A Useful Addition to SDMX?
Steven Vale (United Nations Economic Commission for Europe)
Alistair Hamilton (Australian Bureau of Statistics)
SDMX has typically been seen as providing a means for standardising and facilitating data exchange between statistical organisations. With this purpose in mind, the focus has therefore been on developing data structure definitions that are specific to individual statistical domains, resulting in the inclusion of a classification of statistical subject-matter domains within the Content-oriented Guidelines. Whilst this approach clearly remains important, it may not, however, cover all exchange requirements.
Over recent years, a number of statistical organisations have started to move away from a traditional domain-based organisation model, to a much more process-oriented approach. Instead of being responsible for a single survey or other statistical output, teams of staff are now working on specific parts of the production process (e.g. data collection) across a wide range of surveys or outputs.
In parallel, there have been recent attempts in several statistical organisations to formalize statistical business process models, showing the different phases of statistical production, and the data and metadata flows between them. This has led to proposals for a generic model, applicable across all statistical organisations, which could be used to inform decisions not only on organisational structure, but also on the design and evaluation of statistical business processes, as well as on information systems architecture. The implementation of a generic model makes it possible to envisage the development of generic software tools for each process, and also increases the feasibility of sharing these tools across organisations.
An essential pre-requisite for this approach is a standard means of communication (data and metadata exchange) between tools and processes, a role for which SDMX seems ideally suited. This paper outlines recent progress towards developing a "Generic Statistical Business Process Model", and proposes that, when finalised, this model would form a useful addition to the Content-oriented Guidelines, providing a process-oriented view to compliment the existing subject-matter view. This would help to widen the applicability of SDMX by providing a basis for defining data and metadata structure definitions for exchange between processes, as well as between organisations.
The Joint UNECE / Eurostat / OECD Work Sessions on Statistical Metadata (METIS) have, over the last few years, been preparing a Common Metadata Framework (CMF). Part C of this framework is entitled "Metadata and the Statistical Cycle" This part refers to the phases of the statistical business process (also known as the statistical value chain or statistical cycle) and provides generic terms to describe them. The first draft of a "Generic Statistical Business Process Model" (GSBPM) was presented by the UNECE Secretariat and Statistics New Zealand at the METIS Work Session in Luxembourg in April 2008. Following two rounds of comments, a third draft has been prepared as an input to the METIS Workshop to be held in Lisbon on 11-13 March 2009, where it is intended that the GSBPM will be finalised.
The original intention was for the GSBPM to provide a basis for statistical organisations to agree on standard terminology to aid their discussions on developing statistical metadata systems and processes. However, the model has also been shown to be relevant in other contexts such as harmonizing statistical computing infrastructures, facilitating the sharing of software components, and providing a framework for process quality assessment. This paper explores other possible applications in the context of the development and implementation of SDMX.
The idea behind modelling the statistical business process is not new, the GSBPM builds on work in several national statistical organizations, as well as existing models and standards such as that proposed in the 1999 United Nations guidelines "Information Systems Architecture for National and International Statistical Offices", and the Combined Life-cycle Model developed under the Data Documentation Initiative. The GSBPM is, however, the first attempt to pull these different strands together to create a truly generic, multi-purpose model for the production of official statistics at the sub-national, national and international levels.
The GSBPM is intended to apply to all activities undertaken in the production of official statistics, at both the national and international levels. It is designed to be independent of the data source, so it can be used for processes based on surveys, censuses, administrative records, and other non-statistical or mixed sources. Whilst the typical statistical business process includes the collection and processing of raw data to produce statistical outputs, the GSBPM can also apply to cases where existing data are revised or time-series are re-calculated, either as a result of more or better source data, or a change in methodology. In these cases, the input data are the previously published statistics, which are then processed and analyzed to produce revised outputs. In such cases, it is likely that several sub-processes and possibly some phases (particularly some of the early ones) of the GSBPM would be omitted.
As well as being applicable for processes which result in statistics, the GSBPM can also be applied to the development and maintenance of statistical registers, where the inputs are similar to those for statistical production (though typically with a greater focus on administrative data), and the outputs are typically frames or other data extractions, which are then used as inputs to other processes.
Some elements of the GSBPM may be more relevant for one type of process than another, which may be influenced by the types of data sources used or the outputs to be produced. Some elements will overlap with each other, sometimes forming iterative loops. The GSBPM should therefore be applied and interpreted flexibly. It is not intended to be a rigid framework in which all steps must be followed in a strict order, but rather a model that identifies the steps in the statistical business process, and the inter-dependencies between them. Although the presentation follows the logical sequence of steps in most statistical business processes, the elements of the model may occur in different orders in different circumstances. In this way the GSBPM aims to be sufficiently generic to be widely applicable, and to encourage a standard view of the statistical business process, without becoming either too restrictive or too abstract and theoretical.
The GSBPM is divided into nine phases, each with a number of sub-processes. It also recognizes twelve over-arching processes that apply throughout the nine phases, and across statistical business processes. These over-arching processes can be grouped into two categories, those that have a statistical component, and those that are more general, and could apply to any sort of organisation. The first group are considered to be more important in the context of the GSBPM, and within that group, particular emphasis is put on the processes of metadata and quality management. The second group should also be recognized though, as they can have (often indirect) impacts on several parts of the model.
The diagram below shows the nine phases and associated sub-processes of the model, which are described in more detail in Annex B.
The Generic Statistical Business Process Model
As per the following simplified illustration, the GSBPM and the SDMX List of Subject-matter Domains can be visualised as forming a matrix, where the phases relate to "end to end" workflow processes and the subject matter domains provide a "side to side" perspective on the different statistical activities undertaken by a national statistical organization (NSO).
