Statistics Sweden replaces traditional collection of price data with scanner data for samples of outlets and products in some parts of the consumer price index. A fixed basket approach is used, where the computing of indices is equivalent to the computing methods applied within the previous production system of the national CPI. The use of scanner data has lowered the collection costs with annuals savings of approximately 80 000 EUR per year. In addition, the sample sizes could be increased and the standard errors of index could thereby be decreased at a low cost. About 2 000 unique products are included in 90 product groups each year, and some 100 000 price observations from scanner data are used in the Swedish CPI each month http://www.istat.it/en/archive/168897 http://www.istat.it/it/archivio/168890

1. facebook.com/statisticssweden @SCB_nyheter
The Swedish experience with
scanner data
From sampling to index calculation
Paulina Jonéus

2. COICOP
 COICOP divisions where we use mainly
scanner data
• 01 (daily necessities excl. perishable fruits, vegetables and meat sold
in department stores, supermarkets and hypermarkets)
• 02.2 (tobacco)
• 02.1 (alcoholic beverages)
• 06.1 (medicines)

3. COICOP
 COICOP divisions where a combination of
scanner data and traditional price collection
is used
• 05.5 (lamps and batteries)
• 05.6 (household cleaning products)
• 09.3 (food for domestic animals, soil and nutrient for plants)
• 12.1 (personal hygiene products)

4. The data
 Former monopoly of pharmacies
• Reporting to the Ministry of Finance
• Scanner data since 2010
 Monopoly of liquor stores
• A government owned chain of liquor stores in Sweden
• Index based on scanner data. From 2016, scanner data
to Statistics Sweden

5. The data
 Three chains of daily necessities
• Account for more than 80 % of the consumer market
• Sample from two categories: supermarkets /smaller
markets and hypermarkets
• Scanner data since december 2011
 From five sources we get 19 percent of CPI

6. The Swedish approach
 Replace the manually collected price data with scanner
data for the sample of outlets and products.
• Annual savings 80 000 EUR per year for daily
necessities
• A cheap way to increase (more than double) sample
sizes
• Possible to manage substitutions in the ever changing
market
• Quantity adjustments
• Homogeneous product groups

7. The Swedish approach – sampling
of outlets
 A stratified, sequential Poisson sampling method
Stratum description
Industry
(NACE)
Sample
size Net Collection
method
Hypermarkets, broad assortment 47111 7 Scanner data
Hypermarkets, broad assortment 47111 9 Visit
Supermarkets with broad assortment 47112 52 Scanner data
Supermarkets with broad assortment 47112 32 Visit
Tobacconists 47260 5 Telephone
Health food shops 47291 5 Telephone
Pharmacies 47730 9 Visit
Pet shops 47762 3 Telephone

8. The Swedish approach – sampling
of outlets
 Covering the three chains proportionally to
their market shares
 For scanner data from the supermarkets
and hypermarkets a sample of some 60
outlets is used
 For pharmacies and the liquor stores, data
is used for all outlets

9. The Swedish approach – sampling
of products
 Since 2001, SCB produces Food Sales in the
trade, based on scanner data
• Annual data show total annual sales values per
article (GTIN) during last calendar year
• A comprehensive mechanical coding of the
different articles into statistical product groups
 The register is used as sampling frame for product
sampling for the CPI

10. The Swedish approach – sampling
of products
 Automatic coding is made independently for all
retail chains, using nomenclatures of each retail
chains
 A SAS script searches for parts of product names
to improve the automatic coding
 The files with preliminary codes for the chains are
joined by GTIN, making comparisons of product
group code possible.

11. The Swedish approach – sampling
of products
 Within these groups three annual samples of 800
very narrowly defined products are selected
 A sequential Pareto πps selection within strata
 About 2000 unique products
 Some 100 000 price observations from scanner
data are used each month

12. The Swedish approach – sampling
of products
 Sampling frame is based on annual sales from t-2
 Updated each year with information from late
autumn market analysis
• No over coverage. Disappearing products are removed.
• Possible under coverage. Sometimes hard to find
replacement products.

13. The Swedish approach – temporal
sampling
 Scanner data is received for all weeks on a weekly
basis. For the supermarkets the data for the full
middle three weeks is used
 The monthly average price is calculated as a
quantity weighted arithmetic average
 For the pharmacies an (quantity weighted) average
price is calculated using all data from the 25th of
last month to the 24th the current month

14. The Swedish approach – the
production month
 1st stage: Initiating a production month
 2nd stage: Product life analysis
 3rd stage: Checking of the scanner data set
 4th stage: Select the data for the product-offer
sample
 5th stage: Aggregate prices over three weeks
 6th stage: Send data to the CPI production system

15. Elementary indices
 The quantity weighted arithmetic average of
weekly average prices is computed in a special
SAS-based module
 Elementary indices is then calculated in the
production system, as for all other product groups,
by Jevons index

16. Summary
 A reasonable amount of data at GTIN level
 Take into account differences between outlets
 Narrowly defined products
 Quantity adjustments
 Manage quality changes
 Can be used in combination with manually
collected prices

17. Questions?