Total

Male

Female

Age0-17

Age18-64

Age 65+

Cross sectional

·Over-coverage

·Under-coverage

·Misclassification

Over-coverage: 705 ineligible households were selected out of the total of 11,857 (5.9%) for 2011.

Under-coverage: No estimate available.

Misclassification: None.

Measurement error occurs at the time of data collection. While attempts are made to minimise measurement error, some will occur during data collection from these four main sources:

1. Questionnaire: The design of the questionnaire is particularly important as it can introduce error through ways such as poorly-worded, confusing questions or not allowing for certain legitimate answers. The design of the questionnaire is conducted very carefully in order to minimise any error. The details of this design are covered below.
2. Mode of data collection: For EU-SILC UK, the mode of data collection for the main stage of interviewing is through face-to-face interviewing. This mode of data collection is chosen because it is deemed to be the best mode for obtaining the most accurate data possible. However, this mode of data collection does allow for the interviewer to possibly influence the answers of the respondent. Steps taken to prevent this are outlined below.
3. Interviewer: It may be possible for an interviewer to systematically influence responses in one way or another. To prevent this, UK interviewers undergo rigorous training and monitoring, the details of which are covered below.
4. Respondent: The respondent error is minimised by careful design of the questionnaire (e.g., collecting detailed information about individual income sources in order to construct a total) and interviewer training. See below for details.

Where possible the UK uses questions that have been harmonised across the national social surveys, such as the Labour Force Survey and Household Budget Survey. This is a thorough process whereby representatives from the different surveys and topic experts agree the harmonised standards. Any changes are thoroughly tested via peer review and pilot testing. Cognitive methods are also applied where deemed necessary.

Interviewers who work on EU-SILC UK are recruited only after a rigorous and careful selection process after which they take part in an initial training course. On this initial training course, interviewers are trained to ensure that during the interviewing process they conduct themselves in a manner which will not systematically influence the answers of the respondents. New and existing interviewers attend survey briefing sessions on all aspects of the fieldwork. Furthermore new recruits are always supervised by being accompanied in the field by a field manager. All interviewers who continue to work on EU-SILC are observed regularly in their work and continuously provided with feedback.

% of households having received an amount

% of household with missing values (before imputation)

% of household with partial information (before imputation)

% of households having received an amount

% of household with missing values (before imputation)

% of household with partial information (before imputation)

Data collection is carried out by face-to-face interviewers using Computer Assisted Personal Interviewing (CAPI) on laptop computers. Blaise software (developed by Statistics Netherlands) is used, which is an integrated system for survey processing. The use of Blaise enables a reduction in processing errors as data can be “checked” as it is entered by interviewers. One example of this is that all income data is “checked” at the point of collection to make sure that net values are not greater than gross values for an individual. Another example relates to the receipt of child benefit payments, where if a respondent claims to receive such a benefit and enters an amount which differs to the known amount that someone can receive then this is highlighted for the interviewer to ensure the right answer has been recorded.

Data are converted from Blaise toSPSSand are edited using this software. At this stage the data are extensively checked for consistency and plausibility and relevant edits are performed. Very high and low values for income variables are checked and edited if deemed to be erroneous. Period codes are scrutinised to ensure that they are consistent with the amounts provided by respondents for income variables. Checks are performed to ensure that respondents do not have income variables filled which they should not on the basis of their answers to other relevant questions and any relevant edits are made. The editing process in SPSS deals with only a very small percentage of cases, which is due to in part to the competency of the interviewers and the checks built in to the questionnaire and the Blaise software.

All private households and their current members residing in the UK at the time of data collection. This is comparable with the EU-SILC definition.

In 2011, the definition for what constitutes a household was changed from:

 “one person living alone or a group of people (not necessarily related) living at the same address with common housekeeping – that is sharing either a living room or sitting room or at least one meal a day”

to:

 “one person living alone or a group of people (not necessarily related) living at the same address who share cooking facilities and share a living room or sitting room or dining area”

This reflected a change in the household definition used for the 2011 UK Census. The national definition differs slightly from the Eurostat definition. The UK definition does not state that a group of people living together in a dwelling must share expenditures to be classified as a single household.

A person is in general regarded as living at an address if he or she (or the informant) considers the address to be his or her main residence. There are, however, certain rules which take precedent over this criterion.

Children of any age away from the home in a temporary job and children under 16 at boarding school are always included in the parental household.

From 2008 students who are living in halls of residence are also included as residents of the household sampled even if they are not in situ at the time of the interview.  However, other children aged 16 or over who live away from home for the purposes of either work or study and come home only for holidays are not included at the parental address under any circumstances.

Anyone who has been away from the address continuously for six months or longer is excluded.