Only a few example Statistical Subject-Matter Areas are illustrated above.
As suggested by the heavily simplified diagram, while the different subject-matter domains, and their outputs, can be recognised within an agency, the extent to which they are completely independent activities in terms of their end to end processes is decreasing. It is more and more the case that infrastructure such as the following are used in common by multiple subject-matter domains:
- a single collection design and approval process;
- statistical methodologies (for imputation, confidentiality treatment etc);
- specialist support services (e.g. question and form design experts, data collection services);
- a single data collection process (e.g. a single survey or a single administrative data source) supporting multiple subject-matter purposes;
- definitional metadata (classifications, data elements);
- common input, intermediate and output data repositories;
- common processing systems
It is seldom the case that in the end to end context everything is either "in common" or everything is "completely separate" across subject matter domains. The pattern changes for individual phases, and even from step to step within a phase.
In order to achieve a deeper understanding of disseminated/shared outputs, and to assess their fitness for a particular purpose, it is often essential to have a summary of the salient aspects of the statistical processes that lead to the outputs available along side the outputs themselves. If/where this is to be done, it becomes clear that some common high-level way of structuring that information is important. This is somewhat analogous, but as a separate dimension, to the way existing SDMX content oriented guidelines propose the Statistical Subject-Matter Domain classification be used as a high level structure for organising data and metadata for various purposes.
The fact that these high level categorisations are independent but intersecting is important to recognise. As mentioned earlier, a lot of the high level information about processes, methodologies etc will be in common across multiple subject matter domains (although operational details may vary and warrant separate description). Creating and maintaining completely separate, unlinked, versions of high level process information which is in common across multiple subject matter domains would be both resource intensive and error prone.
The fact that there is an independent but intersecting relationship between business process phases and SDMX Statistical Subject-Matter Domains does not in itself necessarily make GSBPM highly relevant to SDMX. It can be noted, for example, that although the latest SDMX User Guide contains a chapter relating SDMX to statistical business processes, the SDMX initiative has progressed a long way already without incorporating either GSBPM or an equivalent construct related to statistical business processes and phases.
This may relate to the fact the SDMX initiative to date has focused on dissemination and sharing of content beyond agency boundaries. The focus has therefore, until now, been on the outward data and metadata flows shown in Phase 7 (Dissemination) of the above diagram.
It is clear that at that specific point the dimension of subject matter domains is much more important in terms of differentiation of the various outward data and metadata flows. All of these flows relate to a single common phase in the GSBPM.
As noted above, however, even if the only role for SDMX related to those outward facing arrows there would still remain value in the GSBPM providing a high level categorisation of information about the statistical processes that led to those outputs.
The importance of such information is recognised in the existing Euro-SDMX Metadata Structure (ESMS). The ESMS was developed to support the European Statistical System, and "contains a description and representation of statistical metadata concepts to be used for documenting statistical data and for providing summary information useful for assessing data quality and the production process in general". Section 20 of the ESMS focuses in particular on statistical processing.
The ESMS is "not be meant as a single comprehensive source of information". Metadata structured around the GSBPM could easily provide an additional level of detail on statistical processing.
One of the additions to the SDMX V2 package structure compared with SDMX V1 is the "process" package. The SDMX V2 framework states:
“The process class provides a way to model statistical processes as a set of interconnected process steps. Although not central to the exchange and dissemination of statistical data and metadata, having a shared description of processing allows for the interoperable exchange and dissemination of reference metadata sets which describe process-related concepts.”
Reference metadata can be provided for processes and for process steps just as it can be provided for other objects. Under current circumstances, "commonly used" names and descriptions of processes may be quite specific to an agency and hard to interpret accurately outside that agency. If the SDMX V2 process package is to be used effectively in the manner proposed in the technical standards, there appears to be significant value in the application of the GSBPM as a common reference model. This might be achieved by agencies either choosing to name and/or describe processes and process steps based on the GSBPM or by them providing reference metadata or mappings that place "local" processes and process steps within the context of the GSBPM as a high level common framework.
A significant initial driver for SDMX was that members of the SDMX Consortium sought to receive data from other agencies in a richer, more consistent and more useful form than was previously the case. Many of the consortium members, having received data from other agencies, then perform various processing upon those data prior to disseminating the content in whole or in part. The processing may be as simple as compilation and combined presentation, but it often includes derivation of aggregates (various totals and subtotals) and sometimes derivation of new indicators/measures etc. Aspects of quality checking of data received, analysis and commentary etc, may also be required.
In other words, these agencies are receiving content in SDMX form as part of Phase 4 (Collect) of the GSBPM and then stepping through the subsequent phases. Consistent with their sponsorship of the SDMX initiative, more and more commonly at Phase 7 these consortium members are also disseminating content in SDMX form.
A number of agencies, both within and outside the Consortium, who collect (or are planning to collect) most of their inputs in SDMX and also to disseminate using SDMX have, very logically, considered the advantages of working with content in SDMX form during the intermediate phases. Eurostat's thinking around SODI (SMDX Open Data Interchange) is a prime example in this regard.
Having "internalised" SDMX in this manner, the SDMX data and metadata flows no longer relate to just a single phase in the GSBPM. It becomes necessary to manage the progress of the SDMX based content through (at least a subset of) GSBPM phases, including managing and recording the processing applied to the content, and the results of that processing. The GSBPM provides a simple common framework for that. Given these processes are now interacting with content held in an SDMX based form, it makes a lot of sense that the relationship between the processes and the content should be described using SDMX, e.g.
- what processes, in what order, need to be applied to which SDMX datasets;
- what attributes recorded for a particular SDMX dataset, or one of its components, might impact the way a particular process operates on that dataset.
