Analysis of data and evidence for Tower Hamlets by fdh56iuoui


									2004 Local Authority studies
Analysis of data and evidence for
Tower Hamlets

Office for National Statistics, September 2004

Executive summary                                                          3
1 Introduction                                                             4
2 Background information on Population Estimates and One Number Census     4
3 Analysis                                                                 5
 3.1 Enumeration                                                           5
 a) Analysis of Council Tax data                                           6
 b) Analysis of Enumerator Record Books (ERBs)                             8
 c) Analysis of Council Tax based response rates by 2001 Ward              8
 d) Enumeration Information                                                9
 e) Estimated response compared to other areas                             9
 f) Qualitative information obtained about the Census enumeration         10

 3.2 Census Coverage Survey                                               10
 a) Missing sample                                                        10
 b) Sample balance                                                        10
     Dummy form distribution                                              11
     Recalculated Hard to Count score                                     13
 c) Localised undercount and the CCS                                      17
 d) CCS fieldwork and response rates                                      20

 3.3 One Number Census process                                            20
 a) Response rates                                                        20
 b) Collapsing in the ONC process                                         20
 c) Outliers                                                              21
 d) Stratification                                                        21
 e) Household and person imputation results                               24

4 Population definitions                                                  24
5 Processing                                                              25
6 Other - communal establishments                                         25
7 Other -administrative sources                                           25
8 Conclusions and recommendations                                         26

Executive summary

The Office for National Statistics (ONS) has now concluded a series of studies designed to improve
population estimates in the areas that proved to be hardest to count in the 2001 Census in England
and Wales. The results of this analysis, involving experts from local government and other bodies,
has confirmed the findings contained in reports by the Statistics Commission and the Local
Government Association. These reports concluded that the One Number Census (ONC) worked
well in most areas but that there were a few cases where it was not able to sufficiently adjust for
under-enumeration in exceptional circumstances.

More information can be found at

A report was produced for each Local Authority examined, and this report sets out the analysis
and conclusions for the Local Authority of Tower Hamlets. It provides background information
about population estimates and the Census, and describes analysis carried out in certain key areas
- enumeration, Census Coverage Survey (CCS), One Number Census (ONC), population definitions
and processing. A comparison with administrative sources is also shown. Conclusions and
recommendations are detailed at the end of the analysis.

Tower Hamlets was selected for detailed analysis as part of the local authority population studies, as
it was identified as an area where there was a significant risk of an under-estimate of the population
by the 2001 Census.
This detailed analysis identified large differences between Census and Council Tax figures, however,
further analysis concluded that this was due to definitional differences.

Further analysis identified no significant problems with Census or CCS enumeration, no evidence of
localised under-enumeration and concluded that the assumptions made by the One Number Census
were appropriate for Tower Hamlets.

After consideration of all the evidence, it was decided that the current estimate remains the best
estimate of the population of Tower Hamlets in 2001, and therefore no adjustment has been made.

2004 Local Authority studies:
Analysis of data and evidence for Tower Hamlets

1 Introduction                                         Table 2.1
Tower Hamlets (Census 2001 population                  Mid Year Estimate (MYE) series and revi-
196,106) is an area within Inner London,               sions for 2000, 2001 and 2002
bordered in part by the River Thames. There
                                                                           Population            Change since 2000
are some characteristics of Tower Hamlets that                                                                MYE
make population enumeration and estimation             2000 MYE                   186,700
difficult. There are blocks of flats with controlled   Census                     196,106                    9,400
access, and also the area is home to some halls        2001 MYE                   196,600                    9,900
of residence for the University of North London
                                                       Revised 2001 MYE           201,600                   14,900
and the City & Islington college.
                                                       Revised 2002 MYE           206,800                   20,100

In order to estimate census undercount across
England and Wales contiguous Local Authorities
                                                       Tower Hamlets received an adjustment with
(LAs) were grouped together to form Estimation
                                                       the revised 2001 Mid-Year Estimates on 26/
Areas (EAs) which consisted of about half a
                                                       09/03. More information can be found at
million population. In most cases EAs consisted
of several Local Authorities. Where an LA was
sufficiently large the EA consisted only of that
one LA, i.e. the LA was an EA in itself. LAs were
divided into Enumeration Districts (EDs) which
                                                       Table 2.2 shows the Confidence Interval
were pre-planned workload areas of around
                                                       associated with the ONC population estimate.
200 households within which an individual
enumerator worked. EDs did not cross Ward
boundaries. Tower Hamlets is part of a Census          Table 2.2
Estimation Area containing two other Local             95 per cent Confidence Interval for Tower
Authorities (LAs), Hackney and Islington.              Hamlets One Number Census estimate
Tower Hamlets was selected for investigation for       Relative C.I         +/-     Confidence Interval Range

a number of reasons:                                   3.9%               7,648             188,458        203,754

   ·   The area had a large difference between
       the Council Tax data and the Census
                                                       Mid Year Population Estimates
                                                       The following tables provide information on
   ·   There were 10 outliers identified within        Tower Hamlets’ population between 1991 and
       the ONC process for the Local Authority,        2000. These indicate whether the population
       which is a substantial number.                  has increased or decreased throughout the
   ·   The area demonstrated the largest               intercensal period, whether the area is one of
       difference in its Hard to Count profile         large change and what effect migration has had
       between 1991 and 2001.                          on the area’s population.

2 Background information on Population
Estimates and One Number Census
Table 2.1 sets out the Census and mid year
estimates (MYEs) for Tower Hamlets for the
period 2000 – 2002. Note that the 2001 and
2002 mid year estimates shown here are those
based on 2001 Census data, while the 2000 mid
year estimate is based on rolled forward data
from the 1991 Census.

                                                                                                                4                                                                           2004 Local Authority studies: Tower Hamlets

                        Table 2.3
                        Population profile for the MYE series up to 2000 (unrevised)
                        Mid Year                        1991        1992     1993     1994    1995       1996          1997      1998       1999      2000
                        Population Estimates            168.1      168.5     169.6   170.5    172.8     176.6          179.8     181.3      184.9    186.7

                        NB: The mid-year population estimates in this table are those that existed before the 2001 Census,
                        and have since been revised.

                        Table 2.4
                        Average annual changes in components of population estimates (unrevised)
                        Average annual change since mid-91                                            E&W                                    Tower Hamlets
                        Average annual change                                                           0.5                                            2.1
                        Average annual natural change                                                   0.3                                            1.8
                        Average annual change in migration                                              0.1                                           -0.7

                        NB: The numbers in the above table do not add up due to rounding.

