National Quality Report SWEDEN 2005

Transcription

1 National Quality Report SWEDEN 2005 The objective of this paper is to give an overall picture of the quality of the Swedish data for the Structural Business Statistics (SBS). The report consists of three parts. Part 1 gives a description of the survey strategy. Part 2 is a report on the methods used for determination of principal activity for the units involved in SBS. Part 3 contains estimates of quality indicators: coefficients of variation and non response rates. 1. Survey strategy For the years , all enterprises with 50 employees or more and belonging to the target population of SBS were included in the sample by total enumeration. A questionnaire was sent to every such enterprise. For the rest of the enterprises there was no direct data collection from the businesses. Instead administrative data, the so called SRUs Standardiserade RäkenskapsUtdrag in Swedish or Standardised Annual Company Reports in English, was used. The SRUs have been collected on an annual basis by the National Tax Board (National Revenue Service) from all Swedish enterprises since the early 1990s. Although the main purpose is to assist the National Tax Board in its work, the SRU data are available to Statistics Sweden for statistical purposes. The SRU material was tested in 1994/95 by Statistics Sweden for its potential as data source. As a result of these tests, the SRUs are used on a full scale for the SBS since From the reference year 2003 the method has changed. In order to improve the quality and to reduce the response burden on enterprises the administrative data is used to a greater extent. Briefly the new SBS uses a combination of different data sources such as SRU, data from a number of existing surveys at Statistics Sweden and three different sample surveys carried out within the SBS framework. It is of vital importance to the overall quality of the SBS that we achieve high quality data from the biggest enterprises. In order to do this we have been working with two different data collection periods, one in spring where the largest enterprises were surveyed and one in the fall where the other enterprises were surveyed. To separate the data collection period hopefully helps us in that strive. Variables we have for all the enterprises For the base variables the administrative data is then main source, only the largest companies (roughly 500), get a questionnaire. However, though these companies are the very Postal address Office address Telephone Fax Box , SE STOCKHOLM, SWEDEN Karlavägen SE ÖREBRO, SWEDEN Klostergatan Website: scb@scb.se Registered No.: V.A.T. No.: SE

2 2(29) largest, the total turnover for these companies are about 36 per cent of total market producers. Every large enterprise divided into two or more kind-of-activity units (KAUs) received one questionnaire for every separate KAU. There were about 80 such enterprises. For the largest companies there are no missing values, because in principle, it is possible to re-create the variables from the annual reports. However, the SRUs are sometimes missing or rejected, on the average about 15 per cent. The missing SRUs are mostly among the smallest businesses. This means that the impact of the nonresponse on estimates is generally much lower than the reported unweighted nonresponse rates may seem to indicate. Figure 1. Nonresponse rate for the base variables. Total market producers. 100% 90% 80% 70% 60% 50% 40% Non response Questionnaire Administrative data 30% 20% 10% 0% Number of enterprises Turnover Variables we only have for a sample of enterprises The variables in the administrative data is supplemented with three separate samples. These separate questioners gives more detail information about some of the variables collected in the administrative data. For the sample surveys information from SRU is used to construct efficient sampling designs. Through the SRU we know the values for virtually all enterprises in the SBS population and in the estimation phase we in fact treat them as known population totals. The surveys can therefore be limited to obtaining estimates for different kinds of relative distributions rather than estimates of unknown totals. Pre-printed values from the SRU in the questionnaire also help easing the burden for the respondents. The three questioners studies the profit and loss account, shares and stocks and investments. Regarding investments, enterprises in manufacturing and mining and quarrying, with two or more local units were required to fill in a special questionnaire for every local unit. The three

3 3(29) samples are in principle independent of each other. The largest companies get one questionnaire where the three parts are included. 2. Determination of Principal Activity for Enterprises in the Swedish Business Register Data Sources for principal activity An enterprise at birth: Every newly registered legal unit carries a code for its principal activity (by NACE rev.1) as it enters the Business Register (BR) at Statistics Sweden (SCB). This code is provided by the Swedish National Tax Board, and based on information from an activity declaration which is mandatory for every person or group of persons who wants to register a new legal unit in Sweden. As the new legal unit enters the BR the system automatically generates one local unit, one local kind of activity unit, one kind of activity unit and one enterprise unit. All of these units, together with the original legal unit, are connected and given the same activity code as the one carried by the legal unit. This group of connected units Local, LKAU, KAU, Legal, Enterprise - is an enterprise at birth in the BR. Besides activity code, other attributes such as address and number of employees (if any) are attached to the enterprise. The PAYE ( Pay As You Earn ) register is the major information source on employment of the enterprise as long as it is too new to be included in surveys conducted by SCB. Enterprises with more than one local unit: There is no formal way (such as an administrative source or a census) by which SCB can detect if an enterprise has more than one local unit. However, if various regular performed checks indicate that an enterprise with a single registered local unit may be active at several locations (a large number of employees is one indication, feedback from surveys is also used), then this enterprise is contacted and as a result of this contact one or more new local unit(s) may be created for the enterprise. In November/December every year, all enterprises with two or more local units are included in a specific survey (the PAYE survey ). The purpose of this survey is to let the enterprises involved define all their local units to which they later (in the spring), on request, are going to attach the PAYE data for their employees. Every person in Sweden with a PAYE value over a certain limit (presently about SEK) is connected with a particular local unit among those defined from the survey, i.e. the local unit where the person has earned most of his/her money during the previous year. The PAYE survey does not collect information for updating of activity codes. However, there is another annual survey, the BR survey performed in the spring, which does collect information on activity for every local unit of multi-local enterprises. For every local unit involved, all activities previously registered by SCB and ordered by their relative importance