Under a model where content is collected using SDMX agencies could, in fact, be seen as undertaking earlier phases in the context of SDMX also. For example, during the phases related to specifying needs and design, the content to be collected and disseminated, including the structures and concepts related to it, should be based on engagement with the relevant (subject matter) "domain group(s)" and on maximum use of relevant domain specific (and cross domain) concepts, data structures and metadata structures. The Build phase would relate to assembling and testing infrastructure to capture and work with content in an SDMX based form.
SDMX is a highly collaborative initiative. With such agencies seeking to share information on what they are doing, or planning to do, in regard to their "end to end" use of SDMX it becomes more important and more useful to have a shared high level reference/terminology model for referring to "end to end" business processes and the phases within them.
While SDMX consortium members often collect content at an aggregate level from a relatively small number of agencies which are expert in, and committed to, compiling and providing data, the situation is typically very different for NSOs. NSOs may collect data directly from a vast number of individual businesses and households and/or need to collect data related to these individuals via administrative records that are far from ideal for statistical purposes. This tends to make the sub-processes between the Collect and the Disseminate phases much more complex and "heavy duty". Over the years NSOs have assembled an array of specialised processes and processing facilities related to these phases. Typically this complexity and vast existing infrastructure investment makes it much less feasible for NSOs to consider fully "internalising" SDMX in the short term.
As described earlier, however, there is an increasing trend toward generalising, standardising and re-using process steps within statistical production processes and the infrastructure that supports those steps.
This trend can be seen within individual NSOs, where rather than each statistical production process having a completely customised approach to a particular sub-process (e.g. imputation) there might be a small number of preferred methods identified and one or more systematic pathways provided for applying the relevant method to a particular set of data. (The most appropriate method may vary depending according to the size of the sample, the nature of the data or other factors. Typically more than one "preferred method" and more than one "preferred systematic path" needs to be supported but that is still far preferable to supporting "n" methods and paths for "n" different statistical production processes.)
The trend can also be seen across agencies, with increasing numbers of practical examples of, and plans for, NSOs developing and sharing common software to support common process steps.
Both within and across agencies this trend leads toward common ways of describing the data and metadata to be acted upon by the process step and the metadata that will "drive" the process step (e.g. settings for parameters and options, structure mappings etc). It is considered by many that SDMX offers a very strong value proposition in this regard, being standard, open, well supported and relatively straightforward in its core but also very powerful and flexible. In many cases, if SDMX didn't already exist - or wasn't considered - something similar but inferior would need to be invented for the purpose.
As process steps are chained together into a workflow, SDMX starts to become a means of exchanging data and metadata between process steps as well as between agencies. An exchange between agencies may require that the provider and/or the consumer map from/to a format internal to that agency as part of the exchange using the common format. Similarly, if a relevant existing data or metadata repository, or an upstream or downstream process step, doesn't natively support the common (SDMX based or otherwise) way of describing the content to be acted upon by a shared process step then some mapping may be required.
Such mapping is likely to be required in the short to medium term. Alongside newer, shared, open process steps and supporting infrastructure, there will continue to exist process steps and processing facilities that are highly specific to a particular production process and "proprietary" rather than open in nature. Over time, however, there will be an incentive to simplify flows by being able to re-use the same structures and concepts for other processing steps even if those other steps are based on localised supporting infrastructure rather than common infrastructure.
Especially when consideration is given to sharing processing capabilities and supporting software across NSOs it becomes clear that the SDMX Content-oriented Guidelines are equally important to facilitating and shaping exchange of statistical content between processes as they are to facilitating and shaping exchange of statistical content between agencies. Even considering a processing flow which is only shared across statistical production activities within a single agency, however, if the content involved in those flows is shaped by the Content-oriented Guidelines then disseminating to other agencies via SDMX the statistical outputs produced from those processes should be much simpler and involve less "information loss". This underscores the fact that the emerging focus on SDMX for structuring the data and metadata to be acted upon by process steps, and supporting the flow of content between process steps, is symbiotic to the original, and still primary, purpose of SDMX V2 supporting exchange and sharing among agencies.
MSIS (Management of Statistical Information Systems) is a group led by the UNECE, Eurostat and the OECD, with engagement of NSOs, which is currently working toward a system for sharing open-source technical solutions between statistical organizations.
The MSIS Taskforce on Sharing Statistical Software has recognised SDMX as a key enabler of the required level of interoperability to allow such sharing of infrastructure solutions to be viable in practice. The Taskforce also recommends adoption of the GSBPM as an essential step toward achieving the convergence of enterprise architectures within NSOs that would enable such practical sharing.
A Sharing Advisory Board is in the process of being founded by MSIS to define and progress strategies in this area into the future.
In the meantime, at a more tactical level, a project within the ESS (European Statistical System) has been proposed for 2009 to focus on a common reference architecture. As part of the specification SDMX is already proposed as a standard format for exchanging aggregate data between the IT "building blocks" used to support the common "business process architecture".
One of the outputs from the ESS project would be a recommended "metadata architecture" (together with the business architecture describing the Business Process Metadata). The extent to which SDMX may influence that "metadata architecture” remains to be seen.
Bringing it all Together
The preceding two sections have highlighted the strong and increasing connections between SDMX and statistical business processes, where the GSBPM provides a common high level reference framework for the latter.
The fact these connections exist does not necessarily mean they should be primarily recognised and managed "within SDMX". Alternatives might include:
- Building into the documentation of the GSBPM a detailed description of its "fit" with SDMX;
- Preparing a completely separate set of detailed guidelines on the relationship between SDMX and the GSBPM.
