Supplementary memorandum from the Office
for National Statistics
In response to evidence given to the committee
by the National Statistician on 28 January, additional information
was requested on questions 1-5 below (numbers in square brackets
provide a link to the transcript of oral evidence). The National
Statistician also offered to provide additional evidence on the
reliability associated with mid-year population estimates (listed
here as question 6). ONS evidence on all these questions is as
follows:
Question 1: How many enumerators experienced
verbal or physical abuse, in the course of their work? [Q245]
A total of 34 incidents were reported in 2007
by ONS field staff collecting social survey information. For comparison,
in that year ONS sent interviewers to visit approximately 340,000
addresses. Of these incidents, nine related to verbal or physical
abuse, seven to dog or bug bites and 18 to vandalising of interviewer
cars and theft from cars.
Question 2: The proportion of the population
who have ex-directory landlines. [Q251]
In 2006, 90% of households had fixed line telephones[1]
and, according to British Telecom, 48% of all landline users,
including non-BT customers, are ex-directory.[2]
Question 3: Further details about the methodology
used in scaling [/calibrating the IPS estimates using] the Labour
Force Survey (LFS). [Qs 261-267]
Details of the methods used to calibrate International
Passenger Survey (IPS) estimates using three year rolling averages
from the LFS were published in advance of the release of the population
estimates in 2007. The full details are in the paper "The
Use of Calibration in Estimating International In-migration to
UK Countries and the Regions of England" on the National
Statistics website at:
www.statistics.gov.uk/downloads/theme_population/Calibration.pdf
An overview of the uses made of LFS and IPS
figures in making local estimates was provided at the same time
and is at:
www.statistics.gov.uk/downloads/theme_population/FAQs.pdf
Methodological justification
(a) Reducing bias
The benefit of using the LFS is to distribute
international migrants to English regions and Wales based on where
they settled rather than relying on where they said they intended
to live, when they responded to the IPS.
In the IPS, arriving international migrants
are asked for an intended destination within the UK. These answers
give biased estimates for the distribution of in-migrants between
the countries of the UK and the government office regions of England.
Analysis of data from the IPS, the 2001 Census and the LFS showed
that the country and region in which recent migrants actually
lived in 2001 had a different distribution from that based on
where migrants sampled in the IPS stated that they intended to
live. Comparisons of IPS and LFS in other years also showed systematic
differences since the mid-1990s.
Additionally, since 2005, in-migrants in the
IPS have been asked whether they intend to stay in their stated
UK destination for the next 12 months or are likely to move on
to a different area. The data for 2005 suggested that in-migrants
who initially intended to stay in London, in particular, were
more likely to settle elsewhere than those indicating they intended
to stay elsewhere (18 and 7% moving on within a year, respectively).
(b) Effect on precision of estimates
An important consideration, in moving from one
method to another, was to assess whether it would adversely affect
the standard errors of the estimates at the country and region
level. Figures A and B show the relative standard errors for estimates
based on using data from the IPS for 2002 and 2004, comparing
the old method with the new method of calibrating using the new
methodology. These relative errors take account of the variability
introduced because the LFS is a sample survey. These show that
in ten out of the twelve areas shown, the relative errors were
lower in both years using the new methodology. For London and
the South East, the small relative increases in one or both years
reflected a smaller estimate of migration (but no change in the
absolute level of error).
Figure A
RELATIVE STANDARD ERRORS[3]
BY COUNTRY AND REGION: COMPARISON OF ESTIMATES BASED ON USING
ONLY THE IPS WITH THOSE USING IPS CALIBRATED TO LFS, 2002
Figure B
RELATIVE STANDARD ERRORS[4]
BY COUNTRY AND REGION: COMPARISON OF ESTIMATES BASED ON USING
ONLY THE IPS WITH THOSE USING IPS CALIBRATED TO LFS, 2004
For reference, the numbers of contacts made
with recent international in-migrants each year in the LFS (from
the much larger number of people who are actually interviewed)
are shown in Table A. Migrant numbers in the sample are dependent
not only on the LFS sample size but also on the actual number
of recent migrants in any area, which will fluctuate between areas
and over time. The relatively low numbers in some areas in individual
years reflects the relatively low number of migrants in those
areas.