                        Table 2.5
                        Migration profile for MYE series to 2000 (unrevised)
                        Migration               Mid-92          Mid-93     Mid-94    Mid-95   Mid-96          Mid-97       Mid-98        Mid-99     Mid-00
                        Net Internal                                                                                           -2.5         -1.1       -1.6
                        Net International                                                                                      0.8          0.9        1.0
                        Total Net                  -2.3           -1.6       -1.0      -0.5       1.0            0.4           -1.6        -0.1       -0.7

                        Table 2.5 shows that migration was variable                     White circles show postcodes with a Hard to
                        for Tower Hamlets over the intercensal decade                   Count (HtC) index of 1 (the easiest areas to
                        but generally demonstrated a trend of outward                   enumerate), light green circles have a HtC index
                        migration. The general trend of outward                         of 2 and dark green circles have a HtC index
                        migration is a contrast to the overall population               of 3 (the hardest areas to enumerate). Local
                        change and in particular natural change,                        Authority boundaries are marked in blue and
                        shown in tables 2.3 and 2.4. This shows that the                the Estimation Area boundary in red.
                        population in this area has grown rapidly, and
                        this is mainly due to large natural change.           
                        One Number Census Quality Assurance
                        information                                                     Information provided by the Local Authority
                        The One Number Census Quality Assurance                         Tower Hamlets Borough Council have not
                        Information Pack for Tower Hamlets, published                   queried their population estimates.
                        in 2003, can be found in the link below, and
                        includes the following information:
                                                                                        3 Analysis
                            ·      Population Estimates
                                                                                        This section of the report covers the detailed
                            ·      Confidence Intervals                                 analysis carried out for Tower Hamlets. More
                            ·      Diagnostic Ranges                                    information on the overall approach to the LA
                                                                                        studies can be found at
                            ·      Dependency Ratios                                    downloads/theme_population/LAStudy_
                            ·      Age-Sex Profiles depicting the above                 FullReport.pdf
                                                                                        3.1 Enumeration
                            ·      Census Coverage Survey Maps including
                                                                                        When looking at possible ward level undercount,
                                   Hard to Count information
                                                                                        it is necessary to consider that one of the key
                        The map in the link below shows the location of                 assumptions underpinning the One Number
                        the CCS Postcodes within the Local Authority.                   Census (ONC) is that the undercount is

                                                                                                                                                         5                                                                      2004 Local Authority studies: Tower Hamlets

                        approximately randomly distributed - that is,              The map shows that the ward shaded darkest blue
                        the distribution of the undercount is mainly               (Bow East) is the one with the largest difference
                        determined by the age and sex profile of an                between Census and Council Tax figures. The map
                        area and the distribution of hard to count areas           also shows that this ward contains a number of
                        as measured by the hard to count index. The                CCS postcodes. If the CCS has worked as it should,
                        assumption is that there are no other factors that         undercount in this area will have been identified.
                        have a large influence, such as the quality of the         Spitalfields and Banglatown ward has the second
                        work carried out by the census enumerators. If             largest proportional difference but this is not
                        this assumption is not true the resulting ONC              covered by any CCS postcodes so the undercount
                        estimates of population may not adequately reflect         may not have been detected in this area. St
                        the true undercount. The Census Coverage Survey            Katherines & Wapping was the other ward that
                        (CCS), although a large sample nationally, is highly       had a large difference and this was covered by CCS
                        unlikely to be able to measure this additional             postcodes. This should indicate that undercount
                        source of variability at low levels, i.e. Enumeration      will have been identified. The performance of the
                        Districts or Wards. The Local Authority studies            CCS in the Bow East ward and St Katherines &
                        have made detailed comparisons with Council                Wapping is therefore investigated further in section
                        Tax data, and conducted in-depth analyses on               3.2 (c) - Localised undercount.
                        enumerator record books (ERBs). The findings of
                        these pieces of work are described below.                  The table below shows the differences between the
                                                                                   Census and Council Tax for each ward, ordered by
                        a) Analysis of Council Tax data                            percentage difference, so that the ward at the top
                        Tower Hamlets shows an overall difference between          has the largest difference. As shown in Figure 3.1,
                        Council Tax records and the Census results of 3,367        there are three wards that have large differences
                        dwellings, a 4.2 per cent difference. This is the 8th      between the Council Tax records and the Census
                        largest positive numerical difference, and the 7th         database, these are Bow East, St Katherine’s and
                        largest positive proportional difference for the 376       Wapping and Spitalfields and Banglatown.
                        LAs in England and Wales.

                        The map on page 7 highlights the differences
                        between Council Tax and Census by ward. CCS
                        postcodes are indicated by the red dots.

                        Table 3.1.1
                        Differences between Census and Council Tax dwelling counts by 2003 ward
                        2003 Ward                      Census Dwellings   Council Tax Dwellings         Difference     Percentage Difference
                        Bow East                                  4,259                   5,072                 813                   19.1%
                        Spitalfields and Banglatown               2,991                   3,315                 324                   10.9%
                        St Katherine’s and Wapping                5,554                  6,056                  502                    9.0%
                        East India and Lansbury                   4,475                   4,741                 266                    5.9%
                        Bow West                                  4,435                  4,698                  263                    5.9%
                        Millwall                                  6,160                   6,457                 297                    4.8%
                        St Dunstan`s and Stepney                  4,675                  4,890                  215                    4.6%
                        Limehouse                                 5,206                   5,426                 220                    4.2%
                        Shadwell                                  4,492                   4,670                 178                    4.0%
                        Blackwall and Cubitt Town                 5,662                   5,881                 219                    3.9%
                        Bethnal Green North                       4,697                  4,869                  172                    3.7%
                        Whitechapel                               4,505                  4,642                  137                    3.0%
                        Bromley-by-Bow                            4,197                  4,224                   27                    0.6%
                        Weavers                                   4,904                  4,904                    0                    0.0%
                        Mile End and Globe Town                   4,672                   4,636                  -36                  -0.8%
                        Mile End East                             4,268                   4,223                 -45                    -1.1%
                        Bethnal Green South                       5,080                  4,895                  -185                  -3.6%

                        NB: Due to extensive boundary changes between Census day and 2003, most wards on the above table will be
                        different to 2001 wards. The more relevant changes are as follows
                        · Bow East was created from parts of Park and Bow wards
                        · St Katherine’s and Wapping was formed from parts of St Katherine’s and Shadwell.
                        · Spitalfield’s and Banglatown was made from part of the existing ward Spitalfield’s.

                                                                                                                                          6                                                                      2004 Local Authority studies: Tower Hamlets

  Figure 3.1
  Map showing the percentage difference between Council Tax and Census for Tower Hamlets

                                                                         Bow East

                         Bethnal Green North               Bow West

                                          Mile End and Globe Town
                           Bethnal Green South
                                                                    Mile End East

  Spitalfields and Banglatown
                                           St Dunstan`s and Stepney Green
                                                                                                  East India and Lansbury


             St Katherine`s and Wapping

                                                                                               Blackwall and Cubitt Town

                 CCS Postcodes
                 2% - 4%

                 0% - <2%

                 >0% - 2%

                 >2% - 4%

                 >4% - 6%

                 >6% - 8%

                 >8% - 10%

                 >10% - 12%
                                                  Census figures higher in pink areas.
                 19% - 20%                       Council Tax figures higher in blue areas.

                                                                                                                                    7                                                                       2004 Local Authority studies: Tower Hamlets

                        Difference between Census and Council Tax occupied and       (and these are not all visitor households) which
                        unoccupied dwelling counts                                   can also be considered as a type of dummy form
                        Tower Hamlets shows an overall difference                    (an ‘empty’ household).
                        between occupied dwellings according to the
                        Council Tax records and the Census results of -              Many enumeration districts showed many more
                        79 dwellings, a -0.1 per cent difference. Amongst            household forms in the ERBs than in the ED
                        all English LAs, this is the 322nd largest positive          database.
                        numerical difference, and the equal 162nd
                        largest positive proportional difference. This               The Park ward (which later became Bow
                        contrasts with the Council Tax differences                   East and Bow West) and St Katherine’s ward
                        above, indicating that the difference is mainly in           contained a number of EDs with many more
                        unoccupied properties.                                       household forms than in the ED database.

