Establishment of the Statistical Business Register at SORS and Its Impact on Sampling Design. Boro Nikić, Maruša Stanek

Size: px
Start display at page:

Download "Establishment of the Statistical Business Register at SORS and Its Impact on Sampling Design. Boro Nikić, Maruša Stanek"

Transcription

1 Establishment of the Statistical Business Register at SORS and Its Impact on Sampling Design Boro Nikić, Maruša Stanek NTTS 2013

2 Motivation for coordinated sampling current state The statistical office is under increasing pressure to reduce the burden of responding units and obtain precise results about population or domain characteristics at the same time 2

3 Motivation for coordinated sampling - problem If we select samples independently we don t have control over response burden of units as some of them might be selected for many surveys or to take in surveys with varying complexity. We want to spread the burden of respondents by reducing the overlap between samples. 3

4 Sampling process (master) sampling frame type of sampling for survey sampling design sampling allocation selection method 4

5 Business surveys at SURS type of (master) sampling for sampling sampling sampling frame survey design allocation master sampling frame selection method At the beginning of each year two master sampling frames are created by integration of different sources: sampling frame of legal units and sampling frame of local kind of activity units. Inactive units are excluded, some activities of units are changed. 5

6 Sources of master sampling frames type of (master) sampling for sampling sampling sampling frame survey design allocation Business Register Statistical Register of Employment VAT Database Annual Accounts selection method around legal units and local kind of activity units 15% of units are inactive some activities or addresses are wrong some local kind of activity units are missing Information about number of employees is added Information about turnover is added from the value added tax database The Annual Accounting Records of companies, sole proprietors, legal entities of private law, associations and legal entities of public law, collected by the Agency of the RS for Public Legal Records and Related Services (AJPES) 6

7 Business surveys at SURS type of (master) sampling for sampling sampling sampling frame survey design allocation sampling frame selection method It is important to construct a high-quality frame for the target population with: minor overcoverage or undercoverage, access to the units of the frame, auxiliary information, used for special sampling or estimation techniques. 7

8 Business surveys at SURS type of (master) sampling for sampling sampling sampling frame survey design allocation types of sampling selection method There are three main ways used for the surveying of enterprises: as a probability sample survey as a cut-off survey as an administrative based survey 8

9 Business surveys at SURS type of (master) sampling for sampling sampling sampling frame survey design allocation sampling designs selection method Usually stratified sampling design is used, with two strata variables: number of employees or turnover or both, activity classification (2, 3 or 4 digit Nace codes or sectors, depends on domains of interests) 9

10 Business surveys at SURS type of (master) sampling for sampling sampling sampling frame survey design allocation sample allocation selection method Assuming that sample size is determined: units of the biggest size are entirely included in the sample, proportional allocation is used most often for small units, when some auxiliary information is available optimal allocation is used. 10

11 (master) sampling frame Business surveys at SURS type of sampling for sampling sampling survey design allocation selection methods selection method Two most often used selection methods for probability samples: simple random sample without replacement selection method systematic selection method 11

12 Coordinated sampling The aim of coordinated sampling is to: minimize (negative coordination) or maximize (positive coordination) the overlap between samples. In order to coordinate several samples, the selection of a new sample depends on the samples previously drawn. 12

13 Coordinated sampling at SURS First coordinated samples at SURS were chosen in 2012 for 5 surveys based on probability selected samples. Only negative coordination of samples was desired. Results show that response burden of small units can be reduced by coordinated sampling. 13

14 Method of coordinated sampling used at SURS 1) Create common sampling frame 2) Assign permanent random number (PRN) from [0,1] to each unit and sort units by PRN 4) Choose fixed starting points: 0 x 1 x 2 x ) Select n 1 units with random numbers starting from x 1 in each stratum for the first survey, for the second survey n 2 units with random numbers starting from x 2 14

15 Simulations Simulations were done in order to see if the response burden was reduced. Sample size and its allocation among strata remained the same as in coordinated samples. Average burden of 100 stratified simple random samples without replacement for 5 surveys was compared to the burden of coordinated samples. 15

16 Number of units included in samples with probability less than 1* 2012 Independent samples Coordinated samples number of incl. units incl. uncertainly 0x incl. uncertainly 1x incl. uncertainly 2x incl. uncertainly 3x 58 4 incl. uncertainly 4x Independent samples Coordinated samples incl. uncertainly 4x incl. uncertainly 3x incl. uncertainly 2x incl. uncertainly 1x incl. uncertainly 0x * Large units have big impact on the estimates and must be included in samples. 16

17 Problems and challenges Systematic selection method is not possible. Sampling designs should be known in advance (to choose starting points more efficiently). How to include complexity of questionnaire? How to coordinate samples over the years? Updating the frame from observations in the sample and taking it into account for succeeding coordinated samples brings biases. 17