It is suggested, however, that recognising and managing the connections within SDMX is by far the most appropriate option. Reasons include the following:
- SDMX is an "umbrella" set of standards and guidelines, and the connections between data, metadata and processes fit comfortably under that umbrella
- The depth and scope of the GSBPM itself is much more limited
- The SDMX V2 technical standard already has a package related to processes
- The GSBPM as it currently stands doesn't (and, arguably, shouldn't) have anything to say about data, metadata and frameworks and standards for them
- The SDMX Content-oriented Guidelines provide a ready made means of describing the elements of the GSBPM and their relevance.
Annex A provides some thoughts on how this might be achieved initially through some simple and limited updates, primarily relevant minor extensions to the SDMX Content-oriented Guidelines.
A key consideration is that this recognition of the GSBPM would be provided largely "ahead of the game" (although, it would seem, only just ahead of it). This is far preferable to waiting until a number of different implementations have attempted to relate SDMX and statistical processes in completely different and incompatible ways. Under those circumstances it may be much harder to agree, and have adopted in practice, any cross domain concept in this area because several competing implementations may have a strong vested interest in promoting their own approach, in which they have a sunk investment, as the common standard.
In this regard, recognising the GSBPM in the manner proposed in Annex A, or in a similar manner, may be a necessary but not sufficient step.
Given that connections between SDMX and statistical business processes is an area of increasing practical interest and activity, and SDMX is, by nature, a highly collaborative initiative, there should be a high level perspective and opportunity to share information (including linking to it rather than replicating it in full) provided through the SDMX website. (In effect, the website already provides two alternative views on implementation plans, activities and experiences - one "by statistical subject-matter domains" and one "by organisation".)
As mentioned earlier, at this stage there won't be a lot of completed implementations that have a strong link with end to end business processes. There will, however, be early plans, ideas, explorations and experiences to share. Providing such a view may significantly further promote sharing and consistency in regard to high level approaches beyond just including references in the Content-oriented Guidelines.
Providing such a view appears to mesh very well with the intentions of a paper which is due to be presented to the UN Statistical Commission (UNSC) toward the end of February 2009. The main annex of that paper contains many plans and ideas for interacting with "distributed initiatives in a way that strengthens the relevance and usefulness of SDMX to all its constituencies". Part II of the Annex focus on measures to foster participation in SDMX developments, including collaborative approaches to the development of Data Structure Definitions (DSDs) and Metadata Structure Definitions (MSDs).
If the primary thesis of the current paper is correct, a significant proportion of the work on DSDs and MSDs in future will relate to common structures for statistical processes to act upon, not just structures for exchanging data and metadata between agencies. While the UNSC paper focuses on value to be added through more usable and collaborative views of activities grouped according to statistical subject-matter domain and according to organisation, an emerging third perspective based on the relationship between SDMX and statistical business processes may also be a worthwhile addition to the exciting way forward proposed in that paper.
- Recognising the GSBPM would impact the Technical Standards (possibly for Version 2.1 or Version 3) very marginally if at all.
- Section 8 of the Framework Document currently refers to "Dependencies on SDMX Content Standards". Just as 8.3 refers to Statistical Subject-Matter Domains, including possible use as a standard scheme for categorisation and mapping purposes, there could be reference to the GSBPM in future.
- It is proposed that Statistical Business Process be added as a cross domain concept. Similarly to a number of existing cross domain concepts (e.g. Adjustment, Frequency and Unit of Measure) there should be the option to provide coded references (in this case to a list based on the GSBPM codes at the phase and sub-process level) and/or free text based detailed information. A few codes may need to be added to cover, e.g.
- reference specifically to the overarching "Quality Management/Metadata Management" aspect of the GSBPM;
- reference to the statistical process as a whole (rather than a specific phase or sub-process within a phase)
- a code for “not applicable” to indicate that a cross reference to the GSBPM is not possible and/or not appropriate in a particular case
- It is also proposed that a separate annex be added to the Content-oriented Guidelines (similar to the current Statistical Subject-Matter Domains annex) to provide a summary explanation of the GSBPM and the ways it may be relevant to SDMX. The references in the Cross-Domain Concepts and the Cross-Domain Code Lists can be kept very concise by simply providing a link to this new annex for context and additional information.
It is an open question as to whether this new annex should become the authoritative version of the GSBPM or not.
- The current Statistical Subject-Matter Domains annex cross references the UNECE Classification of International Statistical Activities (CISA) but the authoritative version of the latter continues to exist independently. The fact the two entities remain notionally separate makes it theoretically possible for the two to diverge temporarily (e.g. one is updated before the same change is applied to the other) or permanently. In practice it is expected that synchronisation (for Statistical Domains 1-3 in CISA) would be maintained if at all possible.
- Whether an authoritative version of GSBPM would continue to exist separately - and, if so, what synchronisation arrangements might apply - would depend on the preferences of those currently responsible for the GSBPM and for the SDMX Content-oriented Guidelines.
- There should also be a brief specific mention of the GSBPM and the ways it may be relevant to SDMX in the main document related to the SDMX Content-oriented Guidelines.
- This is particularly the case if a new annex related to the GSBPM is added to the Content-oriented Guidelines.
- It is proposed updates to the SDMX Metadata Common Vocabulary (MCV) be minimal.
- Quite appropriately, the individual Statistical Domains from CISA, let alone the Statistical Areas within each Domain, are not included in the MCV.
- Similarly it is proposed that the Phase and Sub-Process terms from the GSBPM are not referenced by separate entries in the MCV.
- The minimalist approach is particularly appropriate because a lot of the terms are "common use" (e.g. “Collect”, “Variables”) - and may be used legitimately in many other contexts in a way that does not align precisely with descriptions given in the GSBPM documentation.
- The GSBPM is acting as a reference model, and providing a meaning for terms in that context, rather than seeking to give a definition to the terms which is authoritative beyond that very specific context.