                                                                                     Spitalfields (which now makes up part of
                        Table 3.1.2                                                  Spitalfields and Banglatown) did not show any
                        Differences between Census and Council                       large differences between the two sources for
                        Tax dwelling counts                                          household forms. However it did contain some
                                          Census     CTB1 Difference            %    EDs that showed differences in the number
                                                                        Difference   of non-response dummy forms that were
                        Occupied          78,530    78,459        -79        0.1%    completed.
                        Unoccupied/        2,251     6,321      4,070      180.8%
                                                                                     This would suggest that the enumeration was
                                                                                     poor in this area and that some households may
                                                                                     not have been correctly recorded. This may be
                        b) Analysis of Enumerator Record Books (ERBs)                a cause for concern if the CCS did not work as
                        Investigations were carried out for Wards                    intended in this area.
                        where a high discrepancy in either direction
                        had been identified between the Census results               c) Analysis of Council Tax based response rates by
                        and the Council Tax. This involved analysis of               2001 Ward
                        Enumerator Record Books (ERBs), investigating                Table 3.1.3 shows a comparison of Council
                        individual records to assess the quality of the              Tax Figures and households counted in the
                        enumeration in the area, to ensure that all                  Census, broken down by ward. An implied
                        addresses had been accounted for and establish               Census response rate has been calculated
                        whether the differences between the Census and               by taking the number of counted Census
                        Council Tax may be definitional (eg differences              households and dividing this by the number of
                        in recording of vacant properties). This analysis            assumed occupied Council Tax addresses (the
                        also allowed an estimate to be produced of the               total number of Council Tax addresses minus
                        number of dummy forms that should have been                  those counted by the Census as vacant, second
                        completed by Enumerators.                                    homes and visitor only households). In wards
                                                                                     where the Council Tax implied response rate is
                        Dummy forms are completed by census                          substantially lower than the ONC household
                        enumerators to account for census forms                      response rate for the LA, it is possible that
                        that either have not been returned (ie a non-                significant enumeration failings, for which the
                        response) or for which the enumerator has                    ONC may not have been able to make a robust
                        determined that they should not be returned                  adjustment, may have occurred. The table is
                        (ie a valid non-response such as a vacant                    ordered by the implied census response rate,
                        household). For 2001, the types of dummy form                with the wards with the lowest response rate at
                        were:                                                        the top of the table.
                            ·   Refusal
                                                                                     The ONC household response rate for Tower
                            ·   Absent Household                                     Hamlets is 76.1 per cent. Looking at table 3.1.3,
                            ·   No contact                                           the wards of Spitalfields and ‘Holy Trinity’
                                                                                     show the largest difference between the ONC
                            ·   Vacant
                                                                                     household response rate and the implied Census
                            ·   Second/Holiday Home                                  response rate, whilst the wards of Millwall and
                                                                                     East India also demonstrate large differences.
                        The first three of these can be considered as
                        some form of non-response. In addition, some
                        census forms are returned with no residents

                                                                                                                                          8                                                                    2004 Local Authority studies: Tower Hamlets

                        Table 3.1.3
                        Analysis of Council Tax based response rates by 2001 Ward
                        Ward Name          2001 Council        Census        Assumed          Counted       Absolute    Implied Census
                                          Tax addresses       Vacants,       occupied           Census     Difference    Response Rate
                                                            2nd homes      Council Tax      households
                                                           and visitor      addresses

                        Spitalfields              4,003            67            3,936           2,554          1,382              65%
                        Millwall                 10,355            118          10,237           6,837          3,400              67%
                        East India                3,016            77            2,939           1,968            971              67%
                        St. Katherine’s           7,865            318           7,547           5,261          2,286              70%
                        Grove                     2,704            17            2,687           1,875            812              70%
                        Bow                       4,356            57            4,299           3,082          1,217              72%
                        Park                      2,710            13            2,697           1,963            734              73%
                        St. Dunstan’s             3,719            68            3,651           2,658            993              73%
                        Blackwall                 3,602            384           3,218           2,349            869              73%
                        St. Peter’s               5,082            53            5,029           3,699          1,330              74%
                        Shadwell                  6,146            172           5,974           4,417          1,557              74%
                        Weavers                   4,904            195           4,709           3,538          1,171              75%
                        Redcoat                   2,656            23            2,633           1,984            649              75%
                        St. Mary’s                2,740            127           2,613           1,973            640              76%
                        Lansbury                  3,941            63            3,878           3,002            876              77%
                        St. James’                3,488            131           3,357           2,605            752              78%
                        Bromley                   4,358            33            4,325           3,457            868              80%
                        Limehouse                 3,900            59            3,841           3,071            770              80%
                        Holy Trinity              4,054            91            3,963           3,441            522              87%
                        Totals                   83,599          2,066         81,533           59,734         21,799              73%

                        NB: It should be noted that the above table shows a comparison with the wards as they were on Census day
                        2001. Since then ward boundaries have changed, the more relevant ones outlined below;
                        · Spitalfields has reduced in size and is known as Spitalfields and Banglatown, and part of the ward is now
                        known as Bethnal Green South
                        · Millwall ward boundary has completely changed its area is now split between new wards Milwall and Blackwall
                        and Cubitt Town
                        · East India has had an extension to it ward boundary encompassing some of what was known as Lansbury and
                        has been renamed as East India and Lansbury.

                        d) Enumeration Information                              used if it was judged the ONC process had
                        Census enumerators collected certain                    failed. This classification is useful for comparing
                        information on households where no contact              response rates across areas, as we would expect
                        was made. According to this data, there were            the response rates measured by the ONC to be
                        1,593 (2.07 per cent) vacant households and 295         similar for these LAs. Further information on the
                        (0.38 per cent) second homes in Tower Hamlets.          contingency and borrowing strength strategy can
                                                                                be found at
                        Census returns indicated that there were 160            pdfs/oncinfopaper.pdf , which includes the
                        (0.21 per cent) visitor only households, and 882        borrowing strength areas in Annex E.
                        (1.15 per cent) empty households (returned
                        household forms which have no usual residents,          The Borrowing Strength areas for Tower
                        or very little information on the form).                Hamlets and their associated ONC response
                                                                                rate figures are shown in Table 3.1.4 below. The
                        These figures are not extreme and are consistent        response rate for Tower Hamlets is generally
                        with expectations for the area.                         lower than response rate for its borrowing
                                                                                strength areas. It is also lower than the mean
                        e) Estimated response compared to other areas           response rate for the borrowing strength areas
                        The ONC contingency strategy used a                     but the high mean response rate is in part due
                        classification of ‘similar’ areas (known as             to the response rate for Greenwich LA which is
                        borrowing strength areas) which were to be              significantly higher than for other areas. Tower

                                                                                                                                     9                                                                     2004 Local Authority studies: Tower Hamlets

                        Hamlets’ response rate is comparable with those            would allow differential undercount to be
                        for the other Borrowing Strength areas.                    quantified. Because under-enumeration is
                                                                                   disproportionately distributed across areas, the
                                                                                   CCS was stratified according to a Hard to Count
                        Table 3.1.4                                                (HtC) index, constructed from the following
                        ONC Response rates for similar (borrowing                  1991 Census variables which are associated with
                        strength) areas                                            undercount:
                        Area                                      ONC Response        ·   Multi -occupancy