Anyone who has been living continuously at the address for six months or longer is included even if he or she has his or her main residence elsewhere.

Addresses used only as second homes are never counted as a main residence.

See section on income reference periods used.

EU-SILC UK, like other official income surveys in the UK, uses continuous interviewing with interviews spread evenly throughout the year. The survey measures current income. For example, with regards to income from earnings and benefits, respondents will provide figures which relate most commonly to the last week, two weeks or month. With earnings in particular, respondents are asked for usual earnings. These figures, which represent current (and usual) incomes, are then annualised (weekly estimates multiplied by 52, monthly by 12 etc). Income from self-employment can be reported for a variety of periods, but it is always up-rated (using the UK’s average earnings index) to the interview date. For income from investment and employee non-cash income respondents are most likely to provide their most recent annual or half-yearly income that they received from this source. This income would be annualised, although there is no up-rating.

This approach is adopted in the UK because it is much easier for respondents to provide estimates of current income, rather than income for a specific reference period, say the most recent financial year. In the UK, only a relatively small proportion of the adult population fill in tax returns, and the rest of the population probably never actually calculate what their annual income is. For this reason, it would be very difficult to collect an estimate of annual income corresponding to a fixed reference year.

So the estimates of income do not correspond strictly to an income reference year. However, we can regard each household’s estimate of annualised current income as corresponding to a 12 month period centred around the interview date. So for a household interviewed in early January 2011, we can regard their income as being measured for the period July 2010 to June 2011, and similarly for a household interviewed in December 2011, the income estimate can be regarded as referring to the period July 2011 to June 2012. Since interviews are spread evenly throughout the year, for any one survey year, the interview reference periods collectively are centred around the calendar year. Therefore it is reasonable to regard aggregate statistics produced from the full annual datasets as measuring annual income in the current survey year. So the EU-SILC UK 2011 survey measures current annual income in 2011.

In the UK, household income statistics, and especially aggregate statistics such as those that are produced from EU-SILC, are generally used and interpreted on the assumption that this distinction between annualised current income, and what might be called a ‘true’ annual income, is small (Bönheim and Jenkins, 2006).

Bönheim, R. and Jenkins, S.P. 2006. A comparison of current and annual measures of income in the British Household Panel Survey. Journal of Official Statistics 22 (4), 733-758.

The reference period for taxes on wealth is based on data provided for the financial years April 2010–March 2011 and April 2011–March 2012. All interviewing for EU-SILC UK took place between 1^st January 2011 and 28^th February 2012.

F

F

F

F

F

F

F

F

F

F

F

F

L – for EU-SILC UK, income from private use of company cars is used to calculate PY020. However, none of the other goods and services highlighted in the EU-SILC definition are used. To collect data on these other goods and services would impart a high burden on the respondent and the accuracy of the collected data would likely be poor.

F

F

F

N/A

L – the only minor difference relates to redundancy payments. For EU-SILC UK, only information on regular redundancy payments is collected. One-off lump sum redundancy payments are excluded.

F

F

F

F

F

F

All income variables are collected at the point of interview. It is not mandatory for respondents to provide any documentation to support their answers. However, interviewers are being encouraged to ask respondents whether it is possible to consult their payslip (if they are working).

No information is collected from registers.

For most income components which are subject to taxation and/or social security contributions, respondents are asked to provide net and gross amounts. The only exception to this is income from interest, dividends and capital investments, which is collected either gross or net, and for which tax paid is then estimated.

Total income for an individual/household refers to income at the time of the interview. If the last pay packet/cheque was unusual, for example it included holiday pay in advance or a tax refund, the respondent is asked for usual pay. No account is taken of whether a job is temporary or permanent.

Gross and net income variables were asked separately, if applicable. If it was not possible to obtain a target variable then an imputation strategy was implemented. See section 12.5.2 for more details.

The EU-SILC Great Britain 2011 survey is based on a probability stratified two-stage sampling design. The sample design in Northern Ireland is a simple random sample. Stratification of the postcode sectors is done by geographical criteria for Great Britain.

In 2011, 11,857 addresses were sampled. Each year approximately 70% of the sample is rolled forward from previous years and the remaining 30% is a new “Wave 1” sample. EU-SILC UK aims to interview all adults aged 16 or over at every household at the sampled address.

Stratification and sub stratification criteria

Stratification involves the division of the population into sub-groups, or strata, from which independent samples are taken. This ensures that a representative sample is drawn with respect to the stratifiers. Stratification of a sample can lead to substantial improvements in the precision of the survey estimators provided that the strata are chosen such that members of the same strata are as similar as possible in respect to the characteristics of interest. The bigger the differences between strata, the greater the gain in the precision of the survey estimates.

