Households' consumption: documentation of statistics

The statistics on households' consumption produce data on the structure of households' consumption, changes in consumption expenditure and differences in consumption by population group. The data collection of the statistics is called the Household Budget Survey. Data collections covering the whole population have been conducted at regular intervals since 1966, roughly every five years. The latest is from 2016.

The population of the statistics on households' consumption consists of households permanently resident in Finland and their members. Excluded from the statistics are Finnish citizens resident abroad and the institutional population, which includes long-term residents of hospitals, old people's homes, care institutions, prisons, etc.

The basic unit of the statistics on households' consumption, on which data are produced, is a household. It consists of persons who live together and who have, in whole or in part, a shared food economy or otherwise use their income together. Consumption data are collected and analysed mainly by household, even though data by household member are also collected.

The published data of the statistics on households' consumption use consumption expenditure in euros, on average, per household or per consumption unit as the unit of measurement.

The reference period of the statistics on households’ consumption is the calendar year. In the data collection, the reference periods are two weeks, one month, three months or one year. The consumption data inquired for a period of under 12 months are scaled to the annual level in the final data. When planning the data collection, attention is paid, for example, to the frequency of goods and services acquisition and the price of them. These influence how the data are collected and what kind of reference period is used in the question.

The statistics on households' consumption cover the consumption expenditure of the household population permanently resident in Finland.

The data of the statistics are published on the level of the whole country and by major region (NUTS2).

The statistics on households' consumption cover the consumption expenditure of private households permanently resident in Finland.

The representative population sample of the statistics on households' consumption describe the economic well-being of households and persons as a whole more extensively than any other statistical data. In addition to the demographic and regional data on households, the statistics also contain data on households':
–    consumption expenditure
–    disposable income
–    interests on housing loans and other loans – housing conditions

The statistics on households' consumption, or the Household Budget Survey, are one of the oldest sample surveys in Finland, their history reaches back to the years 1908 to 1909. Household Budget Surveys covering all households in Finland have been conducted approximately every five years since 1966. Starting from 1966, data are available in electronic format.
Two time series have been produced based on the data, of which the older one contains comparable data for the years 1966, 1971, 1976, 1981, 1985 and 1990. The newer time series contains data for the years 1985, 1990, 1995, 1998, 2001, 2006, 2012 and 2016.

The statistics on households' consumption are produced at intervals of five years, on average. The results of the statistics are released as preliminary data approximately 12 months after the statistical reference year and as final data at a lag of approximately 15 months. In addition, several releases by theme are made from the statistics for each survey round.

In the statistics on households' consumption, sampling, data collection and data processing affect the reliability of the results. The size of random variation in the survey results can be measured with the help of the relative standard error of the mean and confidence intervals.

In the so-called main groups of the 2016 consumption expenditure, the relative standard errors are small apart from education expenditure. The most reliable data come from the biggest consumption expenditure groups (food and non-alcoholic beverage, and housing and energy). The relative standard errors for clothing and footwear and education expenditure are highest.
When using the data of the Household Budget Survey it should be noted that the reliability of consumption data weakens when moving from the main group level to the consumption sub-groups. Standard errors of often purchased commodities are small, but for less often purchased products, standard errors may become quite large. In these cases, the sample may not be sufficient to describe consumption expenditure reliably. Detailed classifications by background variables, such as age, socio-economic group or region, also increase the relative standard errors of consumption items as the observations decrease.
Standard error calculations describe random variation related to the results. In addition, households’ non-response and not remembering cause some consumption items to be only partially collected from sample households, which means that households' average consumption expenditure is underestimated. The error caused by this may in some cases (e.g. alcohol expenditure, games of chance) be considerably bigger than a random error. Underestimation of consumption expenditure items can be studied by comparing the results with corresponding national accounts data, for example. In 2016, the share of total expenditure in the Household Budget Survey of national accounts data was good 80 per cent.

As other sample surveys, the Household Budget Survey contains several possible sources of error. The main sources of error affecting the reliability of the results are related to data content, formulation and order of presentation of the questions, sampling, non-response, data collection and processing of data and analysing the results.