4 4(29) (size, see below) are pre-printed. The respondent is asked to verify or to revise the pre-printed information. The BR survey is the main source for determination of principal activity of enterprises (and their connected units). However, for enterprises involved in manufacturing (and above a certain cut-off limit in size) the Annual Production Statistics in Manufacturing (the PRODCOM Survey ) is used to determine principal activity. Enterprise and Legal unit: In a limited number of cases presently about 50 cases - the enterprise comprises several legal units. Hence for the vast majority of enterprises in the BR there is (at a fixed point in time) a one-to-one relation between enterprise and legal unit. Normally, both have the same principal activity. However, if the enterprise is a holding company (head office), which is regarded as an ancillary activity, then the enterprise and the legal unit, although one-to-one, may have different principal activity. This is the most common exception to the general rule (i.e. same activity) in the one-to-one case. The principal activity for an enterprise is determined from activities registered at the local units connected with the enterprise. Local unit and LKAU: An LKAU consists of the whole or a part of a local unit in the BR. At the present time a very limited number of local units (about 70) are split into several LKAUs. In an overwhelming majority of cases there is a one-to-one relation between LKAU and local unit. Hence the LKAU is the smallest unit of the BR, however mostly coinciding with the local unit. Enterprise and KAU: In most cases there is a one-to-one relation between Enterprise and KAU. However, a number of enterprises presently 75 - are split into several KAUs. In such cases the KAU is an aggregate of LKAUs Classification algorithm The algorithm used to determine principal activity for a unit (local, enterprise etc.) in the BR takes account of the hierarchical nature of NACE rev.1. The simplest way to describe it is by an example. Consider the following fictitious enterprise: From the Annual Commodity Survey we know that the enterprise produces commodities corresponding to the following NACE classes. We also know for each class its share (in production value) of the total production value for the enterprise. NACE class Share of Total Production (MSEK) DG % DG % DG % DH % DK % Total 100% Step1: Determine Primary Section: Section D is the only one represented. Hence the primary section is D. Step2: Determine Primary Subsection. Subsections DG (45%), DH (25%) and DK(30%) are represented. Since DG is the largest, the primary subsection is DG.

5 5(29) Step3: Determine Primary Division within the Primary Subsection. The only division represented within the subsection DG is 24. Hence the primary division is 24. Step4: Determine Primary Group within Primary Division. Groups 241 and 246 are represented within division 24. Group 241 accounts for 25% of total production while 246 accounts for 20%. Hence the primary group is 241. Step5: Determine Primary Class within Primary Group. Classes 2414 and 2416 are represented within the primary group. Class 2414 (15% of total) is larger than 2416 (10% of total). Hence the primary class is This algorithm is applied to any type of unit in the BR for which data of the above type are available: Local Units/LKAUs, KAUs, Legal Units and Enterprise Units. However, the percentage distribution between activities as in the example may be taken from different sources. If the PRODCOM Survey is the data source then the distribution is based on production value per activity relative to the production value for the whole unit. If the BR Survey is used as data source then the percentage distribution is based on the respondent s own judgement of the relative size of the different activities (size is usually by number of employees, hours worked or a similar measure). If the respondent has reported several activities, and their ordering in size, at a local unit but not given any further, quantitative information about relative sizes, then the following standard rule is applied by SCB. If for example two such activities are reported without quantitative information on their relative size then the rule automatically assigns 80% to the primary and 20% to the secondary activity. This standard rule is often used for units the service sector. % for activity i primary second third forth fifth--tenth Number of activities Reported:

6 6(29) 3. Quality Indicators: Coefficient of Variation and Nonresponse Rate Member States shall, by Eurostat requirement, report the coefficient of variation for various characteristics and activity levels. The coefficient of variation (CV) shall take into account three main factors: Sampling error Misclassification error Non-response error The following tables convey the requested estimates of coefficients of variation from the Swedish SBS with reference year Sampling: There is no sampling error in the Swedish SBS-2005 for the base variables since for these variables no sampling procedure has been used. Parts of the value added are based on sampling. The sampling error for these parts have been included in table 3.1 and 3.2. Misclassification is defined in the requirement notes in the sense that a unit may be classified in one activity at the sampling occasion and in another activity at the time of estimation. However, since a census approach is used and the most recent information on activity is used to form domains, there is no contribution to the coefficient of variation from this source in the SBS Unit Nonresponse: There is some nonresponse among units in the SBS Although the main concern with nonresponse is usually its possible contribution to bias of estimates, there may also be random variation stemming from nonresponse. The Eurostat requirements focus on the random component of the nonresponse error. These contributions (which are in fact the only contributions) to the coefficient of variation are estimated in Tables ahead. The unweighted proportion of unit nonresponse by 3 digit NACE is presented in Table 3.4. Item Nonresponse: Member States shall report the item nonresponse rates by 3 digit NACE for the following characteristics: (turnover), (value added), (labour cost), (gross investments), (number of employees). All of these items except for some very minor components of value added can be found in the official Profit and Loss Accounts or in the SRU material. Hence the proportion of item nonresponse for all of these items coincide with the unit noresponse rate of missing SRUs. (An SRU is either completetly accepted, completely rejected or completely missing. SRUs were never partly used in the SBS-2005). Method used for estimation of CVs: Official Profit and Loss Accounts and SRUs are used as imputation sources for the large enterprises which are subject to a conventional postal inquiry. Since all items involved in the present report are found in those sources, it is fair to say that there is no nonresponse and no random variation emanating from the large enterprises. Only small enterprises then contribute to the CV. Strata comprise 3- digit NACE groups for sole proprietors and, for all other legal forms, the combinations of 4- digit NACE sub groups and size group by number of employees (0, 1-4, 5-9, 10-19, 20-49).

7 7(29) Reweighting within strata is done: The weight Number of frame elements/number of sample elements is replaced by the weight Number of frame elements/number of responding elements.