Closed: Population censuses; are statistical dinosaurs able to adapt?

Comments on “Comments on the scientific basis of the register-based census by Anders Wallgren and Britt Wallgren”.

I received a copy of the Comments of Wallgren & Wallgren on my article in the March 2020 issue of SJ IAOS. I appreciate their efforts to comment on my article and below is my reaction on their contribution to the discussion. I will also upload a pdf version of my comments.

Introduction
The objective of my paper was to review the development of the census as a general scientific method and its underlying scientific principles over time, including the recent emergence of alternative procedures to the census, and to make a call for a science-wide review of the scientific principles underlying the different “types of censuses”. It is generally accepted that the census in population statistics is based on well-established scientific principles, and that it has an array of proven methods to deal with issues of quality, including completeness and errors. In the second half of the 20th century a number of European countries offered an alternative to the census, using data derived from existing administrative records instead of direct field enumeration, the so-called register-based census. Although Statistics Denmark was the first country to produce a complete set of population and housing” census-like” tables using registers of data derived from administrative records in 1981 [1, p. 44] , there is at present no standard methodology for their production. Moreover, some of the main proponents of the use of registers to replace direct data collection methods have indicated that there are no theoretical bases for these procedures. In my paper I only presented, briefly, some of the ideas of B. F. M. Bakker (Statistics Netherlands) [2, pp. 3 -5: 3, p. 17] and Anders Wallgren and Britt Wallgren (Statistics Sweden) [4, p. 3]. I will not repeat what I have written on Bart Bakker’s ideas, but for clarity’s sake I repeat my original Wallgren and Wallgren quote: “Although register-based statistics are a common form of statistics used for official statistics and business reports, no well-established theory in the field exists. There are no recognised terms or principles, which makes the development of register-based statistics and register-statistical methodology all the more difficult. As a consequence, ad hoc methods are used instead of methods based on a generally accepted theory.” [4, p. 3, emphasis in the original]. Wallgren and Wallgren have reacted to my article, and it seems that they wish to distance themselves from the statement that “no well-established theory in the field exists”. [5]. In my comments on their comments I will remain within the universe of discourse established by their comments, with one exception. In the final assessment of the theoretical basis of the so-called register-based census I will refer to a 2011 publication of Bart Bakker which was not referred to in my article nor in the comments of the commentators.

Comments
The commentators’ contribution consists of four parts: an “Introduction”; a section called “Is there a theory?” a section called “The traditional census and the register-based census - main differences”; and finally the “Conclusions – what about the scientific basis?”
In the introduction the commentators partly reproduce my quote without comment.
In the section “Is there a theory?” the commentators refer to a 1987 article by Robert M. Groves [6] and indicate that they “interpret the concept “survey theory” in the same way” as he does, without quoting his definition . They state that “Perhaps the present state of register survey theory is similar to the situation [of survey research, added by author] in 1987 when Groves wrote his paper. The first section in his paper has the title “Survey research as a Methodology Without a Unifying Theory”. Our conclusion is that this present state results in a diversity of terms and principles instead of a unified theory.”
In less than 20 years Robert M. Groves’ 1987 article was superseded by the 2004 (text) book on survey methodology of which he is the main author. [7] In the Preface Groves and his co- authors state “…Over the past two decades a set of theories and principles has evolved that offer a unified perspective on the design, conduct, and evaluation of surveys. This perspective is most commonly labeled the "total survey error" paradigm. [7, p. xv] The 1987 concept of “survey theory”, to which the commentators subscribe, is now replaced by the concept of “survey methodology”. “Survey methodology seeks to identify principles about the design, collection, processing, and analysis of surveys that are linked to the cost and quality of survey estimates. This means that the field focuses on improving quality within cost constraints, or, alternatively, reducing costs for some fixed level of quality. "Quality" is defined within a framework labeled the total survey error paradigm (… ). Survey methodology is both a scientific field and a profession.” [7, p.32] According to Groves et al. to achieve high-quality survey results requires the application of the principles from several traditional academic disciplines, among which mathematics (statistics), social sciences, cognitive psychology and computer science. Except for the principles of computer science, the principles of the other sciences were established before 1960.