Initially, postcode sectors for Great Britain were allocated to 30 major strata. These were based on the 10 regions in England (subdivided between the former Metropolitan and non-Metropolitan counties), five subdivisions in Scotland, two in Wales and one in Northern Ireland (Annex 2). In addition, London was subdivided into quadrants (Northwest, Northeast, Southwest and Southeast) with each quadrant being divided into inner and outer areas (Annex 2). Using a finer division of London significantly improves the precision of estimates.

It should be noted that regions and strata do not exactly map onto each other. There are 30 strata in Great Britain but 37 regions. Some strata contain cases from two or more regions and some regions contribute cases to more than one stratum.

Within each major stratum, postcode sectors were then stratified according to selected indicators taken from the 2001 Census. Sectors were initially ranked according to the proportion of households with no car, then divided into three bands containing approximately the same number of households. Within each band, sectors were re-ranked according to the proportion of households with a household reference person in socio-economic groups 1 to 5 and 13 (Annex 3), and these bands were then subdivided into three further bands of approximately equal size. Finally, within each of these bands, sectors were re-ranked according to the proportion of people who were pensioners. The ranking by pensioners and socio-economic group is carried out in reverse order so as to maximise similarity between one band and the next. A systematic sample of postcode sectors (PSUs) is selected from the ordered frame resulting in an implicit stratification of the sample. PSUs were then paired up to form pseudo-minor strata. The implicit stratification of the sample makes it possible to increase the precision of the survey estimates while ensuring good geographical coverage. It is just the major strata that are provided in the microdata D file.

Major strata were then divided into minor strata with equal numbers of addresses, the number of minor strata per major strata being proportionate to the size of the major stratum, so larger PSUs have more chance of being selected. In 2011 there were 912 minor strata. These consisted of 684 strata of waves 1, 2 and 3 from 2010 and 228 new wave 1 strata which replaced the 228 wave 4 strata from 2010.

Each PSU formed a quota of work for an interviewer. Within each of the 228 new PSUs, 23 addresses were randomly selected.

Sample selection schemes

EU-SILC Great Britain uses a two-stage sampling scheme:

1. Selection of Primary Sampling Units (PSUs) utilising a probability proportional to size sampling scheme.
2. Systematic random sampling of 23 addresses within aPSU.

The sample design in Northern Ireland is a simple random sample.

Sample distribution over time

Addresses are selected for the first wave of each panel using a random probability design, the detail of which is outlined in the preceding sections of this report.  The design weight for a household is calculated as the inverse of the inclusion probability for the samples address (e.g., a standard Horvitz-Thompson (HT) estimator).  The HT estimator is then adjusted by a two-step procedure to produce the wave 1 cross-sectional weight.

Non-response to the surveys (GLF and LCS) used to produce the EU-SILC data can introduce bias into the estimator. For the UK data, an attempt is made to correct for this bias through weighting households based on their estimated propensity to respond. For EU-SILC, non-response can occur at any given wave.

A non-response model exists for the GLF which comprises a number of adjustment classes.  These classes were constructed by linking households selected for the 2001 General Household Survey (the earlier version of the GLF) to the 2001 Census.  The Census is mandatory in the UK and so both responders and non-responders to the GLF can be matched to Census records. Response classes were formed based on households’ propensity to respond to the survey, condition on certain combinations of characteristics available in both the Census and the survey.  The reciprocal of the response propensity is used as the non-response weight.

Calibration is used in the weighting procedure both to improve precision and to ensure consistency with known population totals.  The EU-SILC sample is based on the population of private households, which means that the population totals used in the weighting need to be those created from counts of people living in private households.

At the time the weights were being constructed the most appropriate version of the population totals available for weighting were those produced for the Labour Force Survey (LFS).  The LFS derives household population estimates by excluding residents of institutions from population projections based on mid-year estimates.  However, certain groups in institutions are included in the population totals (e.g., nurses in nursing homes).

The population information and EU-SILC UK data were grouped into twelve age by sex categories and into six regional categories to form weighting classes. The initial non-response adjusted HT weight is adjusted, using Stats Canada’s Generalized Estimation System (GES), so that the final weights ensure that the weighted totals for the above demographic categories match the population totals.

Age-group by sex

0-4                                       0-4                              Males and Females                               

5-15                                    5-15                            Males and Females

16-24                            Males                                                     16-24                     Females

25-44                            Males                                                     25-44                     Females

45-64                            Males                                                     45-64                     Females

65-74                            Males                                                     65-74                     Females

75+                                Males                                                     75+                         Females

Regions

Metropolitan

Non-metropolitan

London

South East

Wales

Scotland

Northern Ireland

The strategy used to impute UK EU-SILC was consistent with the options proposed in the following Eurostat task-force documents associated with donor-based imputation methodology:

                                EU-SILC 74/02

                                EU-SILC 136/04

                                EU-SILC 154/05

The UK EU-SILC Imputation Strategy was developed with the primary aims of imputing for all item level missingness, resolving inconsistencies, and preserving both cross-sectional and longitudinal relationships in the responses for the households and persons affected.  The strategy was also designed to preserve the maximum amount of observed data.