- The term "Statistical Business Process" (as a cross domain concept) would need to be added, with a hyperlink to the annex related to the GSBPM.
- There might also be cross references added to the existing MCV entries for "Statistical processing" and "Statistical production".
- While details could be worked through at the same time as the MCV is being updated for other reasons, some of the related terms for "Statistical processing" - where they align closely with terms used in the Phase and Sub-Process levels of the GSBPM - might also have cross references added.
The key point is that the MCV would - where relevant - cross reference documentation related to the GSBPM rather than itself becoming the repository for the detailed definition of each element.
Phase 1 - Specify Needs
This phase is triggered when a need for new statistics is identified, or feedback about current statistics initiates a review. It determines whether there is a presently unmet demand, externally and / or internally, for the identified statistics and whether the statistical organisation can produce them.
In this phase the organisation:
- determines the need for the statistics;
- confirms, in more detail, the statistical needs of the stakeholders;
- establishes the high level objectives of the statistical outputs;
- checks if current collections and / or methodologies can meet these needs, and;
- completes the business case to get approval to produce the statistics.
This phase is broken down into five sub-processes. These are generally sequential, but can also occur in parallel, and can be iterative:
1.1. Determine need for information - This sub-process includes the initial investigation and identification of what statistics are needed and what is needed of the statistics. It also includes consideration of practice amongst other (national and international) statistical organisations producing similar data, and in particular the methods used by those organisations.
1.2. Consult and confirm need - This sub-process focuses on consulting with the stakeholders and confirming in detail the need for the statistics. A good understanding of user needs is required so that the statistical organisation knows not only what it is expected to deliver, but also when, how, and, perhaps most importantly, why. For second and subsequent iterations of this phase, the main focus will be on determining whether previously identified needs have changed. This detailed understanding of user needs is the critical part of this sub-process.
1.3. Establish output objectives - This sub-process identifies the statistical outputs that are required to meet the user needs identified in sub-process 1.2 (Consult and confirm need). It includes agreeing the suitability of the proposed outputs and their quality measures with users.
1.4. Check data availability - This sub-process checks whether current data sources could meet user requirements, and the conditions under which they would be available, including any restrictions on their use. An assessment of possible alternatives would normally include research into potential administrative data sources and their methodologies, to determine whether they would be suitable for use for statistical purposes. When existing sources have been assessed, a strategy for filling any remaining gaps in the data requirement is prepared.
1.5. Prepare business case - This sub-process documents the findings of the other sub-processes in this phase in the form a business case to get approval to implement the new or modified statistical business process. Such a business case would typically also include:
- A description of the "As-Is" business process (if it already exists), with information on how the current statistics are produced, highlighting any inefficiencies and issues to be addressed;
- The proposed "To-Be" solution, detailing how the statistical business process will be developed to produce the new or revised statistics;
- An assessment of costs and benefits, as well as any external constraints.
Phase 2 - Design
This phase describes the development and design activities, and any associated practical research work needed to define the statistical outputs, concepts, methodologies, collection instruments and operational processes. For statistical outputs produced on a regular basis, this phase usually occurs for the first iteration, and whenever improvement actions are identified in phase 9 (Evaluate) of a previous iteration.
This phase is broken down into six sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
2.1. Outputs - This sub-process contains the detailed design of the statistical outputs to be produced, including the related development work and preparation of the systems and tools used in phase 7 (Disseminate). Outputs should be designed, wherever possible, to follow existing standards, so inputs to this process may include metadata from similar or previous collections, international standards, and information about practices in other statistical organisations from sub-process 1.1 (Determine need for information).
2.2. Frame and sample methodology - This sub-process identifies and specifies the population of interest, defines a sampling frame (and, where necessary, the register from which it is derived), and determines the most appropriate sampling criteria and methodology (which could include complete enumeration). Common sources are administrative and statistical registers, censuses and sample surveys. This sub-process describes how these sources can be combined if needed. Analysis of whether the frame covers the target population should be performed. A sampling plan should be made: The actual sample is created sub-process 4.1 (Select sample), using the methodology, specified in this sub-process.
2.3 Variables - This sub-process defines the variables to be collected via the data collection instrument, as well as any other variables that will be derived from them in sub-process 5.4 (Derive new variables), and any classifications that will be used. It is expected that existing national and international standards will be followed wherever possible. This sub-process may need to run in parallel with sub-process 2.4 (Data collection), as the definition of the variables to be collected, and the choice of data collection instrument may be inter-dependent to some degree. Preparation of metadata descriptions of collected and derived variables and classifications is a necessary precondition for subsequent phases.
2.4. Data collection - This sub-process determines the most appropriate data collection method(s) and instrument(s). The actual activities in this sub-process vary according to the type of collection instruments required, which can include computer assisted interviewing, paper questionnaires, administrative data interfaces and data integration techniques. This sub-process includes the design of questions and response templates (in conjunction with the variables and classifications designed in sub-process 2.3 (Variables)). It also includes the design of any formal agreements relating to data supply, such as memoranda of understanding, and confirmation of the legal basis for the data collection. This sub-process is enabled by tools such as question libraries (to facilitate the reuse of questions and related attributes), questionnaire tools (to enable the quick and easy compilation of questions into formats suitable for cognitive testing) and agreement templates (to help standardise terms and conditions). This sub-process also includes the design of process-specific provider management systems.
2.5. Statistical processing methodology - This sub-process designs the statistical processing methodology to be applied during phase 5 (Process), and Phase 6 (Analyse). This can include developing and testing routines for coding, editing, imputing, estimating integrating, verifying and finalising data sets.
