Cet ouvrage fait partie de la bibliothèque YouScribe
Obtenez un accès à la bibliothèque pour le lire en ligne
En savoir plus

Statistical uses of classifications: constraints and examples

8 pages
Classifications play an essential role in statisticians' everyday activity. In particular, classifications of activities and products have many practical uses—often implicit—in statistical operations. In reviewing these applications, we point out the reasons why adjustments and updates are needed when classifications change. Moreover, the implementation of new classifications, along with the necessary backcasting and extrapolations, takes place within a multi-faceted and constraining regulatory framework.
Voir plus Voir moins

Statistical uses of classifications:
1constraints and examples
! Jean Lienhardt*
Classifications play an essential role in statisticians’ everyday activity. In particular, classifications of activities and
products have many practical uses—often implicit—in statistical operations. In reviewing these applications, we point
out the reasons why adjustments and updates are needed when classifications change. Moreover, the implementation
of new classifications, along with the necessary backcasting and extrapolations, takes place within a multi-faceted and
constraining regulatory framework.
lassifications of activities and The APE code (principal c products are an integral part of economic activity) can
the basic organization of statistical serve as identifier in
sources. This article examines several business statistics
France’s official statistical service is
characterized by heavy interdependent
Classifications of activities operations, particularly in the field
and products are used of business statistics. We thus
to define survey coverage need to define relevant criteria for
and statistical indicators identifying statistical units and their
The “sectoral” eU statistical
regulations (i.e., concerning individual For this purpose, one of the most
economic activities) use nAce to Detail from cover of French Classifications of significant information items
Activities and Products (NAF Rev. 2 / CPF Rev. 2)define their scope of application. The associated with statistical units
change in data repository from one in individual data files is the APe
nAF version to the next has two code. The same is true of certain
immediate consequences: aggregated files: the APe code often
serves as the main identifier for a
– A technical consequence: the stratum or aggregate. We therefore Classifications are used
transposition of the regulation within aim to systematically check the to select samples
a constant scope code in databases and registers.
This verification determines the In business statistics, n AF is the – A political consequence: the
quality of information exchanges paramount criterion for most stratified negotiation on extensions and
between statistical files directly samplings. All registers and survey restrictions of coverage entailed by
involving aggregates—for example, launch tools must therefore be the change.
in “Intermediate enterprise immediately adjusted to meet the
Systems” (Systèmes Intermédiaires standards of a new classification. In practice, one can adjust the scope
d’entreprises: SIes).This notably concerns the French of coverage of statistical sources by
business register (SIRene) and related setting up new surveys or maintaining
registers, the survey launch files, and existing surveys for the sake of
sampling frames. The change in data continuity.
repository requires at least an update
A secondary consequence is of these basic tools.
the redefinition of the spheres of
competency of statistical agencies, Beyond, there may be challenging
* Jean Lienhardt is coordinator of Projects for also based on classifications. In France, methodological consequences. For
changes in Activity classifications at InSee.
this involves transfers of competency continuous surveys, which call for 1. o riginally published as “Les usages
statistiques des nomenclatures: contraintes et between ministerial statistical offices partial sample replacement, it is
exemples,” Courrier des statistiques (French
(Services Statistiques Ministériels: important to ensure consistency
series), no. 125, nov.-Dec. 2008, pp. 77-84,
SSMs). between samples. http://www.insee.fr/fr/ffc/docs_ffc/cs125m.pdf.
courrier des statistiques, english series no. 15, 2009 69
Source: InSeeJean Lienhardt
The APE code is used in
Box 1: Using NAF in automatic codingmany ways in statistical
production processes, in Many statistical operations use coding throughout 2008 with the headings
the application “engines” tools to transform respondents’ plain- collected in the declaration packages
language declarations of APe codes filed for new businesses (as the self-themselves
into classification codes. reported headings are not those of
the classification). The census and
A few examples: continuous labor-force survey will use Two variables are generally involved:
similar knowledge databases, compiled – principal economic activity (Activité
in summer 2008.– In the Annual enterprise Survey Principale exercée: APe) coded with
nAF(enquête Annuelle d’entreprise: eAe), For occupation coding, SIcoRe serves
