 |
|
A
Guide to Construction Output Statistics |
|
Back
|
|
Quarterly
Inquiry of Construction Activity
Sampling
and collection
Imputation
and Grossing
Other
Adjustments
DLO
Enquiry
Sampling
and collection
Imputation
and Grossing
Estimate
of Unrecorded Output
Total
Construction Output
Deflation
and seasonal adjustment
Annex
A - Rotational Sampling |
Production
of estimates of construction output
Construction
output is a quarterly series of the output of the construction
industry, in both the private and public sectors. Press Notices are
published on the first Friday of March, June, September and
December. The actual series is a combination of results from 2
surveys, the Quarterly Inquiry of Construction Activity and the
Building and Civil Engineering Employment and Output Enquiry, plus
an estimate for unrecorded output.
The
Quarterly Inquiry of Construction Activity
The
Quarterly Inquiry of Construction Activity (CA) measures the output
(and employment) of private contractors, covering new construction,
improvements and alterations and repair and maintenance. Output
figures include:
The
value of building civil engineering and associated work, excluding
VAT done by the contractor’s directly employed staff which is
chargeable to customers
The
value of materials used, labour costs, overheads and profits;
The
value of work done on the contractors own initiative on buildings
such as dwellings for eventual sale or lease;
The
value of work done on demolition and site preparation;
The
value of work done by the contractor on the construction or
maintenance of its own premises;
The
value of articles made by the contractor and used in construction
work;
The
value of any materials supplied by the contractor free of charge to
subcontractors.
The
following are excluded:
The
value of work done overseas;
The
value of work done by sub-contractors of any type;
The
value of any payments made to labour-only subcontractors;
The
value of articles made for sale or materials sold;
The
value of materials supplied to the contractors free of charge by
firms in the industry;
The
value of land;
Any
architects or consultants fees.
Top
of page |
|
Information
for Great Britain is based on quarterly returns from a sample of
around 12,000 firms taken from the CISTATS universe of around
165,000 construction contractors. This universe is bigger than the
New Orders universe because it also includes firms working solely as
subcontractors, as well as those only carrying out repair and
maintenance. This universe is kept
up-to-date through quarterly exchanges of information with the
Inter-Departmental Business Register as well as information from
data suppliers.
Construction
firms on CISTATS are assigned to size groups (SGs) based on their
employment levels, which are used to produce a stratified
rotational sample, with a rotation period of 8 quarters. Rotational
sampling is explained in Appendix
A. The table below shows the size groups (or “strata”) and the sampling ratios used:
|
Size Group
|
Employment
|
Sampling Ratio
|
|
0
|
1
|
1 in 35
|
|
1
|
2-3
|
1 in 20
|
|
2
|
4-7
|
1 in 10
|
|
3
|
8-13
|
1 in 3
|
|
4
|
14-24
|
1 in 3
|
|
5
|
25-34
|
1 in 2
|
|
6
|
35-59
|
1 in 1
|
|
7
|
60-79
|
1 in 1
|
|
8
|
80-144
|
1 in 1
|
|
9
|
115-299
|
1 in 1
|
|
10
|
300-599
|
1 in 1
|
|
11
|
600-1199
|
1 in 1
|
|
12
|
1200+
|
1 in 1
|
Over
50% of construction firms are in Size Group 0, and a further 40% are
in Size Groups 1 and 2.
The
output information collected is broken down into New Work, Public
Housing Repair and Maintenance, Private Housing Repair and
Maintenance, Public Non-housing Repair and Maintenance and Private
Non-housing Repair and Maintenance. The employment information is
broken down into Working Proprietors, APTCs (administrative,
professional, technical and clerical employees) and Operatives
(manual workers).
Top
of page |
|
Results
Processing - Imputation and Grossing
CA
results are much more complex than New Orders, and take much longer
to run. The following are the key stages run by the statisticians
that are involved in turning the raw data supplied by contractors
into current price contractors output.
Update
survey – this copies the live database into a separate table
which is where the results are run. This can take up to 2 hours if
the system is busy.