                        Tower Hamlets                                    75.6%        ·   Unemployment
                        Borrowing Strength Areas
                                                                                      ·   Country of birth (which is associated
                        Hackney                                          72.1%            with language difficulty)
                        Newham                                           79.7%
                                                                                      ·   Private rented accommodation
                        Southwark                                        76.8%

                        Greenwich                                        85.6%        ·   Number of households imputed in 1991
                        Islington                                        77.9%     ONC imputation rates by key variables can be
                        Mean Response Rate for BS Areas                  78.4%     found at
                                                                                   imputation_rates_by_variable.asp, and
                        The ONC response rates for the other local                 confirm that the above variables were associated
                        authorities within the same Census Estimation              with undercount in the 2001 Census.
                        Area (EA) as Tower Hamlets are in the table
                        below. It shows that the response rate for Tower           The index categorises Enumeration Districts
                        Hamlets is broadly consistent with the other LAs           (EDs) into 3 groups representing the easiest 40
                        in the EA.                                                 per cent, the next 40 per cent and the hardest
                                                                                   20 per cent of EDs nationally. The sample was
                                                                                   then selected separately within each of these
                        Table 3.1.5                                                strata. This meant that the CCS sample had good
                        Response rates for all LADs in the Estima-                 coverage of areas with each HtC index value,
                        tion Area                                                  based on 1991 Census data, which was the only
                        Local Authority                       ONC Response Rate    information available at the time of the 2001
                        Tower Hamlets                                      75.6%
                        Hackney                                            72.1%
                                                                                   a) Missing sample
                        Islington                                          77.9%
                                                                                   As part of the ONC process, investigations
                                                                                   were carried out into sampled postcodes, with
                                                                                   high levels of CCS only, or Census only counts.
                        f) Qualitative information obtained about the
                                                                                   If it were found that the CCS or Census was
                        Census enumeration
                                                                                   enumerated so poorly as to be out of scope of
                        The analysis of fieldwork intelligence indicates
                                                                                   the ONC (ie the ONC would not be able to
                        that there were some problems.
                                                                                   compensate for the undercount), these postcodes
                            ·     The percentage of direct returns,                were removed from the sample.
                                  unplanned Census forms and calls made
                                  to the helpline fell outside the acceptable      There were no missing postcodes and no area
                                  range.                                           was excluded from estimation.
                            ·     Many enumerators were doubling up
                                                                                   b) Sample balance
                                  or had large workloads. This meant
                                                                                   This analysis assessed whether the selected
                                  that fewer visits could be made by the
                                                                                   sample was adequately balanced across
                                  enumerators to check progress.
                                                                                   indicators associated with undercount (both
                                                                                   across the Estimation Area and also the
                        3.2 Census Coverage Survey                                 constituent Local Authorities), and also whether
                        A key element of the One Number Census                     the CCS was successful in achieving high
                        was the Census Coverage Survey (CCS) - a                   response rates, and therefore measuring the
                        post enumeration survey that was designed                  undercount in the Census.
                        to measure undercount in the Census. The
                        aim was to survey a representative sample                  The analyses used 2001 Census data, in an
                        of postcodes across the country, which                     attempt to examine the balance of the sample
                                                                                   across a number of variables.
                                                                                                                                  10                                                                       2004 Local Authority studies: Tower Hamlets

                        Dummy form distribution                                     The map on page 12 shows the spread of
                        As mentioned in section 3.1 (b), dummy forms                dummy non-response forms by ward, with CCS
                        are completed by census enumerators to account              postcodes indicated by red dots. The areas with
                        for census forms that either have not been                  the largest percentage of non-response dummy
                        returned (ie a non-response) or for which the               forms (ie the darkest wards on the map) tend to
                        enumerator has determined that they should not              contain a number of CCS postcodes. The fact
                        be returned (ie a valid non-response such as a              that the CCS visited these areas indicates that
                        vacant household).                                          areas of undercount in these areas are likely to
                                                                                    have been identified. One exception to this was
                        The refusal, absent, no contact and empty                   Blackwall ward that was not covered by any CCS
                        households (which can be thought of as non-                 postcodes and undercount may not have been
                        response dummy forms) provide an indication                 detected.
                        of response rates across an area, and can be
                        used to assess whether there is any significant             Table 3.2.1 below shows the proportion of
                        imbalance within the CCS sample selection by                Census dummy forms across the Estimation
                        comparing the proportions of these dummy                    Area and within the CCS sample for each Hard
                        forms between the sample and the non-sampled                to Count stratum. The difference between
                        areas. This would show whether the CCS is                   proportions is not high so does not give cause
                        likely to provide a sample that underestimates or           for concern.
                        overestimates undercount.
                                                                                    It should be noted that hard to count strata 1
                        The proportions of census households that were              and 2 were collapsed for this Estimation Area,
                        classified as refusals, no-contact, absent or empty         due to the low sample size of hard to count 1.
                        households are examined. The data used are the              See section 3.3 (b) for details.
                        unadjusted census records - that is prior to the
                        ONC adjustments. The CCS sampled postcodes                  Table 3.2.2 below shows the proportion of
                        within an Estimation Area (EA) are compared                 dummy forms for each Hard to Count stratum
                        with the whole of the EA, and the same analysis             of each Local Authority in the Estimation Area,
                        done at LA level. This is also carried out within           both across the whole stratum and within the
                        the Hard to Count (HtC) strata, since we would              CCS sample. Again, the differences between
                        expect the proportions of dummy forms to be                 proportions are not relatively large.
                        different across these strata. It is valuable to
                        look at proportions of dummy forms at both LA
                        and the EA level because the ONC estimation
                        system produced estimates first at the EA level,
                        then apportioned these estimates out to the LAs
                        within that EA.

                        Table 3.2.1
                        Proportion of dummy forms across the Estimation Area and within the CCS sample for
                        each Hard to Count stratum
                        Estimation Area           Hard to Count   Count of dummy        Proportion of            Proportion of      Difference
                                                  Index stratum      forms in CCS      dummy forms               dummy forms          between
                                                                                      across all areas                  in CCS     proportions
                        Islington, Hackney                    2               135               21.4%                   16.8%            2.2%
                        and Tower Hamlets
                                                              3               643               21.2%                   21.2%            0.1%

                        Table 3.2.2
                        Analysis of proportions of dummy forms across the Local Authorities for each Hard to
                        Count stratum
                        Local Authority      Hard to Count      Count of        Count of      Proportion of        Proportion of    Difference
                        District             Index stratum   dummy forms    households in    dummy forms           dummy forms        between
                                                                  in CCS             CCS    across all areas              in CCS   proportions
                        Tower Hamlets                    2            75              464                20.1%             16.2%         3.9%
                        Tower Hamlets                    3           143              740                20.3%             19.3%         1.0%
                        Islington                        2            60              339                16.7%             17.7%         -1.0%
                        Islington                        3           227            1,119                18.2%             20.3%         -2.1%
                        Hackney                          2           633            2,876                22.0%             22.0%         0.0%
                        Hackney                          3           273            1,168                24.8%             23.4%         1.4%

                                                                                                                                           11                                                                    2004 Local Authority studies: Tower Hamlets

      Figure 3.2.1
      Map to show the percentage of non-response dummy forms for Tower Hamlets


