Objectives
The broad sample design objectives were endorsed at the June 2011 Sample Redesign
Board meeting
. In late 2011 the interpretation of the broad objective of 'no
change to the relative priority of the State-level estimates' was further clarified
. It has also been agreed there will be no change to the relative priority
for estimates of employment and unemployment.
3 Sample Design Overview
Scope of sample design
This paper concerns the design of the private dwelling sample for the MPS. The monthly
private dwelling sample for the MPS is selected by sub-sampling dwellings from a master
sample of fine areas (small aggregations of mesh blocks). Each fine area in the master
sample provides a single cluster of dwellings each month over the five-year design period.
The sample design can be summarised by the following aspects:
sampling strata;
stratum cluster sizes, which are number of dwellings to approach per month within each
selected fine area; and
number of clusters to select in each stratum.
The MPS sample also consists of a special dwelling component selected from a frame of
establishments. Importantly, the sample allocation for the private dwelling sample implicitly
determines the sample size of special dwellings (assuming maintaining the strategy of
selecting persons in special and private dwellings with equal probability).
Stratification
Fine areas are assigned to sampling strata defined by the combination of SA4, the finest
geography for routine dissemination of labour statistics, and an 'area type' classification of
areas. Dwelling density and remoteness are two key factors for determining area type. The
stratification and area type classification are discussed in more detail in Appendix 2.
Equal probability constraint
Advice is sought from the Board on whether the sample design should incorporate the
constraint of selecting all dwellings within a State with equal probability (EPS design).
Traditionally the MPS design has been EPS.
Does the Board feel that the EPS constraint should be relaxed, noting the trade off of cost
savings against likely increases in sampling error for other estimates?
Method for choosing cluster sizes
The optimum cluster size for a stratum balances the trade-off between cost and variance
which arises when selecting a sample with geographic clustering.
s.
Interviewer activity simulation cost model
For the 2011 sample design project a new procedure for modelling interviewer costs has
been used for the sample optimisation task. The costs of alternative sample designs have
been compared using a simulation model of interviewer activity. The simulations measure
the time spent by interviewers making telephone and face-to-face call attempts, travelling
and interviewing, as well as measure distances travelled (to account for motor vehicle
allowance payments).
4 Optimal cluster size choices
5. Properties of sample with the recommended cluster sizes
This section presents the State-level distribution of sample for the proposed design, which
produces the RSEs indicated in Table 2.1. Comment is sought from the Board on the
allocation, in particular the high sample allocation in NT. Based on the design outlined
above, the expected numbers of dwellings selected over the coming years are as follows:
Does the Board have any concerns over the slight reduction in overall sample which comes as
a result of how the state accuracy targets have been set?
Advice is sought from the Board on the sample allocation in NT, in light of the sample
allocation and expected sampling skip for the proposed design.
Does the Board have any concerns over the increase required to the NT sample?
6 Impact on cost due to changes to sample clustering
Appendix 1
Appendix 2 Stratification and Area type classification
Stratification scheme
The stratification of the area frame for the 2011 sample redesign is the combination of SA4
and area type class. The number of strata defined within an SA4 is the number of distinct
area types which defined within it. Including the SA4 dimension ensures the sample design
controls the amount of sample selected for each SA4. Incorporating area type facilitates
adoption of the most sampling scheme within each of the variety of areas contained in each
SA4. There are a small number of special strata (Pre-determined growth, Indigenous and
secure apartment building), which span the SA4s within the State.
Area type classification
The area type classes, presented in Table 1 below, primarily aim to combine geographic
areas with similar costs for data collection and possibly similar levels of geographic
clustering of labour characteristics. The area type classes may also divide distinguish areas
in which are more homogenous with respect to the labour characteristics of the persons
within. For example, the demographics of persons in the "Capital City Past Growth" class,
which largely contains outer suburbs, are likely to differ from the demographics of persons
living in the inner suburbs.
The capital city area types are defined for areas within SA4s which are define the capital city
boundary of the State. The variety of area type classes distinguish areas of high dwelling
density, areas with recent high growth (typically outer suburbs), pockets with many secure
apartment buildings and outlying rural areas within the capital city boundary. A key
distinction between the area types outside of the capital cities are Self Representative Areas
(SRA) and non-SRA. The distinction is based on dwelling density, and in non-SRA areas a
modified sampling procedure is used to increase sample clustering. This ensures
interviewer workloads can be created from sample selections which are relatively close
together. The Remoteness and Urban/Rural classifications defined on the ASGC were used
as the basis for classifying areas to the SRA rural area type and the different non-SRA area
types.
Table 2 presents the distribution of dwellings across the area types in each State, providing
perspective on the relative size of the classes. Nationally, the only area type classes with
more than 2.5% of the population of dwellings are "Capital City Settled", "Capital City Inner
City", "Capital City Past Growth", "Rest of State SRA Major Cities", "Rest of State SRA
Medium towns", "Rest of State SRA Small Towns" and "Rest of State non-SRA Least
Remote".
Table 1: Area types for the 2011 MPS stratification
Number
Area Type
Explanation
1
Capital City Settled
Default Capital city areas
2
Capital City Inner City
Population density > 3125 persons per sq
km CHECK
3
Capital City Secure Apartment
SA1s with a high proportion of secure
Buildings
apartment building *
4
Capital City Past Growth
growth of more than 10% since 2006
5
Capital City Rural
Rural outskirts of capital cities
6
Rest of State SRA Major Cities
Population 100k +
7
Rest of State SRA Medium towns Population 12-100k
8
Rest of State SRA Small Towns Population <12k
9
Rest of State SRA Rural
Rural SRA areas
10
Rest of State non-SRA Least
Remote
11
Rest of State non-SRA Remote
12
Rest of State non-SRA Very
Remote
13
Pre-determined growth
Capital city** greenfield areas with large
growth expected in the next 5 years
14
Indigenous
Notes:
* SA1s with at least 100 apartments OR 3 or more buildings containing a total of 75 or more
apartments
** NSW contains pre-determined growth SA1s which are outside the capital city
Table 2: Distribution of State's dwellings across area type classes (% points)
14
0.0
0.0
0.3
0.3
0.5
0.0
13.5
0.0
Summary of method for area type assignment
In general area types have been assigned to areas at the SA2 level. Situations in which
multiple area types can exist in an SA2 are:
the SA2 contains multiple localities, the localities in the SA2 may belong to different area
types
the SA2 contains SA1s belonging to the secure apartment building, pre-determined
growth or Indigenous area types
Appendix 4 Overview of interviewer activity simulations and assumptions
Simulation model
For the 2011 MPS sample redesign the cost of alternative sample designs have been
compared using a simulation model of the activity undertaken by interviewers. The purpose
of the models is to measure the relative rather than absolute PSO costs for sample design
alternatives. The simulation model measures the cost of
interviewer time and
motor vehicle
allowance costs associated with travel. The interviewer time includes time for:
travelling to workloads,
travelling within workloads,
making call attempts by phone and in the field,
conducting interviews
The simulations include all of the activity for callbacks. The simulation model does not
measure indirect costs, which should be similar for the alternative sample designs.