Table A
NUMBER OF INTERNATIONAL RECENT MIGRANTS IDENTIFIED
IN THE LFS SAMPLE BY GOVERNMENT OFFICE REGION AND WALES, 1999-2006
| 1999 | 2000
| 2001 | 2002 |
2003 | 2004 | 2005
| 2006 |
| North East | 11 | 18
| 18 | 8 | 6 |
24 | 19 | 24 |
| North West | 41 | 38
| 43 | 33 | 61 |
37 | 61 | 48 |
| Yorkshire and the Humber | 75
| 50 | 54 | 60 |
49 | 57 | 96 | 54
|
| East Midlands | 26 | 19
| 24 | 45 | 41 |
42 | 49 | 65 |
| West Midlands | 40 | 41
| 41 | 53 | 37 |
46 | 32 | 43 |
| East | 65 | 88
| 48 | 72 | 72 |
46 | 60 | 91 |
| London | 190 | 226
| 147 | 219 | 165
| 172 | 167 | 162
|
| South East | 139 | 121
| 106 | 108 | 112
| 93 | 89 | 108
|
| South West | 54 | 45
| 43 | 53 | 67 |
45 | 76 | 55 |
| Wales | 7 | 15
| 19 | 17 | 11 |
13 | 20 | 13 |
As indicated above, the method of calibration introduced
in 2007 makes use of rolling three-year averages. The use of the
three year averages is particularly important in providing a more
robust indication of recent levels of settlers in the regions
than figures for any single year or use of the IPS alone.
Estimation of migrants settling in local areas
In distributing migrant numbers to local areas, no further
LFS calibration of IPS figures is undertaken beyond that described
above. The method of distribution varies by region and type of
migrant (see Table B), and in one instance (non-student migrants
to London) further use is made of the LFS three-year rolling averages.
This is purely for distributing the calibrated London totals to
intermediate geographies within London. In this instance, the
further distribution of these numbers, from intermediate geographies
down to individual London Boroughs is achieved using 2001 Census
distributions.
Table B
SUMMARY OF DATA USED TO ESTIMATE MIGRANT NUMBERS, BY AREA
AND TYPE OF MIGRANT
| Outside London: distributions of in-migrants
|
| Wales/English region level: | IPS + LFS in combination—calibration using one year
IPS and three year average LFS
|
| Intermediate level (NMGi): | IPS three year average
|
| LA level: | Census |
| London: distributions of non-student in-migrants
|
| Wales/English region level: | IPS + LFS in combination using calibration as above
|
| Intermediate level (NMGi): | LFS three year average
|
| LA level: | Census |
| London: distributions of student in-migrants
|
| Wales/English region level: | IPS + LFS in combination using calibration as above
|
| Intermediate level (NMGi): | N/A
|
| LA level: | Census |
| Distributions of out-migrants
|
| Wales/English region level: | IPS one year
|
| Intermediate level (NMGo): | IPS three year average
|
| LA level: | Propensity to migrate model
|
Source: "Frequently Asked Questions", published by ONS on 24 July at:
www.statistics.gov.uk/downloads/theme_population/FAQs.pdf
|
Question 4: Clarification about the method used by CLG
to allocate resources, including the predicted margin of error
by 2011. [Qs 270-271]
The Department for Communities and Local Government (CLG)
have provided the information included at Annex A.
Question 5: The outcome of the discussions between ONS
and the relevant chief statistician about the methodology used
by the NHS to allocate resources to PCTs, 2008-09. [Q 273]
Advice to Department of Health ministers on the resource
allocation formula for the NHS is provided by an independent committee,
the Advisory Committee on Resource Allocation (ACRA).
The Department of Health have provided information on the
2008-09 allocations, included at Annex B.
Question 6: What is the margin of error in mid year population
estimates and how much does it vary? [Q 258]
Mid year population estimates are produced using the cohort
component method which begins with population estimates by age
based on the 2001 Census and rolls these forward for each subsequent
year by adding births, subtracting deaths, adjusting for migration
and ageing the population by a year.