– occupation coded with the to code declarations of occupation APe codes are determined from the
“occupations and Socio-occupational gathered in surveys or employers’ data gathered. The classification
categories” (Professions et catégories Annual Declarations of Payroll Data change requires a new algorithm for Socioprofessionnelles: PcS) system. (Déclarations Annuelles de Données
constructing the code. Sociales: DADS). SIcoRe is used in
InSee has developed a general- two ways. First, we include it in the
purpose automatic coding software – In many applications, the APe code post-collection processing sequence
program called SIco Re (French for “large” surveys and administrative enters into the definition of quality
acronym for “c omputerized System files. Second, we install it in interviewers’ criteria: minimum representativeness
for coding Survey Responses”). Pcs for “light-weight” surveys.
for a stratum of the Industrial SIcoRe uses “knowledge databases” Automatic processing is supplemented
Production Index (IPI); quality targets containing plain-language headings by operations performed by specialized
for the compilation of summary and their associated codes. It also teams at InSee Regional offices.
uses “ancillary variables,” i.e., statistics; quantitative criterion
information from the same documents For the population census, the in regulated operations to prepare
as the variable to be coded (SIRene continuous labor-force survey, and structural statistics. The classification
“declaration packages” filed by new surveys using the “common core” of update entails an adjustment of quality
enterprises, questionnaires, and so household surveys (Tronc c ommun
thresholds and criteria. on). These variables facilitate code des enquêtes auprès des Ménages:
determination thanks to decision tables, TcM), SIcoRe incorporates up to
which are sets of specific logical rules – The APe code is used recurrently fourteen ancillary variables, of which
for the variable to be coded. two are based on nAF:in coding-related questions: either
– the APe code at the most detailed directly, if the question concerns the
As nAF is used—directly or level of the classification (census,
APe heading in plain language, or
indirectly—in these coding operations, labor-force survey)
indirectly if the APe code is used as the effects of the classification change – a n AF2 variable reflecting the
an ancillary variable to code other are significant. divisions of the new classification
variables—for example, to determine (available in surveys using the TcM).
For APe coding, we need to adapt the socio-occupational categories (PcS)
knowledge databases to the headings The n AF change therefore requires in household surveys. Box 1 describes
and codes of the new classification. us to adapt—and, in most cases, to the procedures, which rely on SIcoRe
Such files are used for the business rewrite—the decision tables that use
software. The classification change register (SIRene ), the population these two variables.
naturally requires a redesign of the census, and the continuous labor-force InSee also transmits results under
related files (knowledge bases), a survey (eec). the ISco classification (see articles by
substantial task. In early 2008, we incorporated the Brousse and Torterat). The PcS-ISco
nAF Rev. 2 headings into the SIRene conversion table uses the detailed nAF
knowledge databases. We assigned breakdown, nAF2, and enterprise size.
new codes to the headings already The table, as well, was redesigned in
determined. We enriched these files 2008.Classifications are a tool
of choice for disseminating
and analyzing results
In See cannot, of course, confine references. Indeed, eU regulations APe often exhibit similar economic
itself to releasing raw data require Member States to disseminate behavior. n AF and the “summary
files—if only because of statistical data under the new categories. economic classifications” (see article
confidentiality rules. We therefore The purpose of this constraint is to by Madinier) were built specifically
prepare statistical results tailored to promote comparability and quality of to allow insightful descriptions of the
the classification categories. It is the workings of the French economy. In results.
role of dissemination applications to this field as well, the classification
construct these results. Databases and The APe variable is also a highly changes have a host of direct
tools for producing the disseminated relevant criterion in economic implications on the ways of viewing
products must be adjusted to the new analysis, as enterprises with the same economic facts.
70Statistical uses of classifications: constraints and examples
Statistical implementation
Box 2: A multi-faceted regulatory framework governs the classifications of NACE Rev. 2: a complex
and determines their implementationoperation requiring a
project structure
The regulatory framework is derived, ever-greater number of key technical
in the first instance, from european parameters of statistical operations
c hanging classifications in a commission regulation 1893/2006 of have been decided at c ommission
decentralized statistical system December 20, 2006, establishing the level.