According to the commentators “The foundations of a theory regarding register surveys and register systems were developed as early as during the 1960s. “ They refer to the seminal 1967 paper of Svein Nordbotten [8] in which he developed the basis of electronic statistical data file systems. He defined the data file, its components, structure and hierarchy, specifying the conditions for the acquisition, distinguishing between continuous and non-continues collection, storage, maintenance and processing of the data, including the combining of data files to create a statistical data system. He also described the relation between continuous data collection (registers) and the census. “The continuous collection scheme contributes to a more balanced distribution of collection activities over time and to the reduction of the number and extent of large censuses. There are, however, characteristics, particularly those for which there are no contemporary administrative needs, which may be most efficiently collected in a census. The censuses may also provide convenient status sets for checking the current registration.” [8, p.744]
The commentators state” With the first register-based census, Statistics Denmark developed new survey methods so that administrative registers could be used to estimate the same parameters as with the traditional census. Our conclusion is that there is a theory, but the theory is not yet well-established.”
Section 3 “The traditional census and the register-based census - main differences” has no direct bearing on the theoretical principles of the census, and will not be discussed in detail. Just one observation the commentators miss the most salient difference between the classical or traditional census and the so-called register-based census. The first covers the total population of a specific territory at a specific moment, while the latter only deals with the registered population.
The commentators in the conclusions state:” The scientific basis of the classical or traditional census is well-established, and advanced methods have been developed to handle issues regarding coverage and other error sources. The register-based census also has a scientific basis. The new methods are developing and will gradually become established. The choice between the two methodologies should be based on a simultaneous judgment of costs and quality.” So what is the scientific basis of the register-based census? The commentators don’t say, but want to compare the two methodologies on costs and quality. However one has a well-established and known theoretical basis, the other has either no scientific basis, as other proponents claim [e.g. 2, pp. 3 -5: 3, p. 17], or, as the commentators maintain, it has an unknown theoretical basis which is under development. This is a methodological impossibility. Comparing methodologies or procedures just on cost and quality is a very reductive view. This issue is what methodology is at the basis of the so-called register-based census! In my article I have referred to Bakker’s description of different methodological options. I would like to refer here to one of the options, namely micro integration, which is fundamental to register-based statistics, According to Bakker micro integration “involves matching data from statistical units at an individual level, with the goal of compiling better information than is possible when using the separate sources.” [9, p. 4]. But what is the status of micro integration? According to Bakker “There has been a long debate over the question of whether micro integration is actually a method or not. To a significant extent, micro integration cannot be generalised, in the sense that it requires knowledge of the sources and the actual situation in society. As a result, specific micro integration rules are continually drawn up and applied.” [9, p.5]
When comparing procedures it is fundamental to incorporate the pre-conditions and operational and technical requirements for each procedure in the assessment. So we have to compare different procedures not only methodologies on costs and quality. According to Wallgren and Wallgren “There is no commonly used way of describing the quality of register-based statistics.” [4, p. 4]. This means that only the cost aspect remains as an element of comparison and assessment of the procedures. One procedure, the classical or traditional census, has an established scientific basis and advanced methods to deal with coverage and other error sources, whereas the other, the so-called register-based census, either is without an established scientific basis, or has an unknown scientific basis that is under development and has no procedures to establish the coverage and for which there is no standard way of describing quality.
Although it is alleged that the so-called register-based census is cheaper than a classical or traditional census, as far as the author knows, a proper cost analysis for the society, not just the statistical office or system, has never been carried out.
Hence, my appeal remains, as science is self-regulating, where are the guardians of the science of statistics to bring order in chaos?