2.6. Processing systems and workflow - This sub-process determines the workflow from data collection to archiving, taking an overview of all the processes required within the whole statistical production process, and ensuring that they fit together efficiently with no gaps or redundancies. Various systems and databases are needed throughout the process. A general principle is to reuse processes and technology across many statistical business processes, so existing systems and databases should be examined first, to determine whether they are fit for purpose for this specific process, then, if any gaps are identified, new solutions should be designed.
Phase 3 - Build
This phase builds and tests the production systems to the point where they are ready for use in the "live" environment. For statistical outputs produced on a regular basis, this phase usually occurs for the first iteration, and following a review or a change in methodology, rather than for every iteration. It is broken down into five sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
3.1. Data collection instrument - This sub-process describes the activities to build the collection instruments to be used during the phase 4 (Collect). The collection instrument is generated or built based on the design specifications created during phase 2 (Design). A collection may use one or more collection modes to receive the data, e.g. personal or telephone interviews; paper, electronic or web questionnaires; SDMX hubs. Collection instruments may also be data extraction routines used to gather data from existing statistical or administrative data sets. This sub-process also includes preparing and testing the contents and functioning of that instrument (e.g. testing the questions in a questionnaire). It is recommended to consider the direct connection of collection instruments to the statistical metadata system, so that metadata can be more easily captured in the collection phase. Connection of metadata and data at the point of capture can save work in later phases.
3.2. Process components - This sub-process describes the activities to build new and enhance existing software components needed for the business process, as designed in Phase 2 (Design). Components may include dashboard functions and features, data repositories, transformation tools, workflow framework components, provider and metadata management tools.
3.3. Configure workflows - This sub-process configures the workflow, systems and transformations used within the statistical business processes, from data collection, right through to archiving the final statistical outputs. It ensures that the workflow specified in sub-process 2.6 (Processing system and workflow) works in practice.
3.4. Test - This sub-process describes the activities to manage a field test or pilot of the statistical business process. Typically it includes a small-scale data collection, to test collection instruments, followed by processing and analysis of the collected data, to ensure the statistical business process performs as expected. Following the pilot, it may be necessary to go back to a previous step and make adjustments to instruments, systems or components. For a major statistical business process, e.g. a population census, there may be several iterations until the process is working satisfactorily.
3.5. Finalise production systems - This sub-process includes the activities to put the process, including workflow systems, modified and newly-built components into production ready for use by business areas. The activities include:
- producing documentation about the process components, including technical documentation and user manuals;
- training the business users on how to operate the process;
moving the process components into the production environment, and ensuring they work as expected in that environment (this activity may also be part of sub-process 3.4 (Test)).
Phase 4 - Collect
This phase collects all necessary data, using different collection modes (including extractions from administrative and statistical registers and databases), and loads them into the appropriate data environment. For statistical outputs produced regularly, this phase occurs in each iteration. It is broken down into four sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
4.1. Select sample - This sub-process establishes the frame and selects the sample for this iteration of the collection, as specified in sub-process 2.2 (Frame and sample methodology). It also includes the coordination of samples between instances of the same statistical business process (for example to manage overlap or rotation), and between different processes using a common frame or register (for example to manage overlap or to spread response burden). Quality assurance, approval and maintenance of the frame and the selected sample are also undertaken in this sub-process, though maintenance of underlying registers, from which frames for several statistical business processes are drawn, is treated as a separate business process. The sampling aspect of this sub-process is not usually relevant for processes based entirely on the use of pre-existing data sources (e.g. administrative data) as such processes generally create frames from the available data and then follow a census approach.
4.2. Set up collection - This sub-process ensures that the people, processes and technology are ready to collect data, in all modes as designed. It takes place over a period of time, as it includes the strategy, planning and training activities in preparation for the specific instance of the statistical business process. Where the process is repeated regularly, some (or all) of these activities may not be explicitly required for each iteration. For one-off and new processes, these activities can be lengthy. This sub-process includes:
- preparing a collection strategy;
- training collection staff;
- ensuring collection resources are available e.g. laptops;
- configuring collection systems to request and receive the data;
- ensuring the security of data to be collected;
- preparing collection instruments (e.g. printing questionnaires, pre-filling them with existing data, loading questionnaires and data onto interviewers' computers etc.).
4.3. Run collection - This sub-process is where the collection is implemented, with the different collection instruments being used to collect the data. It includes the initial contact with providers and any subsequent follow-up or reminder actions. It records when and how providers were contacted, and whether they have responded. This sub-process also includes the management of the providers involved in the current collection, ensuring that the relationship between the statistical organisation and data providers remains positive, and recording and responding to comments, queries and complaints. For administrative data, this process is brief: the provider is either contacted to send the data, or sends it as scheduled. When the collection meets its targets (usually based on response rates) the collection is closed and a report on the collection is produced.
4.4. Load data into processing environment - This sub-process includes initial data validation, as well as loading the collected data and metadata into a suitable electronic environment for further processing in phase 5 (Process). It may include automatic data take-on, for example using optical character recognition tools to extract data from paper questionnaires, or converting the formats of data files received from other organisations. In cases where there is a physical data collection instrument, such as a paper questionnaire, which is not needed for further processing, this sub-process manages the archiving of that material in conformance with the principles established in phase 8 (Archive).
Phase 5 - Process
This phase describes the cleaning of data records and their preparation for analysis. It is made up of sub-processes that check, clean, and transform the collected data, and may be repeated several times. For statistical outputs produced regularly, this phase occurs in each iteration. The sub-processes in this phase can apply to data from both statistical and non-statistical sources (with the possible exception of sub-process 5.6 (Calculate weights), which is usually specific to survey data).