such as the French official statistical new nAce.
For n Ace Rev. 2, the regulatory service is not a routine matter.
In addition to the list of nAce Rev. 2 corpus has been adapted in a total of Indeed, as early as June 2004, InSee
categories itself, the regulation fourteen specific statistical domains. deemed the operation sufficiently
includes many general instructions for They define a nearly complete program
challenging to commission “a
its implementation: date of introduction for coordinated implementation at
forward-looking analysis of the risks set at January 1, 2008, national eU level. However, they maintain
associated with the implementation extensions, “comitology” principles for uncertainty in several major statistical
of 2007-generation classifications of futures adaptations, and so on. and accounting areas, particularly
2activities and products.” agricultural statistics, balance-of-
The regulation also explicitly revises payments statistics, national accounts,
eleven statistical regulations in specific and even short-term surveys. In all these The report by InSee’s Inspectorate-
domains, setting the main parameters fields, specific regulations are being General, released in February 2005,
for nAce Rev. 2 implementation in prepared on separate timetables.highlighted some vital issues. It
each domain: timetable, redefinition of
stressed the need to:
scope, criteria for presenting results, eU regulatory arrangements are simpler
conformity criteria (quality, thresholds), in regard to the product classification
– participate actively in the eU and transition-related issues (double (cPA). Regulation 415/2008 of April
updating of legislation made obsolete reporting, backcasting). 23, 2008, merely establishes the new
by the changes in classifications, cPA and defines its general uses.
This regulatory framework was The application measures will therefore particularly in summary classifications
supplemented by a second set of be decided through discussions in at international level
measures taken under european individual economic sectors. note that
c ommission regulation 973/2007 of the c PA regulation allows national – align the management of
August 20, 2007. These measures extensions, but France has decided
classification changes in France with
are important for, in recent years, an not to avail itself of the option.
the eU timetable
– strengthen coordination
arrangements in the “classification
changes” project and document the 2008 must be available under the production time for an international
process. classification had been considerably new classification. For structural
shortened. Table 1 summarizes the business statistics, France was a year
These recommendations served key dates.ahead of schedule and took steps to
as guidelines for the gradual disseminate initial 2007 results under
implementation of the new system – However, one difference between nAF Rev. 2 by the second half of
3starting in 2006. the present situation and conditions 2008.
prevailing in 1993 concerns the eU
As a direct result of this review, the context. Most statistics prepared – May 2011: publication of new
national timetable for n AF Rev. 2 today conform to eU regulations, national-accounts base under nAF
implementation has been organized which was not yet the case in Rev. 2 (reference period: 2008-2010).
around four main dates: 1993. There is no doubt that the
commitments arising from this To the extent that one can compare
– January 6, 2008: SIRene framework are a powerful incentive the present experience with the 1993
crossover. to abide by the timetable.
crossover—the switch from nAP to
n AF was similar in scope to the – First quarter 2009: publication of
transition from nAF Rev. 1 to nAF first short-term indicators under nAF 2. The eU chose the term “operation 2007”
Rev. 2—the following remarks are in to designate all the classification changes Rev. 2.
implemented on January 1, 2008.order:
3. o ur article deals specifically with the
– october 2009: transmission to classification implementation phase. The design
phase is discussed in the article by T. Lacroix – The implementation of the new eurostat of first results under nAce
in this issue, as well as in J.P. Grouthier and
classification in statistical operations Rev. 2 for structural statistics covering J.P. Lachize, “Le changement de nomenclatures
must proceed at a substantially faster d’activités et de produits dans l’industrie,” 2008. Generally speaking, the
Courrier des statistiques (French series), no. 125,
regulation states that annual statistics pace than in 1993. It has been noted
nov.-Dec. 2008, pp. 77-84, http://www.insee.fr/
on economic activities performed in (see article by T. Lacroix) that the fr/ffc/docs_ffc/cs125k.pdf.
courrier des statistiques, english series no. 15, 2009 71Jean Lienhardt
Flowchart for spread of new classifications in French official statistical service:
main operations and scheduled crossover dates
Business-Register SIRENE partners Processing of 2006 EAEs Industry surveys
Improvement RAMSESE Public-sector sphere (intermediate classification) gradual adaptation
Surveys (EARs) FINESS Agriculture sphere since January 2007