Meeting the aims of the strategy was not trivial as the cross-sectional and longitudinal correlations were both nested and complex. In any one year, the UK EU-SILC dataset contains over 400 routing and income variables; routing variables indicate whether or not the respondent receives an amount whilst the amount itself follows on in one or more consecutive variables.  Missing values may be present in both the routing and the amounts collected.

Further complications include:

• legal constraints which make some combinations of the routing variables invalid;
• highly correlated relationships amongst subsets of the variables, for example, earnings before and after taxation followed by an associated time period for which the payment relates;
• the panel aspect of the survey introduces further correlations between years in addition to those within year.

To meet the aims of the imputation strategy the ONS implemented an iterative, two-stage imputation process: Stage 1 focused on the imputation of missing routing; Stage 2 focused on the imputation of missing amounts and time periods.

The imputation process was supported by statistical tools and used standard statistical techniques for panel data, including:

• SAS (Statistical Analysis System) – to facilitate deductive imputation. This was applied to correct for missing values by implementing propositional relationships in the data based on logical rules and legal constraints. For example, using gross values with auxiliary variables to derive missing net values;
• SPSSAnswerTree - to identify key predictors to partition the data into homogeneous classes for subsequent imputation;
• CANCEIS (CANadian Census Edit and Imputation System) - for stochastic imputation. CANCEIS implements a highly efficient nearest neighbour imputation method that preserves the shape of the distribution whilst also estimates and maintains observed relationships and distributional parameters. Stochastic imputation ensures less distortion in the estimates of variance. Asymmetric trimming was also applied as a refinement to exclude outlying values which might have otherwise caused excessive influence.

The quality of the final data was validated in two ways: by calculating expected values and observing the pre-and post-imputation distributions.

A UK EU-SILC imputed rent variable was supplied for 2011. Estimates of imputed rent were generated through the use of hedonic regression modelling, using the Heckman Two-Step method. The explanatory variables used in the regression were region, type of dwelling (flat, semidetached/terraced house, detached house), ownership of a car, value of dwelling (council tax band, except Northern Ireland), thermal comfort (ability to keep home adequately warm) and seniority (year of contract). The Heckman Two-Step procedure requires the dependent variable, in this case rent, to be converted to a log linear variable. Hence, predicted imputed rent was estimated as log linear variable. A back-log transformation was done to produce imputed rent in its proper form.

In the UK, company cars are taxed based on their carbon dioxide (CO₂) emissions. Therefore, UK EU-SILC assigns the benefit of having access to a company car as being equal to the level of tax. However, it is difficult to estimate the level of tax and therefore the following method is used.

EU-SILC UK asks several questions about company cars. First, the survey establishes whether the household has any company cars. Second, it establishes what the manufacturer’s list price for the vehicle was when it was new. If the respondent is unable to provide an answer, they are asked which price band they think the company car sits in. If the respondent gives a band price the answer is translated into a mid-point price. For example, a Mazda saloon with a band price between £10,001-£13,000 would be given a list price of £11,500.

The estimation of the value of using a company car for private purposes (excluding payment of fuel) is done using the following elements:

                1. Type of fuel used;

                2. Data from VCA (Vehicle Certification Agency, UK);

                3. Price of the car.

Once the price of the car is known (using one of the methods described above) a factor based on fuel type and emissions of the engine is applied to that list price. However, this is problematic. Although data on the make and model of each car is collected, the quality of answers given by respondents is extremely variable, for instance, answers such as ‘a red Ford’ offer little value to a calculation. Therefore the estimates are based on average tax bands for cars of certain price bands.

The factors used for 2011 are shown in below. Diesel cars have a 3% uplift from petrol cars, which reflect the extra tax charged for these vehicles. Electric vehicles have no tax charged on them whilst hybrid vehicles and vehicles which run on fuel consisting of up to 85% denatured ethanol (E85) incur a tax rate of 15% regardless of the list price of the vehicle.

Tax rate based on CO₂ emission rates (%)

These percentage rates are the factors that are applied to the car prices to produce a monetary benefit for each company car in a household.

Car benefit = Car price × CO₂ tax emission rate

Version control

Version

Date delivered to Eurostat

Revision summary

V1

9/11/2012

Provisional data sent to be included in EU aggregate data.

V2

20/11/2012

First full sent after final quality assurance with some minor adjustments made to V1.