Errors related to sampling may be random or systematic. Random variation is caused by only part of the population being included in the survey. The size of the random error is influenced by the size of the sample, the method for drawing the sample and estimation of the results. The size of the random error is further affected by variation in the observation values of the variables (dispersion). A random sampling error is assessed with the help of standard errors of estimates.
Systematic errors in the results are caused by coverage, i.e. frame, errors in the sample. The share of coverage errors of total errors related to the Household Budget Survey is estimated to be small.

In the different stages of the Household Budget Survey, attention was paid to finding and minimising the errors. At the start of the study, the focus was on careful planning of the data content and testing of the questionnaire. Efforts have been made to minimise the risk of errors in the data collection stage through interviewer training and standardisation of the interview. Several interviewers had experience of previous Household Budget Surveys. In addition, the computer-assisted (CAPI) interview standardises the interview and enables the use of various limitations to the values entered on the form. Because the data of the Household Budget Survey are mainly collected through interviews and consumption diaries, the work input and motivation of the interviewers are key for the whole survey.

Despite careful planning of the data content and the questionnaire, it is not possible to reliably collect all consumption expenditure of households. This is partly due to households' memory errors or non-response. The Household Budget Survey produces too low consumption, for example, for alcoholic beverages and games of chance and other similar products that are often seen as socially undesirable. The consumption expenditure of the Household Budget Survey is compared with the consumption data of the household sector in the national accounts of the corresponding year. Processing errors here refer to errors during data storage, processing and coding. The aim was to reduce them by means of various checking rules. Systematic errors in the results may also be caused by non-response, that is, a sample household being left outside the survey. Non-response is generally understood as one key indicator of the quality of the data. The increase in non-response has been the most significant factor weakening the quality of survey data in recent years.

The statistics on households' consumption are produced at intervals of five years, on average. The results of the statistics are released as preliminary data approximately 12 months after the statistical reference year and as final data at a lag of approximately 15 months.

Time series corrections caused by methodological changes etc. are made when necessary.

As a rule, there has been no delay between the release calendar and the actual release date in the statistics on households' consumption. In individual cases, the release has had to be postponed due to late delivery of the administrative data used by the statistics.

In the main groups of consumption expenditure, relative standard errors do not exceed four per cent except for education expenditure. Compared with 2012, standard errors did not change much. The relative standard error of all consumption expenditure was 1.15 per cent in 2016 (1.28 per cent in the 2012 survey).

Over-coverage includes deceased persons, the institutional population and Finnish citizens resident abroad. Non-sampling errors are partly due to the fact that it may be a year between the sampling stage and the interviewer’s contact with the sample person, even though the addresses were updated twice. Over-coverage errors can also be caused by the difference between the pursued population of the survey and the population of the register data used in the sampling frame: registers cover the whole population, while the household population has been defined as the population of the Household Budget Survey. Correspondingly, under-coverage consists mainly of the immigrants registered in the Central Population Register between the time of sampling and the time of the survey. The significance of coverage errors in the total error related to the data has been estimated as low compared to other sources of error.

The gross sample consisted of 8,216 persons, some of whom were not included in the population of the survey. This over-coverage includes deceased persons, the institutional population and Finnish citizens resident abroad. The number of over-coverage was 193 persons, so the net sample consisted of 8,023 households.

The size of the net sample was 8,023 households, of which a total of 4,350 households were non-response cases. Thus, the total non-response of the survey was 54.2 per cent.

There is quite little item non-response in the interview, because all questions are discussed with the interviewer in a computer-assisted interview. It is difficult to examine item non-response, or individual missing responses, in the Household Budget Survey because it cannot be ascertained from the purchase data of the receipt collection period whether the household has not recorded or not purchased the product in question within two weeks.

Item non-response was imputed for the variables describing consumption. If the interviewee had answered “do not know” or refused to answer a question, the data were imputed using the donor imputation method.