AMELI TSE2G - y surveys

(EMBs): Feb. 2009

- annual surveys (EABs):

June 2009
SIRENE register crossover (Jan. 6, 2008)

Related services ODISSEE, online services, SIAM (Jan. 2008)

Crossover tables and files (Feb. 2008)
OCEAN (sampling frame) CITRUS

(Jan. 2008) (March 2008)

OCEAN ACEMO UNEDIC Soc. sec. agencies EAE 2007

DARES Q4 2008 URSSAF / Q3 2008 (end-2008)
(Feb. 2009) (Oct. 2008) IPVI-S Jan. 2009

(March 2009)

Data on
DMMO Jan. 2009 EPURE ICT Q1 2009 DGI

temp work

EMMO Q1 2009 Q1 2009 (June 2009) (in 2009)
CA3 indices I PI J an. 2009

Jan. 2009

(June 2009) (May 2009)

(March 2009) ICOM J an. 2009
(Apr. 2009)


APE(T) coding
ACEMO Employment estimates DADS-U 2008 SUSE 2007
in surv eys and

annual Annual at Dec. 31, 2007 (July 2009) (Sept. 2009) (Apr. 2009)
(2009) Quarterly Q2 2009 (Aug. 2009) SIE 2007
(May 2009)

(Jan. 2009)

Income statistics ESANE 2008 Short-term surveys

EEC Q1 2009
(from May 2010)
ECMOSS 2009 (Jan. 2010)
(March 2010)
(Feb. 2009)
(in 2010)

Census: 2009

annual survey

(May 2009)

- Provisional 2010 account

- Semi-final 2009 account (in May 2011)
Quarterly accounts
- Final 2008 account
base 2005 of quart. nat. accts

(in May 2011)
- Base 2005 of national accounts

Note: for a key to all abbreviations, see bilingual list of abbreviations in this issue (to access online version, return to Contents page)