                                            St. James'                             Bow

                              St. Peter's

              Weavers                          Holy Trinity                              Bromley

               Spitalfields                                                  Limehouse
                                                          St. Dunstan's
                               St. Mary's      Redcoat                                   Lansbury         East India


                 St. Katherine's                                                                      Blackwall

                  CCS Postcodes
                  9% - 10%

                  >10% - 13%

                  >13% - 16%

                  >16% - 19%

                  >19% - 22%

                  >22% - 25%

                  >25% - 28%

                                                                                                                                 12                                                             2004 Local Authority studies: Tower Hamlets

                        To explore this further, the distribution of the   The assumption underpinning the analysis is
                        proportions of dummy forms are shown below         that the recalculated score is highly correlated
                        in figures 3.2.2 to 3.2.3 on pages 14 – 16, with   with the real undercount - that is, that these
                        the proportion for Tower Hamlets as a whole        variables are associated with undercount in
                        shown in figure 3.2.2 and broken down by hard      2001. From the patterns observed in the CCS,
                        to count stratum for figures 3.2.3 (b) and (c)     there is evidence to suggest that this assumption
                                                                           is reasonable. The score is calculated as the sum
                        Figure 3.2.2 shows that there is some CCS          of the proportions of:
                        sample imbalance. In the whole EA there are
                                                                              ·   Unemployed persons
                        a number of EDs that have a dummy form
                        response proportion of 30 per cent or higher. In      ·   Persons whose country of birth was a
                        the CCS sample there are very few EDs that have           non-English speaking nation
                        dummy form proportion above 30 per cent.
                                                                              ·   Privately rented households
                        Figure 3.2.3 (b) shows that there is some             ·   Dummy form ‘non-response’ households
                        CCS sample imbalance in the HtC 2 stratum.            ·   Multi-occupied dwellings
                        The CCS sample only contains Enumeration
                        Districts that have a dummy form proportion        The derived score is calculated for all postcodes
                        of 30 per cent or less, whilst the whole LA has    across the Estimation Area, and comparisons can
                        some EDs that have dummy form response             be drawn between the score distributions for the
                        proportions of 60 per cent. However, the           Estimation Area/Local Authority and the CCS
                        population of this stratum is small.               sampled postcodes.

                        Figure 3.2.3 (c) shows that again there is some    Section 3.3 (d) Stratification, also looks at the
                        CCS sample imbalance. There are no EDs             recalculated hard to count scores but provides
                        in the CCS sample that have a dummy form           a general analysis on how the area has changed
                        response proportion of above 30 per cent,          in terms of hard to count between 1991 and
                        whilst in the LA there are some EDs that have      2001. This section focuses on comparing the
                        dummy form response proportions of over 60         proportions of hard to count areas in the CCS
                        per cent. However, this apparent imbalance may     sample and the rest of the EA or LA
                        not indicate an underestimation risk if there is
                        sample in other LAs that represent these harder    Figure 3.2.4 on page 17shows the recalculated
                        to count areas.                                    Hard to Count score distributions for the
                                                                           whole EA and for CCS sample postcodes. The
                        Figure 3.2.3 (a) - Distributions of the            distributions are reasonably similar, indicating
                        proportions of dummy forms in each 2001            that the CCS sample was balanced across the EA
                        Enumeration District for the HtC 1 stratum         as a whole.

                        There is no HtC 1 stratum as this was collapsed    Figures 3.2.5 (b) and (c) on page 18 show the
                        with HtC 2.                                        recalculated Hard to Count score distributions
                                                                           for the population and sample across the whole
                        Recalculated Hard to Count score                   Estimation Area by 2001 hard to count groups.
                        The Hard to Count score was derived using a
                        number of 1991 Census variables which were         Figure 3.2.5 (b) shows that the distributions
                        associated with undercount. The score was used     are not very similar for the Hard to Count 2
                        to determine the level of the Hard to Count        stratum, caused by the small population and
                        index for each 1991 Enumeration District in        sample sizes. There are large spikes in the CCS
                        England and Wales. The index was then used         sample distribution that suggests the CCS has
                        within the CCS sampling strategy as the primary    over-targeted some areas with the same HtC
                        stratifier within each Estimation Area, and as a   score.
                        stratum for estimation.
                                                                           Figure 3.2.5 (c) shows that the distributions are
                        As the data used to derive the HtC index was       not identical for the Hard to Count stratum
                        from the 1991 Census, there was a risk of the      3. The CCS sample covers most of the range
                        sample being unbalanced with respect to the        that the whole EA covers. The CCS sample has
                        ‘real’ hard to count information. We can assess    a higher percentage of EDs with higher scores
                        this by using a new hard to count score, derived   when compared to the whole EA.
                        from the 2001 Census data, at postcode level.

     Figure 3.2.2
     Distributions of the proportions of dummy forms in each 2001 Enumeration District

                                                                                         2004 Local Authority studies: Tower Hamlets
     Figure 3.2.3 (b)
     Distributions of the proportions of dummy forms in each 2001 Enumeration District for the HtC 2 stratum

                                                                                                               2004 Local Authority studies: Tower Hamlets
     Figure 3.2.3 (c)
     Distributions of the proportions of dummy forms in each 2001 Enumeration District for the HtC 3 stratum

                                                                                                               2004 Local Authority studies: Tower Hamlets                                                            2004 Local Authority studies: Tower Hamlets

                        Figure 3.2.5 (a) is not relevant, as there is no   The following table shows the level of
                        Hard to Count group 1 stratum in Tower             undercount identified by the CCS. The One
                        Hamlets.                                           Number Census was based on dual system
                                                                           estimation, which combines the numbers of
                        Figure 3.2.6 on page 19 shows the recalculated     people enumerated by the Census and/or CCS
                        Hard to Count score distributions for Tower        and estimates those people missed by both. The
                        Hamlets Local Authority and for the CCS sample     DSE (dual system estimate) column shows this
                        areas within Tower Hamlets. The distribution       figure for each ward. A measurement of the
                        has a number of spikes but does cover the range    Census undercount implied by the CCS can
                        of scores that are shown in the whole LA.          therefore be obtained by dividing the DSE by
                                                                           the number of people counted in the Census,
                        c) Localised undercount and the CCS                and using this figure as a multiplier. The final
                        Potential areas of localised undercount were       column in the table shows this figure. If the
                        identified in a number of ways including           CCS had found a level of undercount that was
                        comparing the ONC household response rate          similar to that implied by the Council Tax data,
                        with the response rate implied by Council Tax      then there is evidence to suggest that the ONC
                        (see section 3.1 (c), table 3.1.3). Areas with     has made a robust adjustment. If, however, the
                        high proportions of dummy forms were also          CCS had no sample in these wards or it had not
                        investigated. Where the CCS had visited these      found a reasonable level of undercount, then the
                        wards, it was investigated to check the level      ONC may not have been able to make a robust
                        of undercount that had been observed by the        adjustment.
                        CCS, and whether this looked plausible when
                        compared to the levels implied by the Council      In Tower Hamlets, Bow East ward had the largest
                        Tax.                                               difference between Census and Council Tax
                                                                           figures. Ward boundaries have changed since
                                                                           Census day 2001, and the ward of Bow East in
                                                                           2003 is made up of parts of several 2001 wards