The data from the statistics on households' consumption are delivered to Eurostat so that their content (concepts, definitions and classifications) is comparable with surveys made in other EU countries. Methodologically, the surveys vary considerably from country to country.

The data of the statistics on households' consumption published nationally in Finland differ from the data supplied to Eurostat. The national classification of consumption expenditure differs partly from the eCOICOP classification used by Eurostat.
The biggest differences compared to the national version relate to the definition of consumption expenditure. Finland's version is considerably more detailed. In addition, in the national survey, consumption expenditure includes the following items of consumption expenditure that are not included in consumption expenditure in Eurostat's classification:
–    some tax-like payments (e.g. vehicle tax, dog tax)
–    membership fees, fines and other transfer charges
–    other items outside consumption expenditure (e.g. payments for estate agents and housing agents, payments for building permits, interests). The reason for these differences is maintaining the comparability of the national time series.
 

Two national time series have been produced based on the data of the statistics on households’ consumption, i.e. the data of the Household Budget Survey, of which the older one contains comparable data for the years 1966, 1971, 1976, 1981, 1985 and 1990. The newer time series contains comparable data for the years 1985, 1990, 1995, 1998, 2001, 2006, 2012 and 2016.

When comparing the data of the statistics on households' consumption with the consumption expenditure of the household sector of the national accounts, differences in defining the sector, data collection and definitions, as well as in the compilation methods of the statistics, should be taken into account. In different statistical reference years, there are some differences in total expenditure between the national accounts and the Household Budget Survey. In 2016, the consumption expenditure estimated based on the Household Budget Survey was good 80 per cent of the corresponding sum of national accounts. Due to differences in calculation methods, the deviation cannot be regarded as very significant. However, the differences vary greatly depending on consumption group. The differences in consumer durable goods and non-durable goods are around 10 per cent. In semi-durable goods and services it is around 30 per cent. This is probably due to changes in the data collections. For example, the sample size, participation in the surveys and modes of data collection have varied from one Household Budget Survey to another. In national accounts, the Household Budget Survey is used for level revision of private consumption expenditure.

The data of the statistics are internally coherent. There are no differences in definitions or examination periods within the same set of statistics.

The statistics on households' consumption are a sample survey for which data are collected with three complementary methods:
–    Interviews with the households
–    Receipts collected and consumption diaries filled in by the households
–    Administrative registers
Efforts have been made to use register data as much as possible to reduce the response burden of households and the costs of data collection. However, actual consumption data cannot be obtained from registers. By contrast, in addition to sample data, the household's income data and data on the members' level of education derive from the register.

The Household Budget Survey is a sample survey. The population of the survey comprises households permanently resident in Finland, i.e. the so-called household population. In 2016, there were 2.677 million households. Excluded from the survey are Finnish citizens resident abroad and the institutional population, which includes long-term residents of hospitals, old people's homes, care institutions, prisons, etc.

In the survey of 2016, the sampling design was a two-phase stratified sampling. In the first phase, a so-called master sample was formed by selecting 100,000 target persons aged 16 or over by means of systematic sampling from Statistics Finland’s population database. In the second phase, the actual sample for the Household Budget Survey was selected by stratum from the master sample. The stratification was made according to the areas of the Regional State Administrative Agencies (AVI) (Greater Helsinki separately) and, in addition, by dividing household-dwelling units into persons aged under 65 living alone and other household-dwelling units. The total number of strata was 16. Within the stratum, the probability of each household entering the sample depends on the number of household members aged 16 or over.

In 2016, the original gross sample consisted of 8,216 persons, some of whom were not included in the population of the survey. This so-called over-coverage includes deceased persons, the institutional population and Finnish citizens resident abroad. The number of over-coverage was 193 persons, so the net sample consisted of 8,023 households. 
 

The data for the statistics on households' consumption 2016 were collected with a telephone/face-to-face interview. In addition, data was obtained from purchase receipts collected by the household and from purchase entries made in the consumption diary. Data were also retrieved from administrative registers.