Note: A word on the terminology, in English language literature the generic common language term “survey” describing any type of inquiry or inventory is used, simultaneous with the more technical term as in “sample survey”. A census becomes a “100 % survey”. The methodology of the census was internationally standardised in the second half of the 19th century. Sample survey methodology was formalised in the second half of the 20th century, when exactly is a matter of opinion.

References:
1. Lange A. The population and housing census in a register based statistical system. Statistical Journal of the IAOS. 2014; 30: 41-45. Doi 10.3233/SJI-140798.
2. United Nations Economic Commission for Europe. Conference of European Statisticians. Statistical Office. Joint UNECE/Eurostat Expert Group Meeting on Register-Based Censuses (The Hague, The Netherlands, 10–11 May 2010. Statistics Netherlands. Micro-integration: State of the Art. Working paper 10. 12 May 2010.
3. Bakker, BFM. Estimating the validity of administrative variables. Statistica Neerlandica. 2012; 66(1): 8-17.
4. Wallgren A, Wallgren B. Register-based statistics. Statistical methods for administrative data. Second edition, Chichester, UK: John Wiley & Sons Ltd: 2014
5. Wallgren A, Wallgren B. Comments on the scientific basis of the register-based census. Comments uploaded on Population censuses; are statistical dinosaurs able to adapt? https://officialstatistics.com/news-blog/population-censuses-are-statis…
6. Groves, R. M. Research on Survey Data Quality. Public Opinion Quarterly. 1987: 51: 4-Part 2: S156-S172.
7. Groves, R.M. et al. Survey Methodology. Hoboken, New Jersey. John Wiley & Sons, Inc. 2004.
8. Nordbotten, S. Purposes, Problems and Ideas Related to Statistical File Systems. Bulletin of the International Statistical Institute. 42. 1967: 733 – 750.
9. Bakker, B. Micro integration. Statistical Methods (20118). The Hague. Statistics Netherlands. 2011.

Alphonse L. MacDonald
Honorary Senior Adviser Methodology to the Director, Algemeen Bureau voor de Statistiek, Paramaribo, Suriname

Commentary by Anders and Britt Wallgren

My position on the issues raised for discussion are clear. It is necessary that as statisticians we stick to the tenets and instruments of science, and do so consistently!

1.Should the census be defined by its unique methodology or are the outputs the main element of its definition?
An activity or instrument cannot be defined by its results. To obtain data, or statistical tables, one has to carry out some sort of operations are activity. So, unless there is something like a "census table" which can be uniquely and independently define. the census will remain an instrument (a set of activities) to obtain data with some specific predetremined characteristics.

2. Population censuses; are statistical dinosaurs able to adapt?
Based on the contribution is appears that colleagues are aware that dinosaurs have beautifully adapted! Modern birds such as the colourful hummingbirds, the powerful Andean condor etc all are descendants of dinosaur's. There are of course also the chicken!

2. Some authors question the theoretical soundness of the register based census. Readers are invited to contribute to the discussion to give their opinion if such a theoretical base is really needed for a census, and why so?
My position is very clear, and I have not heard anything against the validity of my reasoning.

Colleagues have been avoiding a direct answer, but their descriptions of what they consider the census seems to indicate that they do not consider the register-based census as a census. We will need a more direct discussion on the underlying principles of the so-called register-based census.