         Figure 3.2.4
         Distributions of the recalculated Hard to Count score for the Estimation Area

                                                                                                                         17                                                          2004 Local Authority studies: Tower Hamlets

                   Figure 3.2.5 (b)
                   Distribution of the recalculated Hard to Count score for the Estimation Area by Hard to Count group 2

                   Figure 3.2.5 (c)
                   Distribution of the recalculated Hard to Count score for the Estimation Area by Hard to Count group 3

                                                                                                                       18                                                                  2004 Local Authority studies: Tower Hamlets

         Figure 3.2.6
         Distributions of the recalculated Hard to Count score for the Local Authority

                        Table 3.2.3
                        Census Coverage Survey implied undercount by 2001 ward
                        Ward name        No. CCS postcodes   Counted in Census              DSE    DSE/Counted in   Census undercount
                        Bow                              4                  68              84.2            1.238               19.2%
                        St Katherine’s                   2                   9              29.3            3.256               69.3%
                        Shadwell                         3                  69              80.9            1.172               14.7%

                        Ward name        No. CCS postcodes   Counted in Census              DSE    DSE/Counted in   Census undercount
                        Bow                              4                 169             219.8            1.300               23.1%
                        St Katherine’s                   2                  39             104.0            2.667               62.5%
                        Shadwell                         3                 113             128.1            1.134               11.8%

                        (Bow and Park). St Katherines and Wapping                The CCS has found a high level of undercount
                        also had large differences between the Council           in Bow East ward, with a mean undercount of
                        Tax figures and Census figures. This comprised           19.2 per cent for households and 23.1 per cent
                        of two 2001 wards, St Katherines and Shadwell.           for persons across the wards from parts of which
                        The CCS has therefore been investigated in each          it was formed. This is a hard to count 2 or 3
                        of these wards. There was no CCS sample in the           area, and so this level of undercount would be
                        Park ward.                                               expected. The CCS has also found extremely
                                                                                 high levels of undercount in the St Katherines
                                                                                 ward. Overall, the difference between the ONC

                                                                                                                                  19                                                                       2004 Local Authority studies: Tower Hamlets

                        estimate and the Council Tax total for this ward          a) Response rates
                        suggests that the ONC adjustment was robust.              The estimated person level response rate for
                                                                                  Tower Hamlets is 75.6 per cent.
                        d) CCS fieldwork and response rates
                        No problems were reported that were specific              Across the Local Authority, dummy form
                        to the CCS enumeration, many of the same                  analysis suggested a household response rate
                        problems faced by Census fieldstaff (Section 3.1          of 79.7 per cent, whereas the ONC estimated
                        (f)) were common to CCS fieldstaff.                       a response rate of 76.1 per cent (ranked 16th
                                                                                  across all LAs in terms of the difference). There
                        Table 3.2.4 below shows the response rates,               were 165 unfilled dummy forms after the
                        matching outcomes and sum of the dual system              imputation process. 3,710 households were
                        estimates by Hard to Count group. There were              imputed into random postcodes.
                        large CCS only and Census only counts and this
                        has led to large estimates of persons missed in           Table 3.3.1 below shows the estimated response
                        both, through the ONC process. The CCS has                rates by Hard to Count group for this Local
                        performed in line with expectations.                      Authority.

                        3.3 One Number Census process
                        The One Number Census project (ONC) was                   Table 3.3.1
                        set up by ONS for the 2001 Census to address              Estimated Response rates by Hard to Count
                        the fact that it is inevitable that some people           group for Tower Hamlets.
                        and households will not be counted in any                 Hard to Count                 Household        Person Response
                        population census. By conducting a large                  group                      Response rate                  rate
                        post-enumeration survey (the Census Coverage              2                                 79.3%                  80.6%
                        Survey, CCS) and combining the results of both            3                                 74.4%                  74.6%
                        the Census and CCS in what is known as a dual
                        system approach, the aim of the project was
                        to estimate and adjust the Census database for            b) Collapsing in the ONC process
                        undercount, and to ensure that robust results             For the purposes of One Number Census (ONC)
                        could be obtained for each local authority area.          estimation the population was divided into 37
                                                                                  age-sex groups. In addition, each postcode was
                        Detailed information on the One Number                    classified into one of three Hard to Count (HtC)
                        Census can be found in the following links:               levels, 1 being the easiest and 3 the hardest. This
                                                                                  means that there were 111 separate estimation
                        A Guide to the One Number Census:
                                                                                  strata in any given Estimation Area (EA).
                                                                                  However, in some cases it was not possible to
                                                                                  produce good quality estimates of these groups
                                                                                  separately. In these cases groups were combined,
                        One Number Census methodology and Quality
                                                                                  referred to as “collapsing strata”. In Tower
                        Assurance process report:
                                                                                  Hamlets, Hard to Count strata 1 was collapsed
                                                                                  with Hard to Count strata 2. More details of how
                                                                                  this was applied for Tower Hamlets can be found
                        Analyses into how well the ONC process worked
                        for Tower Hamlets were carried out as detailed            collapsing_strata.pdf
                        in the sections below.

                        Table 3.2.4
                        Response rates, matching outcomes and sum of the missed person estimates by Hard to
                        Count group for the CCS sample areas in the Local Authority.
                        Local Authority   Hard to Count     CCS person    CCS only count       Census only    Matched count          Estimate of
                                                  group   response rate                             count                         persons missed
                                                                                                                                          in both
                        Tower Hamlets                 2          84.3%                188              148               838                 43.3
                        Tower Hamlets                 3          84.9%                456              238             1,317                 76.1
                        Hackney                       2               -                0                 0                   0                0.0
                        Hackney                       3          71.2%                578              615             1,485                219.6
                        Islington                     2          71.1%                86               151               384                 39.8
                        Islington                     3          73.7%                481              512             1,408                162.4

                                                                                                                                              20                                                              2004 Local Authority studies: Tower Hamlets