2007 Statistical uses of classifications: constraints and examples
These constraints come on top of (www.insee.fr) were of great help—an process must also take account of
the need for continuing dialogue asset that was not available to our the other parameters involved in
between some eighty applications predecessors in 1993 (see article by the change in the accounting base,
and statistical operations and several Dutrech). namely, changes in concepts and in
hundred people, responsible for 350 the content of statistical sources.
basic tasks at In See and in ministerial – For product reclassification,
– The flowchart tells us nothing about statistical offices (SSMs). For all these we revised the industry surveys
how long it will take to integrate the reasons, France’s official statistical (Enquêtes de Branche: EDBs) when
new classification into a given source. service decided to set up a closely submitting the 2007 and 2008 waves
The operation may consist in a “big coordinated and largely decentralized for official approval. Revisions began
bang” (as with SIRene ), a multi-year project structure. in July 2006. The surveys are an
program (as with the industry surveys, essential resource for compiling many
revised in a three-stage process), short-term indicators and statistics.
or a gradual integration of the new NAF Rev. 2 deployment
– We released the first eAe-based classification in the successive in official statistical service
results for individual industries in 2008. flows constituting the source. The
Last year was also largely devoted to latter scenario requires a transitional Table 2 shows a simplified timetable for
preparing the classification change arrangement with double coding. The the incorporation of the classification
source crossover is not complete for short-term indicators. Most of changes in statistical operations.
until the last flow is incorporated. For these will be disseminated starting in These operations are located at many
instance, the continuous labor-force March 2009 (monthly indicators for different points in the official statistical
survey will complete the crossover January 2009) or June 2009 (quarterly service.
in June 2009 but will be ready under indicators for Q1 2009). The change
n AF Rev. 2 for 2008 results. Similarly, in classifications will coincide with The exhaustive inventory prepared
the change in the indicators’ five-year in early 2007 found approximately
base.eighty such points. The diagram Table 1
showing the diffusion of the new
c lassification – In June 2009, In See will also publish classifications in the official statistical
change in: 1993 2008the first employment estimates under service completes our description
n AP-n AF1 n AF1-n AF2the new classification. As with the (Flowchart). The flowchart is arranged switch switch
short-term indicators, the preparation according to the four main dates
SIRene of these data involves considerable mentioned and indicates the technical
business January January
backcasting, a prerequisite for the or logical dependency relationships register 1993 2008
crossoverpublication of results.between statistical operations.
May May– eU institutions have settled on of national-Let us recall the key stages in the 1999 2011
accounts baseMay 2010 as the crossover date for process.
short-term surveys of businesses Process
duration 6.3 years 3.3 years– o n January 6, 2008, the business and households. Two years (2008
register (SIRENE) switched to n AF and 2009) will hardly be too long
to prepare new, robust “balance of Rev. 2. The transition was prepared
4 opinion” series.throughout 2007. The Annual
Business Survey (EAE) executed
– eU institutions also eventually in 2007 to collect 2006 data played
decided that EU national accounts a special role in this preparation (see
should be available by September article by Pinel).
2011 under the new classification, and
French annual and quarterly national – The switchover is critical to n AF
Rev. 2 implementation for two accounts by May 2011. This choice
reasons. First, the SIRene register is a compromise between proposals
plays a key role both administratively for earlier dates (such as the one
and statistically. Second, the number for France) and the desire of certain
of economic agents impacted by organizations to await the next revision SIRENE logo
the APe code change is substantial. of the System of n ational Accounts
The process was backed up by an (Sn A), i.e., after 2012. However, the
4. For a description of the process, see information campaign prepared very method of construction of the
Bernadette Rocca, “Le changement de
as early as o ctober 2006. The high national accounts, whose annual data nomenclature dans le répertoire Sirene: un
travail au long cours,” Courrier des statistiques point of the campaign was the direct sets rely on one another in three-year
(French series), no. 125, n ov.-Dec. 2008,
notification of enterprises concerned moving periods, makes it difficult to
pp. 61-66, http://www.insee.fr/fr/ffc/docs_ffc/
in January 2008. The new technologies agree on a specific schedule. The cs125j.pdf.
c ourrier des statistiques, english series no. 15, 2009 73
Source: In SeeJean Lienhardt
Table 2
(Simplified) timetable for classification changes
Preparation for SIRene business-register crossover
2006 Annual enterprise Survey (eAe) under “intermediate classification” (disseminated
under nAF Rev. 1 end-2007)
Business-Register Improvement Surveys (eARs)
January 6:
SIRene crossover; enterprises and local units (establishments) classified under new
classification, nAF Rev. 2
2007 eAe: disseminated under nAF Rev. 2 (except manufacturing) in H2 2008
Prodcom 2008 (annual production survey): disseminated under nAF Rev. 2 in H1 2009
Early 2009:
Monthly short-term statistics
Short-term employment statistics
2009 monthly industry survey (eAB): collected and disseminated under nAF Rev. 2
enterprise demography
June 2009:
Quarterly short-term statistics
2009 population census logo
employment estimates
continuous labor-force survey (eec) (2008 and Q1 2009 results)