In the interview, the structure of the household according to register data is checked and data are collected on the members of the household (occupations, activity on the labour market, etc.) and data are inquired on households' expenditure, such as housing, insurance and large purchases, and owning durable consumer goods.

Statistics Finland’s own interviewers carry out the interviews. After the interviews, the households collected receipts or recorded purchases in the consumption diary concerning all their expenses during two weeks. The sample households of the survey were randomly divided into 26 receipt collection periods of 14 days throughout the survey year. Thus, as comprehensive and reliable data as possible are obtained on the consumption in different seasons. The receipts were returned either to the statistical interviewer with whom an agreement had been made about the collection of receipts or directly to Statistics Finland. The receipts were scanned and converted to electronic format at Statistics Finland. Administrative registers were utilised in collecting background data. For example, data on the income and education of household members and on the use of public services were obtained from them.

The data collection for the statistics on households' consumption is conducted roughly every five years. The latest is from 2016.

The Statistics Act (280/2004) obliges Statistics Finland to treat the data as confidential. Respondent households cannot be identified from the published survey results and tables. Statistics Finland's trained interviewers and other employees are bound by an obligation to observe secrecy decreed in law. The survey data based on interviews and receipt and expenditure information for two weeks are supplemented with data from administrative materials available to Statistics Finland. Statistics Finland can release data only for scientific research and statistical surveys. Other data can later be combined to the released data. Releasing is always done so that direct identification of the survey participants is impossible.

The average duration of the telephone interview for the Household Budget Survey data collection was around one hour. The interview stage was followed by a two-week period of collecting receipts and keeping a consumption diary. Participants were instructed to collect each receipt with product data during two weeks. Other expenses were to be recorded in the diary.
We strive to minimise the response burden by e.g. questionnaire designing and developing data collection tools.

In the 2016 statistics on households' consumption, item non-response was imputed for the variables describing consumption as in the 2012 survey. If the interviewee had answered “do not know” or refused to answer a question, the data were imputed using the donor imputation method (Banff procedure in SAS). Only variables of which it was known that the data should be greater than zero were imputed. For example, a few per cent of the data related to housing were imputed with this method. Of insurance data, the share of imputed data was nearly 14 per cent. The data donor was retrieved by variable by first deciding on so-called must match variables, based on which a donor that is as similar as possible is sought.
Imputation is an integral part of other data editing, that is, the detection and correction of inconsistent or erroneous data. For all corrected variables, information about the correction was entered into the so-called flag variable in the database. The possible values of the flag variables were: 1: Imputed with Banff (donor imputation) 2: Data retrieved from administrative dataset 3: Data distributed to several variables (e.g. insurance packages) 4: Other imputation method (e.g. mean or median) 5: The data have been corrected (e.g. outliers).

The distribution of data to several variables (flag=3) was primarily used for insurance packages, that is, in a situation where the household told what insurances it had and what their total sum was. Generating data with the help of administrative data was used, for example, in connection with variables concerning housing (e.g. year of construction, floor area). Other imputation methods were mainly used for durable consumer goods, of which under two per cent were imputed.

When estimating the total sums and averages of consumption expenditure, the inclusion probabilities and response probabilities related to sampling and additional information available from the population are utilised in weighting the data of each household.
The bias caused by the skewed non-response was corrected with the help of re-weighting. In the re-weighting, the so-called correction for non-response is first made by multiplying the inclusion probabilities with the estimated response probabilities. After this, the weights corrected for non-response are calibrated. The purpose of calibration is to improve the estimates calculated from the sample by means of additional information obtained from the target population. The weighting coefficients of the households having responded to the survey were adjusted so that, with regard to the most important background variables, the marginal distributions of the households having responded corresponded to the marginal distributions of the entire household population. The CALMAR calculation software was used for calibration.