                        c) Outliers                                          the analyses in section 3.1 showed that there
                        Each postcode in the CCS is used to represent        were not significant pockets of localised
                        undercount in a number of postcodes in the           poor response. This lessens the risk of
                        population. When an unusual observation              underestimation due to the presence of a large
                        occurs in a postcode it is likely that it is not     numbers of outliers.
                        representative of other postcodes. In order that
                        the unusual postcode (referred to as an outlier)     d) Stratification
                        does not have an undue influence on the rest of      The ONC estimation process used pre-defined
                        the population, it is removed from the sample.       stratification. These were the Hard to Count
                                                                             (HtC) index and 37 five-year age-sex groups.
                        Within the ONC strategy, outliers were               The HtC index was based on 1991 Census data,
                        identified using pre-defined ‘ratio’ limits for      and was used to draw the sample and to form
                        individual observations. The ratio was defined       estimation groups. Whilst this was the best
                        by the ratio of the dual system estimate to the      stratification that could be used at the time,
                        census count for each age-sex observation in         there may be areas where the Hard to Count
                        each postcode. These limits were different for       index gave a poor stratification. Analyses have
                        each Hard to Count stratum. The limits were          been undertaken to examine the change between
                        a ratio of 3 for the Hard to Count 1 stratum, a      the 1991 HtC distribution and that implied by
                        ratio of 4 for the Hard to Count 2 stratum and a     the 2001 Census data.
                        ratio of 5 for the Hard to Count 3 stratum. If an
                        observation was classified as an outlier through     Figure 3.3.1 on page 22 shows how the HtC score
                        this method, it was not used in the calculation      distribution for Enumeration Districts (EDs)
                        of any model parameters. For example, the CCS        has changed from the 1991 Census to the 2001
                        finds 6 people that the Census missed, and both      Census for the whole Estimation Area. A score
                        the Census and CCS find 1 person. The Census         of greater than 0.43 on figure 3.3.1 indicates
                        will show 1 person, and the dual system estimate     that the ED would be a HtC 3 area using 2001
                        is 7 persons. The ratio for this observation is      data. It can be seen that the Estimation Area was
                        then 7, which is classified as an outlier. This      harder to count in 2001 than it was in 1991, with
                        methodology was based on the data from the           more EDs in HtC 3.
                        simulation studies, which assumed an overall 95
                        per cent Census coverage.                            Note that this analysis looks at the overall change
                                                                             between 1991 and 2001 in the distribution
                        The Estimation Area had 30 outliers, ranked          of hard to count areas across the Estimation
                        equal 2nd across the 101 Estimation Areas. The       Area and LA as a whole. Part of section 3.2 (b)
                        outliers within this Estimation Area contained       also looked at the recalculated hard to count
                        8.8 per cent of the persons found in the CCS         distribution, but in terms of the CCS sample
                        that were missed by the Census, ranked equal         compared with the rest of the EA and LA.
                        3rd amongst all Estimation areas. This is a high
                        figure, which may have affected the estimates        Figure 3.3.2 on page 22 shows how the HtC score
                        in this area. There were 10 outliers in Tower        distribution has changed from the 1991 Census
                        Hamlets, ranked equal 5th across all 376 Local       to the 2001 Census for Tower Hamlets. As with
                        Authorities.                                         the EA graph, this shows that Tower Hamlets
                                                                             is harder to count in 2001 - EDs scoring above
                        Because this Estimation Area had a large             0.43 would fall into HtC stratum 3, and the 2001
                        number of outliers, there was an increased           score shows a higher proportion of EDs above
                        risk of underestimating the population. This         this level, and many less EDs below this level.
                        was because in areas where response in the
                        census was poorest, the ‘fixed cut off ’ outlier     Figure 3.3.3 on page 23 shows how individual
                        strategy tended to identify more outliers than       EDs have changed from the 1991 score to the
                        it was designed to. It was therefore a possibility   2001 score for the Local Authority District.
                        that some of these observations were valid           The graph shows that the area has been subject
                        representations of the level of undercount in        to much change in both directions, with a
                        non-sampled areas. To determine whether              clear trend that HtC scores increased over the
                        this was the case, the enumeration data were         10-year period. This may indicate a risk of
                        examined for evidence of localised pockets of        underenumeration, although the area is still
                        non-response - which would be represented            mostly HtC 3.
                        by outliers for several age-sex groups within
                        the same CCS postcode. In Tower Hamlets,

                                                                                                                              21                                                     2004 Local Authority studies: Tower Hamlets

                  Figure 3.3.1
                  Distribution of 1991 and 2001 Hard to Count scores for EDs within the Estimation Area

                 Figure 3.3.2
                 Distribution of 1991 and 2001 Hard to Count scores for Tower Hamlets

     Figure 3.3.3
     Scatter plot of 1991 and 2001 Hard to Count scores for each Enumeration District in Tower Hamlets

                                                                                                         2004 Local Authority studies: Tower Hamlets                                                                2004 Local Authority studies: Tower Hamlets

                        e) Household and person imputation results               2) There was a high level of dependence for
                        Detailed information on imputation can be                   missed households, which results in an
                        found at                  underestimate of total households and
                        pdfs/sc9908.pdf                                             total people. Because the under-estimate
                                                                                    of people will be within wholly missed
                        The ONC imputation process imputed people                   households, the CCS pattern will show an
                        as part of wholly imputed households and also               overestimate.
                        into counted households. There was no explicit
                                                                              In general, of these two scenarios, the second
                        restriction on how many people were imputed
                                                                              is the most likely to have occurred. However,
                        into these two categories, the only restriction
                                                                              a firm conclusion could only be reached if
                        was on the total number of people imputed
                                                                              supported by other external evidence, of
                        into a Local Authority. Since the imputation
                                                                              dependence or of over-estimation of households.
                        of households (rather than people) was carried
                        out as the first step, any people remaining to
                                                                              Within this Estimation Area, the imputation
                        be imputed were then placed into counted
                                                                              process imputed 18.9 per cent of the imputed
                        households. Therefore, the patterns observed
                                                                              people into counted households. The CCS
                        in the CCS may be very different to those
                                                                              measured this proportion to be 16.5 per cent.
                        created by the imputation methodology. There
                                                                              This was the 23rd largest difference in this
                        are a number of factors that will contribute
                                                                              direction across all 101 Estimation Areas.
                        to any extreme differences, mainly relating
                        to households and people being estimated
                                                                              At LA level, the imputation process imputed 13.3
                        separately. This means that any under or over-
                                                                              per cent of the imputed people into counted
                        estimation of either will affect the proportions of
                                                                              households. The CCS measured this proportion
                        households or people that are imputed.
                                                                              to be 17.7 per cent. This was the 203rd largest
                                                                              difference in this direction across all 376 Local
                        For areas where the imputed proportion is
                                                                              Authority Areas.
                        significantly higher than that measured by the
                        CCS there are three interpretations:
                                                                              There is not a large difference between the
                           1) There was an underestimate of total             imputation process and the CCS with Tower
                              households, therefore more imputed              Hamlets’ figures (ranked 203th of 376) so there
                              people went into counted households.            is insufficient evidence of a problem here.
                           2) There was a high level of dependence
                              between Census and CCS for people
                                                                              4 Population definitions
                              within counted households, and so the
                              CCS pattern shows an underestimate.             The 2001 Census was conducted on a usual
                              More information on dependency can              residence base, that is, people were asked to
                              be found at              fill in details on a form at their place of usual
                              census2001/pdfs/dependency_paper.pdf            residence.

                           3) There was an overestimate of total people,      It is likely that areas which have high numbers
                              therefore the remainder were placed into        of mobile people where usual address is not easy
                              counted households.                             to define will be difficult to count. People with
                        Of these three scenarios, generally the first two     second homes, students who live at different
                        are the most likely to have occurred. However,        addresses during term time and holidays fit this
                        a firm conclusion could only be reached if            category, as do many members of the armed
                        supported by other external evidence, of              forces who are often moved from base to base.
                        dependence, under-estimation of households or         Quantifying the numbers of people that are
                        over-estimation of people. Conversely, for areas      wrongly missed off forms for this reason is
                        where the imputed proportion is significantly         very difficult, due to the lack of evidence that is
                        lower than that measured by the CCS, there are        available - the 2001 Census was conducted solely
                        two interpretations:                                  on a usual residence base, with little information
                                                                              on visitors collected, therefore comparisons with
                           1) There was an overestimate of total              figures collected on a different base cannot be
                              households, therefore too many imputed          made.
                              people went into wholly imputed

                                                                                                                                  24                                                                   2004 Local Authority studies: Tower Hamlets