Summer 2009:
“Total coverage” structural statistics for 2007. Here are some examples that illustrate
the complexity of these technical May 2010: surveys of current economic conditions
Mid-2010: DADS, ecMoSS constraints:
May 2011:
2011 Annual and quarterly national accounts; – The calendar for preparing the
new accounting base published under nAF Rev. 2 SIRene switchover made 2006 a
special year. The 2006 eAe became Population census: continuous integration of nAF Rev. 2 until end-2012 (but first results
2012 under nAF Rev. 2 in H2 2009) the pivotal year for structural business
statistics compiled under a dual
classification (readable under nAF
Rev. 1 and nAF Rev. 2) (see article the population census will be able of the backcast-series offering to
by Pinel).to use nAF Rev. 2 by 2009, but the the national council for Statistical
transition will last until end-2012. Information at a fairly early stage of
– At the same time, SIRene 5nAF revision.
administrators took steps to offer
– The flowchart (which is, in part,
a register accessible via both
something of a forecast) also tells us little The nature of the series to be
classifications by early 2008 and
about the many round trips between reconstructed and the length of
December 31, 2008, the latter date
the statistical sources involved; these the backcasting period are among
being applied in an internal procedure
“dialogues” between sources are the regulatory parameters imposed
to applications scheduled for
due to the interdependence between by eurostat, at least for the main
crossover in early 2009.
successive years and to calibrations. statistical domains. The regulations
This interdependence requires a on structural statistics and on
– Although comparability over time
reconstruction of the missing links, short-term statistics both contain is not the main goal of structural
notably through backcasting. such recommendations. Many analyses, most of these sources
other eU regulations and additional require multi-year operations to
provisions introduce requirements achieve consistency. For example, Classifications: why that necessitate backcasting. o ne the 2007 “Intermediate enterprise backcast? notable example consists of the System” (Système Intermédiaire
general agreements on national d’entreprises: SIe), build under the
The issue of backcasting is
accounts. new classification, must rely on
inseparable from that of classification
the 2005 and 2006 SIes, compiled
changes. Indeed, it is the issue of Taken as a whole, these measures under the old classification, which
most immediate concern to faithful define a minimum standard. But for therefore need to be backcast. A
users of statistical products. For this
a wide variety of technical reasons similar observation applies to annual
reason, InSee presented an overview
specific to the decentralized enterprise surveys (eAes).
organization of France’s official
statistical service (SSP), significant – c onversely, special care will be
5. See cnIS document entitled Perspectives
room for maneuver remains at national needed for the maintenance of
en matière de séries rétrospectives, october
5, 2007. level. statistical sources under the old
Source: InSeeStatistical uses of classifications: constraints and examples
classification on which national
Box 3: Backcasting – example of short-term activity indicatorsaccounts will rely until early 2011
(i.e., well beyond the crossover date
Short-term statistics are subject to SIRene-based conversion files were for these sources). That is the case
an eU regulation. InSee has been available by early 2008 (VAT).for SIe, for which the years 2007 and
producing them under nAF Rev. 2 62008 will need to be “extrapolated.”
since January 2009. The classification Backcasting methods reflect two
The same is true of most short-term change is associated with a five-year approaches: a “micro” approach and a
indicators.in the base (base 2005). “macro” approach:
– The “micro” approach consists in
– Given this maintenance requirement, backcasting each basic component Three activity indicators—IPI, Ico M
we should note that a postponement of an index at the most detailed (index of orders booked), and VAT
level possible. We can, for instance, (sales indices)—were backcast in five of the national-accounts conversion
backcast data at the “enterprise” level, stages up to their publication in March would have a significant cost impact
then aggregate them. This method 2009:on the entire sequence of operations
relies on actually observed data, but – Q1 2008: preliminary operations: 7upstream. We took this cost into requires thorough consistency checks definition of new boundaries of control
account when preparing the tentative on the aggregated results.series in IPI and IcoM rebasing;
schedule. – The “macro” approach consists in planning for a new method to determine
backcasting at a relatively aggregated final sales indices.
level. Its advantage is that it is “lighter” – Another long-standing requirement – Before June 2008: calculation of
as it requires a smaller volume of data concerns short-term sources, such index weightings (IPI, sales index) for
to process. But it rests on the heroic base 2005. This requires the use of as the industrial production index
assumption that the switch from one annual business statistics for 2005 (IPI) and short-term surveys. These classification to the next occurs in a (SUSe file).rely on weightings determined from uniform manner over the long run.
– July-october 2008: backcasting of structural statistics under the new
indices.8classification. For reasons of timing The choice of method depends
– n ovember-December 2008: validation precisely on the data availability consistency, the structural statistics
of aggregated results. period:are backcast.
– november 2008-February 2009: – For VAT indices, data at enterprise
start-up of new computer applications. level are available from January 2004
– The backcast data described above on. We can therefore perform “micro”
are largely used in file-construction We defined a strict timetable, particularly backcasting for the period 2004-2008.
