WP2-frame quality

WP2-frame quality
Johan Fosen*, in cooperation with Li-Chun Zhang
Budapest, 21-22 April 2016
1
* Contact information: [email protected]
Outline
• Frame errors as opposed to output errors
• Frames. Definition
• Frame errors
• sources for quality measurement and
frame types
• Types of frame errors and frame types
2
Frame errors as opposed to output errors
• WP3: quality of output statistics, e.g. of
register-based employment statistics
• WP2: quality of frames, e.g. of
• Census
• Business register
• Population dataset
• Address register
3
Frames. Definition
• Frame: any list, material or device that delimits, identifies, and allows
access to the elements of the target (survey) population.
• e.g. Census, business register, population dataset, Address register.
• Lessler and Kalsbeek (1992) :
• The target population is finite and of identifiable elements.
• The sampling units are not necessarily the (target) population elements.
• It must be possible to locate and distinguish the population elements.
• Some mechanism must exist for linking the target population to the
sample.
• More than one type of linkage may exist between population elements
and sample units.
• Auxiliary information for design and estimation must be known
throughout the population
4
Types of frame error
• Coverage error
• alignment error
• domain classification error
• unit error
• contact information error
5
Frame error: Coverage error
• Correct frame: unit is in the frame and unit
is in the population
• Undercoverage:
• {unit not in frame} AND {unit in population}
• Overcoverage:
• {unit in frame} AND {unit not in population}
6
Frame error: alignment error
• BU = base unit = atomic unit
• Most common in frames for social statistics:
person
• CU = Composite unit: other type of units.
Aggregations of BU
• e.g. address, family or dwelling household
7
Frame error: alignment error, continued
• Connection between BU’s and CU’s: alignment
table
BU (person)
Adam Smith
CU-1
(home
address)
Smith-SO19xxx
CU-2
(business
address)
-
Eva Hanford
Smith-SO19xxx
Highfield-SO17xxx
Telepho
ne Nr.
123456
132415
324151
Mark Smith
Smith-SO19xxx
London-WS5Dxxx
-
Alan Smith
Smith-WC1Exxx
London-WC1Exxx
654312
Sarah
Sommers
Sommers-L17xxx
London-NQ6Axxx
-
…
…
…
8
Frame error: domain classification error
• a unit’s frame-domain should equal its populationdomain.
• Otherwise: domain classification error.
•
Frame
Domain
Classification
1
2
…
H
Population Domain Classification
1
2
…
H
N11
N21
N12
N22
N1H
N2H
NH1
NH2
NHH
9
Frame error: unit error
• Sometimes a CU is non-existing in registers
and needs to be constructed, e.g. living
household
• unit error: errors in the construction, e.g.
when considering two living households as
only one.
10
Frame error: unit error, continued
BU (person)
Adam Smith
Dwelling
household
123
Living
household
(constructed)
4567
True
living
household
4567
Eva Hanford
123
4567
4567
Mark Smith
123
4567
4568
Alan Smith
124
4570
4570
Sarah
Sommers
125
4570
4570
…
…
…
11
Frame error: Contact information error
BU (person)
Adam Smith
CU-1
(home
address)
Smith-SO19xxx
CU-2
(business
address)
-
Eva Hanford
Smith-SO19xxx
Highfield-SO17xxx
Telepho
ne Nr.
123456
132415
324151
Mark Smith
Smith-SO19xxx
London-WS5Dxxx
-
Alan Smith
Smith-WC1Exxx
London-WC1Exxx
654312
Sarah
Sommers
Sommers-L17xxx
London-NQ6Axxx
-
…
…
…
12
sources for quality measurement and
frame types
Sources for quality measurement
Frame type
Census
Postenumeration/
coverage
survey
X
Quality
survey
Ongoing
survey
"Sign-oflife"
-survey
X
Business reg.
X
X
Pop.dataset
X
X
Address register
13
Types of frame errors and frame types
Types of frame errors
Frame type
Coverage
Census
X
Business reg.
X
Pop.dataset
X
Alignment
X
Address
register
14
Domain
classification
X
Unit
Contactinfo
References
• Lessler, J.T. and Kalsbeek, W.D. (1992). Nonsampling Error in
Surveys. Wiley.
• Zhang, L.-C. (2011). A unit-error theory for register-based
household statistics. Journal of Official Statistics, 27, 415-432.
• Zhang, L.-C. (2012). Topics of statistical theory for registerbased statistics and data integration. Statistica Neerlandica, 66,
41-63.
• Zhang, L.-C. (2016). On the quality of frames for social statistics,
in progress.
15