The Census results incorporated estimates of under-enumeration,
based on a sample survey, the Census Coverage Survey (CCS), and
for this reason are subject to sampling error. Figures indicating
the margins of error as a result of this, at national and local
levels, were published at:
www.statistics.gov.uk /census2001/annexa.asp
These error estimates do not however incorporate the additional
uncertainties associated with the adjustments made to the Census
results for current population estimation purposes. These adjustments
are described in Section 2 of the Local Authority Population
Studies: Full report, available from:
www.statistics.gov.uk/downloads/theme_population/LAStudy_FullReport.pdf
Births and deaths in the UK are registered and are assumed
to be complete. So the main source of statistical uncertainty
is associated with the international migration component. This
uses Total International Migration (TIM), which is compiled from
several sources of information. The principal source is information
on the intentions of recent migrants sampled in the IPS, but account
is also taken of groups not covered by the IPS, such as asylum
seekers and those arriving from Ireland, and adjustments made
for those who change their intentions. TIM estimates are distributed
to local areas using additional sources (eg Labour Force Survey
and Census). Survey estimates such as the IPS and LFS have known
sampling errors. All sources also have non-sampling errors such
as under-coverage. No single error measure is available to summarise
this complex process.
However, the sampling error of the main individual component,
the IPS, is available. The effect on these sampling errors of
using the LFS to calibrate country and regional distributions
was described in the reply to Question 3. This analysis incorporated
the effect of sampling variation in the LFS.
In addition to sampling variation in any particular data
source and year, the following needs to be taken into account
in considering the overall margin of error in population estimates:
— Any residual discrepancies between Census and
the rolled forward estimates (for example definitional differences,
sampling and other errors in the Census estimates and sampling
and other errors in the post-Census corrections). These are considered
to be relatively small at national level, but may make a larger
relative contribution to the limited number of local authority
estimates where there were particular problems at Census.
— Accuracy of local migration estimates from the
Census, used in distributing migrant estimates as described in
Table B (for example due to differential levels of imputation
and sampling errors in the imputed values).
— Changes since Census in the local authority distribution
of migrants within intermediate areas (Table B indicates how these
distributions are used).
— The cumulative effect of sampling and other errors
in annual population estimates as we move further from the base
Census year. As is standard practice following a Census, the size
of these errors will be assessed using the 2011 Census and revisions
made accordingly.
These are complex issues and ONS have not published comprehensive
estimates of the margins of error, as indicated to the Committee
by the National Statistician.
International Passenger Survey (IPS)
For 2005, the relative standard error for the total IPS in-migration
estimate of 496,000 migrants was 4%. This gives a range of between
459,000 and 533,600 as the 95% confidence interval[5]
for the IPS estimate of the number of migrants entering the UK
during 2005.
The relative standard error was 5% for the 2005 out-migration
estimate of 328,000 migrants. This gives a range of 297,000 to
360,000 migrants as the corresponding 95% confidence interval.
Further details of the effects of sampling error on the migration
estimates by various characteristics are given in Table 4.2 at
pages 31-32 of the Annual Reference Volume MN Series no. 32, International
migration: Migrants entering or leaving the United Kingdom and
England and Wales, 2005. This is available from:
www.statistics.gov.uk/downloads/theme_population/MN_no32.pdf
Entries in this table show that estimates based on the sampling
of passengers on certain routes have much larger errors associated
with them. For other tables in this publication, as a guide, the
standard error for an estimated 1,000 migrants will be in the
region of 40%. This reduces to about 10% for an estimate of 40,000
migrants. Thus, generally speaking, the larger the sample supporting
a particular estimate, the smaller its sampling error.
4 March 2008
Annex A
METHOD USED BY CLG TO ALLOCATE RESOURCES
Formula Grant is a fixed amount set by the Spending Review
that is then distributed to local authorities by Communities and
Local Government. Allocations are no longer made on an annual
basis, but are multi-year settlement allocations made at the beginning
of the Spending Review period. For example, provisional allocations
for 2008-09, 2009-10 and 2010-11 were announced in December 2007.