The following marginal distributions for 2016 obtained from registers were used in the weight calibration:
1)    distribution of household-dwelling units by region (NUTS3) - Greater Helsinki separately: Helsinki, Espoo (+ Kauniainen) and Vantaa each formed their own “region”
2)    distribution of size of household-dwelling units: 1, 2, 3, 4, 5, 6, 7, 8+ persons
3)    distribution of the socio-economic group of household-dwelling units
4)    distribution of level of education of household-dwelling units
5)    population distribution by gender and age
6)    total of taxable earned income of household-dwelling units
7)    total of taxable capital income of household-dwelling units.
In addition, in the calibration, the sums of the weight coefficients of the receipt collection periods are forced to be the same in order for the data to represent all periods equally well.

The interview, receipt and diary data of the statistics on households' consumption are checked in multiple phases. Basic checks include: 
–    review of the comments written by the interviewer in connection with the interview
–    checking the minimum and maximum of interview data
–    checking the 15 biggest and smallest values of each code in the receipt and diary data – logical checks (e.g. that each household has certain mandatory data).

A separate user handbook (in Finnish) has been made for the statistics on households' consumption in each survey round and in it the survey method is described as accurately as possible: https://stat.fi/til/ktutk/men.html

Statistics Finland

Social Statistics

The compilation of statistics is guided by the Statistics Act. The Statistics Act contains provisions on collection of data, processing of data and the obligation to provide data. Besides the Statistics Act, the Data Protection Act and the Act on the Openness of Government Activities are applied to processing of data when producing statistics.

Statistics Finland compiles statistics in line with the EU’s regulations applicable to statistics, which steer the statistical agencies of all EU Member States.

Further information: Statistical legislation

The data protection of data collected for statistical purposes is guaranteed in accordance with the requirements of the Statistics Act (280/2004), the Act on the Openness of Government Activities (621/1999), the EU's General Data Protection Regulation (EU) 2016/679 and the Data Protection Act (1050/2018). The data materials are protected at all stages of processing with the necessary physical and technical solutions. Statistics Finland has compiled detailed directions and instructions for confidential processing of the data. Employees have access only to the data essential for their duties. The premises where unit-level data are processed are not accessible to outsiders. Members of the personnel have signed a pledge of secrecy upon entering the service. Violation of data protection is punishable.

Further information: Data protection | Statistics Finland (stat.fi)

Statistics Finland's general guidelines are applied in the publication of tabulated statistical data of the statistics on households' consumption.

An individual household cannot under any circumstances be identified from the published data.
Release of data for research use is done centrally by Statistics Finland's research services. The survey data of the Household Budget Survey are mainly household-specific and households are identified with consecutive numbers. Some of the classification data of the household are person-based. The data do not contain identification data. In addition, before granting a licence to use the data and releasing the data for research use, the data protection of the data is assessed in accordance with the normal practice of the research services and the data are made less detailed if necessary.
Unit-level data are released to Eurostat. The released data do not contain identification data.

Statistics Finland publishes new statistical data at 8 am on weekdays in its web service. The release times of statistics are given in advance in the release calendar available in the web service. The data are public after they have been updated in the web service.

Further information: Publication principles for statistics at Statistics Finland

The data of the statistics on households' consumption are submitted to the Consumer Price Index, the international price comparison and national accounts. In addition, the data of the statistics are submitted to Eurostat and Eurostat publishes the data on its web pages.

Eurostat, the Statistical Office of the European Union, is responsible for compiling statistics on the HBS and for the release of its statistical data for research use. Research use requires an application for licence to use statistical data.

The data of the statistics are also published as articles and blogs in Statistics Finland's Tieto & Trendit periodical and in social media.

Statistical data are published as database tables in the StatFin database. The database is the primary publishing site of data, and new data are updated first there. When releasing statistical data,

existing database tables can be updated with new data or completely new database tables can be published.

In addition to statistical data published in the StatFin database, a release on the key data is usually published in the web service. If the release contains data concerning several reference periods (e.g. monthly and annual data), a review bringing together these data is published in the web service. Database tables updated at the time of publication are listed both in the release and in the review. In some cases, statistical data can also be published as mere database releases in the StatFin database. No release or review is published in connection with these database releases.

Releases and database tables are published in three languages, in Finnish, Swedish and English. The language versions of releases may have more limited content than in Finnish.