The "Process" and "Analyse" phases can be iterative and parallel. Analysis can reveal a broader understanding of the data, which might make it apparent that additional processing is needed. Activities within the "Process" and "Analyse" phases may commence before the "Collect" phase is completed. This enables the compilation of provisional results where timeliness is an important concern for users, and increases the time available for analysis. The key difference between these phases is that "Process" concerns transformations of microdata, whereas "Analyse" concerns the further treatment of statistical aggregates.
This phase is broken down into seven sub-processes, which may be sequential or parallel, and can be iterative:
5.1. Standardize and anonymize - This sub-process is where statistical units are derived or standardized, and where data are anonymized. Depending on the type of source data, this sub-process may not always be needed. Standardization includes converting administrative or collection units into the statistical units required for further processing. Anonymization strips data of identifiers such as name and address, to help to protect confidentiality. Standardization and anonymization may take place before or after sub-process 5.2 (Integrate data), depending on the requirements for units and identifiers in that sub-process.
5.2. Integrate data - This sub-process integrates one or more data sources. The input data can be from a mixture of external or internal data sources, and a variety of collection modes. The result is a harmonised data set. Data integration typically includes:
- matching / record linkage routines, with the aim of linking data from different sources referring to the same unit;
- prioritising when two or more sources contain data for the same variable (with potentially different values).
Data integration may take place at any point in this phase, before or after any of the other sub-processes. There may also be several instances of data integration in any statistical business process.
5.3. Classify and code - This sub-process classifies and codes the input data. For example automatic (or clerical) coding routines may assign numeric codes to text responses according to a pre-determined classification scheme.
5.4. Edit and impute - This sub-process applies to collected micro-data, and looks at each record to try to identify (and where necessary correct) missing data, errors and discrepancies. It can also be referred to as input data validation. It may be run iteratively, validating data against predefined edit rules, usually in a set order. It may apply automatic edits, or raise alerts for manual inspection and correction of the data. Where data are missing or unreliable, estimates are imputed, often using a rule-based approach. Specific steps include:
- the identification of potential errors and gaps;
- the selection of data to include or exclude from editing and imputation routines;
- editing and imputation using one or more pre-defined methods e.g. "hot-deck" or "cold-deck" imputation;
- writing the edited / imputed data back to the data set, and flagging them as edited or imputed;
- the production of metadata on the editing and imputation process.
Editing and imputation can apply to unit records both from surveys and administrative sources, before and after integration.
5.5. Derive new variables - This sub-process creates variables that are not explicitly provided in the collection and are needed to deliver the required outputs. It derives these new variables by applying arithmetic formulae to one or more of the variables that are already present in the dataset. It may need to be iterative, as some derived variables may themselves be based on other derived variables. It is therefore important to ensure that variables are derived in the correct order.
5.6. Calculate weights - This sub process creates weights for unit data records according to the methodology created in sub-process 2.5: Statistical processing methodology. These weights can be used to "gross-up" sample survey results to make them representative of the target population, or to adjust for non-response in total enumerations.
5.7. Calculate aggregates - This sub process creates aggregate data and population totals from micro-data. It includes summing data for records sharing certain characteristics, determining measures of average and dispersion, and applying weights from sub-process 5.6 to sample survey data to derive population totals.
Phase 6 - Analyse
In this phase, statistics are produced, examined in detail, interpreted, and made ready for dissemination. This phase includes the sub-processes and activities that enable statistical analysts to understand the statistics produced. For statistical outputs produced regularly, this phase occurs in every iteration. The Analyse phase and sub-processes are generic for all statistical outputs, regardless of how the data were sourced.
The Analyse phase is broken down into six sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
6.1. Acquire domain intelligence - This sub-process includes many ongoing activities involved with the gathering of intelligence, with the cumulative effect of building up a body of knowledge about a specific statistical domain. This knowledge is then applied to the current collection, in the current environment, to allow informed analyses. Acquiring a high level of domain intelligence will allow a statistical analyst to understand the data better, and to identify where results might differ from expected values. This allows better explanations of these results in sub-process 6.4 (Interpret and explain).
6.2. Prepare draft outputs - This sub-process is where domain intelligence is applied to the data collected to produce statistical outputs. It includes the production of additional measurements such as indices or seasonally adjusted series, as well as the recording of quality characteristics.
6.3. Verify outputs - This sub-process is where statisticians verify the quality of the outputs produced, in accordance with a general quality framework. Verification activities can include:
- checking that the population coverage and response rates are as required;
- comparing the statistics with previous cycles (if applicable);
- confronting the statistics against other relevant data (both internal and external);
- investigating inconsistencies in the statistics;
- performing macro editing;
- verifying the statistics against expectations and domain intelligence.
6.4. Interpret and explain - This sub-process is where the in-depth understanding of the outputs is gained by statisticians. They use that understanding to interpret and explain the statistics produced for this cycle by assessing how well the statistics reflect their initial expectations, viewing the statistics from all perspectives using different tools and media, and carrying out in-depth statistical analyses.
6.5 Disclosure control - This sub-process ensures that the data (and metadata) to be disseminated do not breach the appropriate rules on confidentiality. This may include checks for primary and secondary disclosure, as well as the application of data suppression or perturbation techniques.
6.6. Finalize outputs for dissemination - This sub-process ensures the statistics and associated information are fit for purpose and reach the required quality level, and are thus ready for dissemination. It includes:
- completing consistency checks;
- determining the level of release, and applying caveats;
- collating supporting information, including interpretation, briefings, measures of uncertainty and any other necessary metadata;
- producing the supporting internal documents;
- pre-release discussion with appropriate internal subject matter experts;
approving the statistical content for release.
Phase 7 - Disseminate
This phase manages the release of the statistical products to customers. For statistical outputs produced regularly, this phase occurs in each iteration. It is made up of five sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
7.1. Update output systems - This sub-process manages the update of systems where data and metadata are stored for dissemination purposes, including:
- formatting data and metadata ready to be put into output databases;
- loading data and metadata into output databases;
- ensuring data are linked to the relevant metadata.