Multi-year settlements have been welcomed by Local Government
as they provide predictability and stability in the funding that
is provided by Central Government.
In order to provide these multi-year settlements, projections
of population and the taxbase (number of Band D equivalent properties)
are used as the main drivers. For all other indicators, the data
remains the same in all years. Therefore, rather than using the
mid-2006 estimates of population as the main measure of population
in 2008-09, 2009-10 and 2010-11, the 2004-based sub-national population
projections for 2008, 2009 and 2010 have been used.
For each authority, the distribution takes into account the
relative needs and potential to raise resources locally relative
to all other authorities providing the same services. There is
also a central allocation and a floor damping mechanism. There
is no guaranteed amount of grant per head.
The calculation of relative needs is based on mathematical
formulae for each of the main services provided. For each formula,
except for Highways Maintenance, Local Authority Education Functions
and Capital Finance, the client group for the service is measured
by the ONS sub-national population projections. Additional socio-economic,
demographic and geographic factors are used to reflect the variations
in need and costs.
In addition, to be able to determine the relative needs amount
from the relative needs formulae, the relative resources amount
and the central allocation, the population projections are also
used as part of the calculations.
The final allocation of grant incorporates a floor damping
mechanism. This sets a lower limit (floor) to all authorities
change in grant from year to year. This means that authorities
whose grant, after the relative needs amount, relative resource
amount and central allocation has been calculated (referred to
as grant before damping), is below this floor, receive an additional
amount to ensure that they receive the floor. As this mechanism
is self-financing, authorities whose grant before damping is above
the lower limit have their grant scaled back. The decision on
what floor levels are set is taken by Ministers.
The revised 2004-based sub-national population projections
published on the 27 September 2007 have been used as they were
the most up-to-date data available when the calculations of the
2008-09, 2009-10 and 2010-11 settlement were undertaken in late
2007.
It is not possible to quantify what difference there would
be for individual authorities, had different population data been
used in the distribution of formula grant. This is because different
decisions may have been taken by Ministers on the methodology
used and the level of floor damping; and no alternative population
estimates are available for years later than 2006.
However, at the overall England level, there would be no
difference, as the amount of grant to be allocated would not change.
Annex B
INFORMATION FROM THE DEPARTMENT OF HEALTH ON RESOURCE
ALLOCATION TO PCTS IN 2008-09
ACRA is reviewing the resource allocation formula for the
NHS. They requested an extension to its work programme, which
Ministers agreed to, meaning that it was not possible to implement
the new formula for 2008-09 and still give the NHS sufficient
time to plan for the year ahead. Therefore, it was decided that
it was better to announce a one-year allocation to PCTs and allocate
the following two years as soon as possible after ACRA had finished
their work. The resource allocation formula for the NHS was frozen
for 2008-09, no part of the formula was updated included populations,
and every Primary Care Trust (PCT) received the same uplift to
their budget of 5.5%. The formula was frozen to avoid changes
in 2008-09 that may then have to be reversed in 2009-10 when the
new formula is implemented.
It is right to give ACRA additional time to finalise their
work to ensure that they will produce a formula which is as robust
as possible. An additional benefit of the delay in allocations
for 2009-10 and 2010-11 is that we will be able to use the latest
population projections which will be available in 2008. Allocations
for 2009-10 and 2010-11 will be made to the NHS by summer 2008
and will use the most up-to-date population projections data available
from the ONS.
1
ONS Omnibus Survey Back
2
Source: BBC News 24, Joining the ex-files, 11 July 2006, http://news.bbc.co.uk/1/hi/magazine/5168570.stm Back
3
The relative standard error is the ratio obtained by dividing
the standard error of an estimate by the estimate itself. Back
4
The relative standard error is the ratio obtained by dividing
the standard error of an estimate by the estimate itself. Back
5
If the survey were repeated many times under the same conditions,
then 95% of intervals constructed in this way would contain the
true value. Back
| 