Information about changes in the publication schedules of releases and database tables and about corrections are given as change releases in the web service.

Data collected for the statistics on households' consumption are released for research use in accordance with Section 13 of the Statistics Act. The release procedure is explained in the guidelines issued by Statistics Finland on granting user licences to Statistics Finland's unit-level data. Release of data for research use is done centrally by Statistics Finland's research services.
The survey data of the Household Budget Survey are mainly household-specific and households are identified with consecutive numbers. Some of the classification data of the household are person-based. The data do not contain identification data.
Unit-level data are released to Eurostat. The released data do not contain identification data.

Revisions – i.e. improvements in the accuracy of statistical data already published – are a normal feature of statistical production and result in improved quality of statistics. The principle is that statistical data are based on the best available data and information concerning the statistical phenomenon. On the other hand, the revisions are communicated as transparently as possible in advance. Advance communication ensures that the users can prepare for the data revisions.

The reason why data in statistical releases become revised is often caused by the data becoming supplemented. Then the new, revised statistical figure is based on a wider information basis and describes the phenomenon more accurately than before.

Revisions of statistical data may also be caused by the calculation method used, such as annual benchmarking or updating of weight structures. Changes of base years and used classifications may also cause revisions to data.

The Household Budget Survey has conventionally produced data for the compilation of macroeconomic indicators 
(Consumer Price Index, level and structure of household sector consumption in national accounts), for decision-making within the European Union to describe differences in welfare by population group and region in the Member States (especially poverty and social exclusion), and nationally to describe the preconditions and distribution of economic well-being. The data are also widely used for planning and monitoring social policy measures. The Household Budget Survey also produces data for statistics describing the use of energy, use of information technology, transport, mass media and culture as well as for market analyses.

National accounts produce macro level data on the consumption expenditure of the household sector at the level of the whole national economy annually. The data of the Household Budget Survey are used in the level revisions of household sector consumption. The main source for the structure of the commodity basket of the Consumer Price Index is national accounts data on private consumption, which are supplemented with more detailed data from the Household Budget Survey.
The Household Budget Survey describes households' resources of economic well-being, which is why it is close in terms of content to the income distribution and wealth surveys. Households' consumption also includes activities which link the Household Budget Survey to time use, cultural and leisure surveys, as well as surveys and statistics describing the environment and the mass media.

Universities and research institutions use the data of the Household Budget Survey when studying, for example, the distribution of well-being by population group or region, poverty and social exclusion or social problems. The data are also used in the so-called lifestyle survey and econometric surveys. Consumption data are also linked to the microsimulation model (SISU). In addition, Statistics Finland makes data analyses and printouts as information service, based on the needs of the users.

Users' views and opinions are not collected regularly. Based on the number of parties ordering chargeable data, the content is perceived to be relevant. In connection with changes concerning the statistics, an expert group has been heard.

Official Statistics of Finland (OSF) are a comprehensive collection of statistics describing the development and state of society. They comprise nearly 300 sets of statistics on 26 different topics. The producers of Official Statistics of Finland have approved a common quality assurance in which they commit to common quality criteria and quality assurance measures. The quality criteria of Official Statistics of Finland are compatible with the European Statistics Code of Practice. The good practices followed in the statistics are presented in Statistics Finland's Quality Guidelines for Official Statistics handbook.

Quality management requires comprehensive guidance of activities. The quality management framework of the field of statistics is the European Statistics Code of Practice (CoP). The quality criteria of Official Statistics of Finland are compatible with the European Statistics Code of Practice.

Further information: Quality management | Statistics Finland (stat.fi)

Data are released to all users at the same time. Statistical data may only be handled at Statistics Finland and information on them may be given before release only by persons involved in the production of the statistics concerned or who need the data of the statistics concerned in their own work before the data are published. Further information: Publication principles for statistics

Unless otherwise separately stated in connection with the product, data or service concerned, Statistics Finland is the producer of the data and the owner of the copyright. The terms of use for statistical data.