Pre-imputation
and Imputation – CA firms which are in the sample but have not
responded have a value imputed for them. For each Size Group, the
system calculates an imputation link by taking the average movement
between the previous quarter and the current quarter for those firms
which have responded. This is then applied to those firms which
responded in the previous quarter but not in the current one. Where
a firm did not respond in the last quarter either an average figure
is imputed, which is the same for each affected firm in the Size
Group. During pre-imputation, the statistician examines the
imputation link for each question in each Size Group, and adjusts it
if necessary. This takes up to 3 hours. Once this has been done
CISTATS uses the imputation links to impute missing values. This is
known as imputation and is an automatic process and can take several
hours. It is usually run overnight.
Construct
– a small number of the largest firms have been marked as key
responders, and cannot be imputed for. For each of these firms that
has not responded, the statistician must construct a response. This
takes about 30 minutes depending on how many firms need to have
responses constructed.
Weights
– The first stage of grossing is the calculation of the
grossing weights, which will be used to multiply up the sample total
to the universe total. This is an automatic process. CA uses 3
different grossing weights: model, design and outlier.
The
model weight (g) adjusts the grossing according to known independent
properties of the companies sampled. The formula is:
|
g |
=
X/((N/n)* sum x) |
|
where |
|
X |
=
Size Group total for an independent stratification measure (in
this case, ONS Total employment for the Size Group) |
|
x |
=
ONS employment for individual companies in the sample |
|
N |
=
Population Size |
|
n |
=
sample size |
For
companies in Size Groups 6 and above, where there is 1 in 1
sampling, the model weight is 1 because X = x and N = n
The
design weight (a) is the main grossing weight. The formula is:
|
a |
=
N/(n(1+hd/(n-d))) |
|
where |
|
h |
= factor
for births and deaths (0 for big firms, 1 for small firms) |
|
d |
= number of
deaths (closedowns) recorded |
|
N |
=
population size |
|
n |
=
sample size |
Again
for Size Groups where there is 1 in 1 sampling (ie big firms) the
design weight = 1
The
outlier weight (w) ensures that responses which are deemed to be
outliers (see below) are not grossed as they are not representative.
If a response is not an outlier, then w = 1. The actual outlier
weight is rather complicated, but in simple terms it is 1/a (where a
is the design weight).
Outliers
– the next stage of grossing is to set outlier limits, that is
the cut-off point whereby everything above that level is deemed to
be an outlier. The statistician sets an outlier for each Size Group,
based on the output per person values for each return. For example,
if an outlier limit of £60,000 is set, then all returns where the
output per person is greater than this are deemed to be outliers,
and will be grossed accordingly.
Grossing
– grossing is now an automatic process, which is run overnight.
Each return for a firm in Size Groups 0-5 is grossed so that:
Grossed
value = a*g*w*ungrossed value
where
a, g and w are the grossing weights described above. Because
non-responders are imputed for, returns for firms in Size Groups 6
and above, where there is 1 in 1 sampling, are not grossed.
Top
of page |
|
Following
grossing, results tables showing various breakdowns of output and
employment can be run. Results table CO04 gives grossed output by
type of work and size group. Table CO29 shows employment by employee
type and size group. These tables are used as the basis for further
adjustments. At the moment, adjustments are made to remove the
effects of methodological improvements made during the development
of CISTATS, so that output information is consistent with
information produced under the old system. At some stage, we will
stop making these adjustments, and start publishing figures on the
new methodology. The adjustments made are:
Outlier
adjustment – CISTATS views outliers differently than the old
system, and as a result there are fewer outliers than there would
have on the old methodology. Returns which would have been outliers
on the old system are not outliers on the new system, and hence are
being grossed as normal returns. An adjustment is made to remove the
effect of the grossing.
Grossing
adjustment – CISTATS grosses in a more complex way than the
old system, which just used N/n (ie, population size divided by
sample size). An adjustment is made to mimic the effect of the old
grossing method.
Rotational
sampling – the old system used a random sampling technique.
With rotational sampling firms stay in the sample for a set period,
and hence are more likely to have returns in consecutive quarters.
This means that more firms use the imputation link rather than the
average imputation method (see above). An adjustment is made to
remove this effect.