                        Students                                                 non residential, demolished, derelict or ‘Late
                        A quality assurance of student estimates was             returns’ (i.e. returned after the processing cut off
                        carried out as part of the ONC process. More             date)) these were excluded from the analysis. If,
                        information can be found in the following links:         however, the absence could not be explained the
                                                                                 details of the individual records were recorded
                                 and summary reports were produced for each
                        students.pdf                                             Estimation Area (EA), recording the number of
                                                                                 forms missed.
                        key_findings.pdf (under section 4.2)                     The main reasons for forms not being processed
                        The number of students that live in Tower                   ·   Forms were not received from the field or
                        Hamlets during vacation time but were living                    received too late
                        away in term-time was 755. This is ranked 371st
                        among all LAs, so is not particularly high.                 ·   Forms lost in the system during system
                        An analysis on student numbers based on                     ·   Some forms were misrecognised and
                        comparisons with census figures and council                     therefore delivered as part of a different
                        tax records has been conducted. This analysis                   EA/LA
                        highlights areas that have a large difference
                        between these two figures.                               For the EA as a whole, there were 2,155
                                                                                 unprocessed forms, 1,431 of which were late
                        This analysis shows that Tower Hamlets (shown            forms. In Tower Hamlets LA there were 690
                        below) is ranked 25th highest of the 376 LAs in          unprocessed forms distributed over a number of
                        terms of the difference between student houses           EDs, which is 0.86 per cent of total households.
                        in council tax records and student properties            There were 2 EDs with 25 or more forms
                        counted in the Census. The second column                 in a block missing. These numbers are not
                        shows households which are occupied solely by            significant, and are within the acceptable level
                        students and therefore gaining an exemption              for ONC imputation.
                        from council tax. There will also be many
                        households in Tower Hamlets that contain
                        students living with non-students that will not          6 Other - communal establishments
                        appear in these figures.                                 This section contains information on communal
                                                                                 establishments (such as residential homes,
                                                                                 student halls).
                        Table 4.1
                        Comparison of students in the Census and                 The 2001 Census shows that Tower Hamlets
                        Council Tax records                                      has 1,986 persons living within 162 Communal
                                                                                 Establishments. There is some evidence of
                        Tower               Total    2001 Census    Difference
                        Hamlets        Exemption    - All Student                problems with the enumeration of Communal
                                          Class N     Properties
                                           (CTB1                                 Establishments in Tower Hamlets.
                                                                                 The ONC Quality Assurance process included
                                            1,304            832           472
                                                                                 an assessment of Communal Establishments, in
                                                                                 particular Halls of Residence. As a result Tower
                                                                                 Hamlets received an adjustment. For more
                        5 Processing
                                                                                 details see
                        As the data was processed, checks were carried           pdfs/onc_key_findings.pdf
                        out to ensure that the delivery of the data for
                        the Estimation Area data was complete. The
                        data was checked to ensure that the Census               7 Other -administrative sources
                        Household form numbers were in sequence
                                                                                 The ONC Quality Assurance (QA) process
                        with no unexpected gaps. The identities of the
                                                                                 involved comparisons with various
                        missed forms were then compared to other data
                                                                                 administrative data sources, as detailed in the
                        and information (Enumerator Record Books
                                                                                 individual QA information packs for each Local
                        (ERB), enumerator-completed summary forms
                        and the geography database). If Household
                        forms had been missed for a valid reason (i.e.

                                                                                                                                     25                                                                2004 Local Authority studies: Tower Hamlets

                        Further analysis of administrative sources was        Table 7.1 compares the 2001 MYEs and Census
                        completed as part of the LA Studies program.          counts, as appropriate, with each of the
                        The administrative sources used, both by the          administrative data sources and the previous
                        ONC QA processes and the LA Studies, are              (2000) MYE. This table also gives Tower
                        detailed below.                                       Hamlets’ ranking amongst local authorities in
                                                                              England and Wales, for each comparison, with
                        Council Tax - Council Tax dwelling counts             LAs ranked first having the largest difference
                        for 2001 were compared with the Census                between the two counts.
                        results for each Local Authority. More detailed
                        comparisons of Council Tax counts of occupied         The administrative sources show a mixed
                        and vacant dwellings with the Census results          picture. Some of the sources suggest that the
                        were also completed.                                  Census has underestimated the population,
                                                                              whilst other sources indicate that the Census
                        Electoral Roll - The revised 2001 Mid -Year           has overestimated the population. They also
                        Estimates (MYEs) for people aged 18 years and         vary when compared to the national averages,
                        over were compared with the Electoral Roll data.      some are lower than the national average, whilst
                                                                              others are higher than the national average. This
                        Patient Register - The revised 2001 MYEs have         is not necessarily a cause for concern given the
                        been compared with the NHS Patient Register           variability in the quality of the administrative
                        data (adjusted and unadjusted) for 2001.              sources, mainly due to definitional and timing
                                                                              differences. There is a large difference between
                        Pensions - The revised 2001 MYEs for people           the number of households on Council Tax and
                        aged 65 years and over have been compared             the Census, although when occupied counts are
                        with the year 2000 Department for Work and            compared as in Table 3.1.2, there is very little
                        Pensions (DWP) Pensions data.                         difference.

                        Child Benefit - The revised 2001 MYEs for
                        people aged 0 – 14 were compared with the 2001        8 Conclusions and recommendations
                        Child Benefits data.                                  There were 3,367 more dwellings recorded on
                                                                              Council Tax data than counted in the Census.
                        Schools Census - The revised 2001 MYEs                However, most of this difference is vacant
                        for people aged 5 – 14 years inclusive were           households, and there are 79 more occupied
                        compared with the 2001 Schools Census data.           households recorded in the Census. The CCS
                                                                              sampled in most of the wards that had the

                        Table 7.1
                        Comparisons of Revised 2001 MYE with administrative sources
                        Source compared       Ranking (/376)       Source     Revised 2001 MYE    Source as % of   National Average
                        to 2001 MYE                             Population                                  MYE

                        2000 MYE                        374         186,700            201,600            92.6%              101.3%
                        2001 Electoral Roll             356         139,291            151,400            92.0%              98.5%
                        data (18+ Only)

                        2000 Pensions data               22          18,631             18,500           100.7%              98.5%
                        (65+ Only)

                        2001 School Census              125          27,779             27,000           102.9%              100.6%
                        Data (5-14 Only)

                        2001 Child Benefit               67          42,200             42,800            98.6%              99.2%
                        Data (0-14 Only)

                        2001 Health                      78         213,411            201,600           105.9%              105.0%
                        Register Data

                        2001 Adjusted                    29         218,588            201,600           108.4%              104.4%
                        Patient Record

                        Source compared       Ranking (/376)        Source             Census     Source as % of   National Average
                        to Census Data                          Households         Households            Census

                        2003-04 Council Tax               2          86,666             80,800           107.3%              101.2%

                        NB: The Mid Year Estimates in this table have been rounded

                                                                                                                                26                                                        2004 Local Authority studies: Tower Hamlets

                        largest differences and there is evidence that the
                        undercount was picked up by the CCS.

                        There was a large change in the Hard to Count
                        scores for Tower Hamlets between 1991 and
                        2001, and there was also a large number of
                        outliers. However, these problems alone do not
                        give cause for concern, and are not backed up by
                        other evidence.

                        None of the other analyses carried out in this
                        report have highlighted significant risks, and
                        we have found no basis on which to make an
                        adjustment to the estimate of the population of
                        Tower Hamlets.


To top