Alphonse

On the relevance of census taking and census results

We have experienced a growing demand for regional and small area statistics. Urbanisation is one of the trends behind the growing demand. Globalisation and urbanisation goes hand in hand. Globalisation puts an emphasis on availability of comparative statistics on nations, regions and cities. International organisations (UN, OECD, EU and many more), goverments, business organisations, research and so forth are important users, heavy users of census data. Not to forget that many of these users analyse and develop new value added products and services out of the census data. Census data is specially valuable because of the time series and offerings for longitudinal studies informing us about changes and trends.
Statistics of high granularity in terms of content and geography are an important tool in running a city, advancing economy, social well-being and sustainability in a region, empowering citizens and local communities. Everyday life is local and thus small area statistics come close to citizens lives and their living and working areas. It seems that still today the censuses are almost the only prime source offering full data for comparative fine-graned statistics worldwide.
It would be desirable to get census data more frequently than every ten years. Fortunately, there are good prospects for getting census data more frequently and in addition to a more affordable price. I´m referring to the modernasation of official statistics going on in almost every country. National Statistical Offices are utilising administrative sources to a larger extent. They are also advanced in intergrating statistical and geographical data. They have a lot of research going on utilising totally new sources for official statistics and accordingly developing new forms of cooperation and new methodologies. Much of this development work is based on international cooperation from the very first beginning paving the way for comparative statistics also internationally.
The Statistical Journal of the IAOS, Volume 36, issue 1 (2020) comprehensively deals with the theme Population and Housing Census. The Review Article presenting Population and Housing Census Articles published in earlier volumes of the Journal is most valuable. It reminds us of the determined and rewarding work focussed on Population and Housing Cenus through years and decades.

On the census technology:

The main technological innovations in the 2020 Census in Mexico include the following:

The use of multimodal methods for data collection: (i) CAPI (Computer-Assisted Personal Interviewing) was the main enumeration method; (ii) Paper questionnaires were also employed, but expected to have a smaller presence; (iii) CATI (Computer-Assisted Telephone Interviewing) and (iv) CAWI (Computer-Assisted Web Interviewing) were both self-enumeration methods, designed to ease participation. Numerals (iii) and (iv) were COVID-19 pandemic-ready but their potential was reduced since it was difficult to redirect public attention; the organic communication strategy was designed to focus people’s attention on the CAPI enumeration method mentioned before.

The use of individual and unique QR codes to carry out a correct identification of each visited dwelling along with its precise georeferencing information. This was key for the self-enumeration methods.

Carrying out data collection operations with mobile computing devices is estimated to generate savings of up to US$40 million, compared to the previous paper-based census.

A multi-tiered system integrating layers of mobile devices, up to 10 applications, and middle database systems to manage participating staff and to process the stream of collected data until it reaches the main database.

Those were key technological innovations employed in the 2020 Census. Supportive technologies were used as well in different stages: conceptual, methodology, management, field staff training, real-time monitoring of operations, data dissemination, and last, but not least, technological architecture and information security.

The use of technology in the 2020 Census in Mexico is a strong proof of automation, not only of the census itself but of all the processes of daily life, from production to smart homes and cities. Automation makes all processes more efficient and reduces costs. Strategic government programs in education, health, or security, among others, would welcome the savings.

The use of GPS to locate dwellings and their inhabitants was indeed a great idea. It would be interesting to compare the geographic information of previous censuses with the current one (once completed) with the help of geospatial information. This would enable us to see how population areas have grown on a national scale, both in size and population. The objective would be to obtain models with the information from previous censuses up to the current one. It could be possible to predict and even simulate future censuses with the help of Artificial Intelligence as increasing computational power is available.

Perhaps human capital is still regarded as very valuable for the 2020 Census. The transition, however, shows that very soon it will not be as such. For instance, it can be expected that individuals will be able to provide census data by themselves given the downward trend in the prices of mobile IT technologies and the wide acceptance of them.

As well, expected reforms in law and regulations will someday enable linking administrative data with the National Inventory of Dwellings utilizing data marts, new mathematical programming languages, and machine learning.

We can see that the 2020 Census is equivalent to 10.9% of the ten-year cost of the Metropolitan Fund, 5.4% of the cost of the INE in the same period, and only 0.1% of federal contributions over ten years for the states and municipalities of Mexico.

Thus the cost of the 2020 Census in Mexico may seem expensive in the first place, however, with the above comparison, we can see that for the ten-year period in which the censuses information is used, the cost no longer seems to be so high.