Closedown
adjustment – on the old system an adjustment was made to allow
for closedown responses coming in slower than actual responses. On
the new system this adjustment is not really necessary because of
the change in the imputation method. However the adjustment is still
made for continuity.
New
Work/R&M switch – on CISTATS, the allocation between new
work and repair and maintenance appears to be skewed towards new
works. Ad hoc adjustments are made to reduce this effect.
Once
these adjustments have been made, the result is current price
recorded contractors output.
Top
of page
|
|
The
Building and Civil Engineering Employment and Output
Enquiry is usually known as DLO, for Direct Labour
Organisation. DLOs are public sector organisations which
employ their own construction workers rather than
contracting out construction work to the private sector.
In the past this included the public utilities such as
the water boards however now these have largely been
privatised it covers largely local authorities.
Top
of page
|
|
As
there are only about 320 DLOs, this quarterly Enquiry is a census,
using 1 in 1 sampling. The questions asked are the same as for the
CA Inquiry.
Top
of page
|
|
Results
Processing - Imputation and Grossing
DLO
results are processed in a similar way to CA, going
through first Update Survey, and then Pre-imputation and
Imputation. No grossing is necessary because of the 1 in
1 sampling.
No
adjustments are made to DLO. The result is current price
DLO output.
Top
of page
|
|
Estimate
of Unrecorded Output
The
Builders Address File, which CISTATS uses as the basis or universe
for all its contractor based surveys is based on the VAT register,
and hence only includes those firms over the VAT threshold. It is
clear from comparing the ONS surveys of employment with the figures
of construction employment collected in the CA and DLO surveys, that
DTI miss a large chunk of construction employment and thus a large
proportion of construction output.
In
order to compensate for this missing “unrecorded” output, we
make an estimate of it. This is done by taking the difference
between total construction employment (including the self-employed)
as measured in the Labour Force Survey, and total construction
employment reported to CISTATS. This gives us a measure of
unrecorded employment. This is then multiplied by the average output
per person for Size Groups 0-2, giving an estimate of the output
produced by the construction workers missed. This is current price
unrecorded contractors output.
Top
of page |
|
Total
construction output is the sum of recorded contractors output,
unrecorded contractors output and DLO output. However because of the
questions asked on the CA and DLO forms, we only have total new
work, rather than information about the various sectors. Similarly
we don’t have a regional breakdown of new work.
In
order to obtain this, we use a matrix of work, region, class and
duration information from New Orders surveys over the past three
years. Very few projects last
longer than three years so it is not necessary to go back further
than that.
Repair
and Maintenance is automatically available broken down in public and
private, housing and non-housing. A regional breakdown for
contractors R&M is calculated using the address of the
contractor as R&M is assumed to be largely carried out by local
firms. DLO R&M is broken down in the same way.
Top
of page |
|
Output
is deflated and seasonally adjusted in a similar way to New Orders. The price indices used are slightly
different, and are listed below:
| MATHO |
Materials
Index: Housing |
| MATNH |
Materials
Index: Non-Housing |
| MATRD |
Materials
Index: Roads |
| LABC |
Labour
Costs Index: Building |
| LABRD |
Labour
Costs Index: Roads |
| MATRM |
Materials
Index: Repair and Maintenance |
| PIPSH |
Price
index for Public Housebuilding |
| SDD |
Price
index for Scottish Public Housebuilding |
| RCPI |
Price
Index for Road Construction |
| PUBLD |
PSA
Index: Public Works |
| BCIS |
BCIS
Index: All |
| BCISPRVI |
BCIS
Index: Private Industrial |
| BCISPRVC |
BCIS
Index: Private Commercial |
|
|
The
constant price figures are then seasonally adjusted, again by sector,
using factors produced using standard seasonal adjustment software
called X11ARIMA. This process smoothes the series so that seasonal
effects such as weather conditions, holidays and the beginning and end
of the financial year do not affect comparisons across time. For
example, the level of Public Housing orders rises significantly during
February and March, the end of the public sector budget year, and if
this effect was not removed it would appear that Public Housing orders
were rising.
|
|
|
|
|
|
Updated
on 13/12/02
|
|