processes. These “intermediate” for the actual backcasting operation. However, this requires heavy processing,
Many procedures had to be performed data are not intended for wide so “macro” backcasting over the period
in advance. This was possible thanks 1998-2007 is preferable.dissemination.
to the fact that certain industries do – For IPI and IcoM, we shall use
not vary from one classification to “macro” backcasting, as individual – Some operations draw on data
the next (IPI) or to the fact that the enterprise data are not available.
gathered over several collection
periods. They too require backcasting
to ensure internal consistency. This
involves recoding data stocks in the
meaning, all short-term series will for the source or statistical survey
Labor-Force Survey, reinitializing
be backcast. The length of these concerned. The method is costlier
cohorts in the “Information System
but yields higher-quality results. We backcast series will vary in proportion
on new enterprises” Survey (Système
chose it for structural statistics (see to the intensity of their use in the
d’Information sur les n ouvelles
article by Pinel).national accounts. Ultimately, short-
entreprises: SIne), consolidating
term surveys and quarterly national
labor-cost and wage-structure – With statistical backcasting, we accounts are rooted in the analysis
surveys, and so on. can determine aggregated results of temporal regularities. For all these
from overall correspondence tables. sources, In See will disseminate – closer to the needs of data
This method may suffice for certain backcast data at the same time as end-users, some operations seek
data under the new classification.to prepare consistent statistics
immediately on publication. For
6. We use “extrapolation” here to denote the
employers’ Annual Declarations of reconstruction under the former classification
of data initially compiled under the new Backcasting instrumentsPayroll Data (Déclarations Annuelles
classification, a process based on pre-
de Données Sociales: DADSs), for switchover data.
There are two distinct methods for example, the results published cover 7. A delay was needed for the 1993 classification
change; but that is not, of course, a working backcasting:two consecutive years, necessarily
hypothesis for the current change.
under the same classification. 8. Strictly speaking, some “base 2005” short-
– Backcasting by individual term indicators in n AF Rev. 2 should be
weighted by data derived from the national
– Intrinsically, because it is the time imputation makes it possible to
accounts using the same base. This reweighting
dimension that gives them their recode all units of the microdata file is, of course, impossible.
courrier des statistiques, english series no. 15, 2009 75Jean Lienhardt
sources when the desired quality Table 3
level (expressed at n AF “division”
Availability of backcast series: selected dates (forecasts)or “group” level) is not essential.
The case of short-term indicators The combination of regulatory requirements and national choices and technical constraints have
resulted in the following tentative schedule for the main statistical operations (start of backcasting illustrates an alternative use of the
period and publication dates):
two methods (Box 3).
start of backcasting publication date
IPI & IcoM (orders booked) January 1990 March 2009
The best correspondence table,
IcT (labor-cost index) Q1 2000 June 2009
naturally, is SIRene itself. The
VAT (sales index) 1995 April 2009
register is a file built from actual IPVI (industrial prices) 1995 March 2009
data, with exhaustive coverage of IPVI-S (service prices) 2000 March 2009
the broadest possible field. However, IVU (unit value index) 1994 March 2009
the file is not perfect. The situation customs statistics January 1999 March 2009
Aggregated employment estimates 1970 September 2009prior to end-2007 is not always well
Detailed estimates 19902009known, for some APe code updates
eSTeL (annual) 1989 July 2009under nAF Rev. 1 have not been
AceMo (earnings) Q4 1998 February 2009incorporated. Moreover, at January
AceMo (annual) 1999 variable
6, 2008, we did not always have
eMMo 1996 Q3 2009
conclusive information for assigning
DMMo (detailed & aggregates) 1996 & 1983 March 2009
the correct nAF Rev. 2 code. For lack
Labor-Force Survey Q4 2007 June 2009
of a better alternative, we therefore census 2004-2008 cycle Q3 2009
had to use an imputation model that DADS 2007 April 2010
minimizes the risk of assigning the ecMoSS (earnings & cMo) 2006 July 2010
wrong APe codes under nAF Rev. 2 ePURe (aggregated series) Q1 2004 June 2009
ePURe (microdata files) 2007 June 2009to the enterprises concerned.
Data on temporary work January 1995 July 2009
R&D Survey 2001 H2 2009In early 2008, InSee released two
Annual enterprise Survey 2003 H2 2008aggregated tables, one built on
SUSe-SIe 2005 H2 2008
SIRene, the other on the 2006 eAe.
LIFI-RecMe 2006 May 2009
The tables “count” the switches from
enterprise demography 1993 Starting February 2009
nAF Rev. 1 to nAF Rev. 2 observed
SIne survey (partial) 1994 H1 2010
during the implementation of the two
Short-term surveys variable May 2010
sources. The first table can serve as Annual national accounts 1959 May 2011
an additional resource for individual Quarterly 1978 May 2011
coding. The second, which takes
into account both the number of
enterprises and the number of potential switches from nAF Rev. 1 Most of these instruments (files and
to nAF Rev. 2 assessed from the tables) have been disseminated in the employees, should be useful in all
category definitions. This is called official statistical service or as part statistical backcasting operations.
an analytical table. Lastly, we shall of In See ’s commercial publication
We developed a third type of table program, starting in February 2008, be able to design special tables for
after an active promotion campaign. nbased solely on the logical content certain applications such as enterprise
of the classifications, expressing the demography.