As a final point, censuses are the main tool for government and other agencies to design, implement, and follow up public policies. The numbers presented above are only a sample of how much some institutions that use census information spend, but they are not the only ones. Census information is addressed in many other tasks, an example is its use in emergency situations such as that of Covid-19, where census data are useful in determining the location of the most vulnerable population by age, for example. The monitoring of some goals set by the SDG is also done with census data. To cite just an example, its data serves to follow up the objective 11 on "Sustainable Cities and Communities", particularly to assess the growth of the urban slick and of the low-stratum population with access to public transport.

The previous makes clear the need for censuses and their relative high cost, but just if this is not compared with elements like those addressed above.

Below is a comparative table of the cost of the 2020 Census against the budget for metropolitan areas (Metropolitan Fund), the National Electoral Institute, and the Federal Budget for States and municipalities.

On the relevance of census taking and census results:

In Mexico the cost associated with the 2020 Census was around USD 331 755 468 ($7 992 818 613 MXN), numbers include 2018, 2019, and 2020. Such a cost would seem high, however, when compared to the budget allocated to other institutions or programs that use censuses information for their tasks, we see that the cost in proportion is no longer perceived as high.

Census information is mainly used for:

 Distribution of resources (Fiscal Coordination Law). Censuses information is used for the distribution of funds for the states of Mexican federation.
 National Electoral Institute (INE, by its acronym in Spanish) uses it for the definition and integration of electoral districts and therefore it also has to do with reapportioning seats in Congress.
 Delimitation of Metropolitan Zones. Census information on the number of inhabitants is used to delimit the metropolitan areas. The resources that are given to these areas are made through the so-called Metropolitan Fund (Fondo Metropolitano).

Since the 2020 Census in Mexico is conducted every ten years, the following is a comparison with the budget of some institutions that use censuses information, as well as the budget of the federation to the states.

Below is a comparative table of the cost of the 2020 Census against the budget for metropolitan areas (Metropolitan Fund), the National Electoral Institute, and the Federal Budget for States and municipalities.

On the necessity of a census methodology:

So far there is no methodology about the register-based census, what is available, are the different ways to obtain them in each country. Do we need a methodology? Countries that use this method will answer no, as it saves time, money, and effort; purists, surely yes, because it does not comply with the scientific method.

Censuses have long been used in different cultures and eras as a way of accounting for people or things. Since the emergence of the scientific method, different individuals or organizations have established a whole theoretical methodology, reaching what is now known as traditional or classic censuses.

After the Second World War, some European countries began to produce register-based censuses at a much lower cost than traditional ones. Many specialists question whether this type of census has a theoretical methodology like the traditional censuses.

From the view of the scientific method, there is no theoretical methodology for register-based censuses, since it is not reproducible everywhere. One of the main reasons is the disparity in the development of the different countries, thus, while countries like Germany or Denmark have very good administrative records, there are countries in the third world that do not even have them, for example, African countries that do not keep track of actual birth dates in rural areas. In general, there is not a standardization national or international of administrative records, plus a dubious quality of them. Many countries have established a unique identification number for each person and a set of codes for other indicators, which are the same in all administrative offices, while other still have different identifiers for people and indicators. In addition to going against the statistical paradigm: it measures what it has, instead of defining a concept or method to measure it.

Last but not least, in most countries there is a legal laxity for the transfer of administrative records to the national statistical offices, which does not help to establish a register-based census. In fact, much of the transfer of administrative records by government agencies in European countries is due to the goodwill of their current authorities, without having a legal obligation to do so, which makes this practice uncertain in the long term. However, there is one case worth mentioning in the use of administrative data: the Central Statistics Office (CSO) of Ireland. The Statistics Act of 1993 —which governs its mandate— states that the CSO has:

 The obligation to co-ordinate official statistics compiled by public authorities.
 The right to assess the statistical potential of the records maintained by public authorities, as well as to ensure that this potential is realized in so far as resources permit.
 The right of access, subject to some limitations and conditions, to administrative records held by public authorities for statistical purposes.
[Italics added to cite fragments]

Thus, the CSO case is an example of legal faculties that must be assigned to national statistics offices over other public bodies that can generate administrative data for statistical purposes.