Note: formatting, loading and linking of metadata should preferably mostly take place in earlier phases, but this sub-process includes a check that all of the necessary metadata are in place ready for dissemination.
7.2. Produce products - This sub-process produces the products, as previously designed, to meet user needs. The products can take many forms including printed publications, press releases and web sites. Typical steps include:
- preparing the product components (text, tables, charts etc.);
- assembling the components into products;
- editing the products and checking that they meet publication standards.
7.3. Manage release of products - This sub-process ensures that all elements for the release are in place including managing the timing of the release. It includes briefings for specific groups such as the press or ministers, as well as the arrangements for any pre-release embargoes. It also includes the provision of products to subscribers.
7.4. Market and promote products - Whilst marketing in general can be considered to be an over-arching process, this sub-process concerns the active promotion and marketing of the statistical products produced in a specific statistical business process, to help them reach the widest possible audience. It includes the use of customer relationship management tools, to better target potential users of the products, as well as the use of tools including web sites, wikis and blogs to facilitate the process of communicating statistical information to users.
7.5. Manage customer queries - This sub-process ensures that customer queries are recorded, and that responses are provided within agreed deadlines. These queries should be regularly reviewed to provide an input to the over-arching quality management process, as they can indicate new or changing user needs.
Phase 8 - Archive
This phase manages the archiving and disposal of statistical data and metadata. Given the reduced costs of data storage, it is possible that the archiving strategy adopted by a statistical organisation does not include provision for disposal, so the final sub-process may not be relevant for all statistical business processes. In other cases, disposal may be limited to intermediate files from previous iterations, rather than disseminated data.
For statistical outputs produced regularly, archiving occurs in each iteration, however defining the archiving rules is likely to occur less regularly. This phase is made up of four sub-processes, which are generally sequential, but can also occur in parallel, and can be iterative:
8.1. Define archive rules - This sub-process is where the archiving rules for the statistical data and metadata resulting from a statistical business process are determined. The requirement to archive intermediate outputs such as the sample file, the raw data from the collect phase, and the results of the various stages of the process and analyse phases should also be considered. The archive rules for a specific statistical business process may be fully or partly dependent on the more general archiving policy of the statistical organisation, or, for national organisations, on standards applied across the government sector. The rules should include consideration of the medium and location of the archive, as well as the requirement for keeping duplicate copies. They should also consider the conditions (if any) under which data and metadata should be disposed of. (Note - this sub-process is logically strongly linked to Phase 2 - Design, at least for the first iteration of a statistical business process).
8.2. Manage archive repository - This sub-process concerns the management of one or more archive repositories. These may be databases, or may be physical locations where copies of data or metadata are stored. It includes:
- maintaining catalogues of data and metadata archives, with sufficient information to ensure that individual data or metadata sets can be easily retrieved;
- testing retrieval processes;
- periodic checking of the integrity of archived data and metadata;
- upgrading software-specific archive formats when software changes.
This sub-process may cover a specific statistical business process or a group of processes, depending on the degree of standardisation within the organisation. Ultimately it may even be considered to be an over-arching process if organisation-wide standards are put in place.
8.3. Preserve data and associated metadata - This sub-process is where the data and metadata from a specific statistical business process are archived. It includes:
- identifying data and metadata for archiving in line with the rules defined in 8.1;
- formatting those data and metadata for the repository;
- loading or transferring data and metadata to the repository;
- cataloguing the archived data and metadata;
- verifying that the data and metadata have been successfully archived.
8.4. Dispose of data and associated metadata - This sub-process is where the data and metadata from a specific statistical business process are disposed of. It includes;
- identifying data and metadata for disposal, in line with the rules defined in 8.1;
- disposal of those data and metadata;
- recording that those data and metadata have been disposed of.
Phase 9 - Evaluate
This phase manages the evaluation of a specific instance of a statistical business process. It logically takes place at the end of the instance of the process, but relies on inputs gathered throughout the different phases. For statistical outputs produced regularly, evaluation should, at least in theory occur for each iteration, determining whether future iterations should take place, and if so, whether any improvements should be implemented. However, in some cases, particularly for regular and well established statistical business processes, evaluation may not be formally carried out for each iteration. In such cases, this phase can be seen as providing the decision as to whether the next iteration should start from phase 1 (Specify needs) or from some later phase (often phase 4 (Collect)).
This phase is made up of three sub-processes, which are generally sequential, but which can overlap to some extent in practice:
9.1. Gather evaluation inputs - Evaluation material can be produced in any other phase or sub-process. It may take many forms, including feedback from users, process metadata, system metrics and staff suggestions. Reports of progress against an action plan agreed during a previous iteration may also form an input to evaluations of subsequent iterations. This sub-process gathers all of these inputs, and makes them available for the person or team producing the evaluation.
9.2. Prepare evaluation - This sub-process analyzes the evaluation inputs and synthesizes them into an evaluation report. The resulting report should note any quality issues specific to this iteration of the statistical business process, and should make recommendations for changes if appropriate. These recommendations can cover changes to any phase or sub-process for future iterations of the process, or can suggest that the process is not repeated.
9.3. Agree an action plan - This sub-process brings together the necessary decision-making power to form and agree an action plan based on the evaluation report. It should also include consideration of a mechanism for monitoring the impact of those actions, which may, in turn, provide an input to evaluations of future iterations of the process.
 The diagram is also appropriate for international statistical organizations that collect data covering a number of different domains.
 An example of reference metadata structured in accordance with ESMS is http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/EN/hlth_hlye_esms.htm