On the definition of a census

What defines a census? Is the unique methodology or the outputs? It should be noted that both elements are. On the one hand, methodology, with its five essential features, facilitates over time and spatial comparability and is adaptable to modern needs and technology. On the other hand, there are the outputs, while censuses are universal and make individual enumeration, register-based census runs the risk of not considering the population that -for some reason- is not part of these records, for example, populations in illegal migration. In fact, for vulnerable populations, censuses are the only mechanism that governments have to know their conditions, geographical position, and number; therefore, censuses continue to be essential tools in the development of public policies and social welfare. Let’s see why:

According to the UN definition, five essential features characterize a census:

 Individual enumeration. It is still one of the key features of a census. No other statistical generation mechanism has been able to replace it.

 Universality. Population Censuses have a universal scope and cover an entire population in a given period, including population groups that are not subject to administrative registration such as street population. Some information would be impossible to grasp through a register-based census so, universality is a key feature when measuring the evolution and compliance of Sustainable Development Goals (SDGs).

 Simultaneity. During the 2020 Census, INEGI officials collected information around the country from March 2 to 27, allowing for a photographic description of the population at a specific point in time. Fulfilling this condition in register-based censuses is not possible. Information available from different historical moments makes information unable to compare.

 Periodicity. The 10-year periodicity is one of the features that might be insufficient in the face of rapid social changes. This is a critical issue in the design and establishment of medium- and long-term policies. However, the INEGI updates basic socio-demographic information with census-like exercises every five years since 1995.

 Produce small area statistics. The 2020 Census in Mexico integrates a useful disaggregation level thanks to Basic Geostatistical Areas (AGEB) that are nothing more than a group of blocks. Also, the 2020 Census includes urban and rural environment questionnaires that allow the INEGI to know infrastructure in localities and its state. This information is not possible to obtain through a register-based census.

These elements signify a wide range of theoretical, methodological, and procedural advantages for censuses, as well as advantages in terms of quality of results over a register-based census. Censuses ease international comparability. In the 2010 World Population and Housing Census Program, 214 countries carried out a census and only 21 did not; and nine countries conducted register-based censuses.

Censuses have allowed the observation of a non-heterogeneous reality that has only been possible through clear definitions, standardized procedures, and a specific methodology for place and time. Register-based censuses lack these features. Another advantage of censuses is that their methodology includes mechanisms for evaluating and verifying data quality. Until now, quality evaluation mechanisms for register-based censuses are in its first development stages.

So far, there is no other statistical project that allows such a level of disaggregation, quality, and quantity of information such as the one obtained by population censuses. That information is fundamental for public and social policy development that is suitable to specific needs of population groups. The information quality obtained from it and, by its kind, is invaluable and necessary for the design, implementation, and evaluation of public policies, which is why both, methodology and outputs, define a census.

Covid-19 will probably affect the timeliness of censuses in many countries where data collection is taking place in 2020. In the United States, for instance, where the 2020 Census data collection is in progress, operations have been adapted and delayed because of Covid-19 (https://www.census.gov/newsroom/press-kits/2020/covid-19-census-bureau…). The United Nations Statistics Division maintains a web page with information on the impact of Covid-19 on censuses planned in 2020 (https://unstats.un.org/unsd/demographic-social/census/COVID-19/).

Could COVID-19 affect timelines for the 2020 Population and Housing Census?

The United Nations Economic Commission for Europe (UNECE) since 2018 promoted work on the future of population censuses beyond the 2020 round. The UNECE Steering Group on Censuses produced a draft report that is currently under electronic consultation by the members of the Conference of European Statisticians. The report discusses basically all the themes mentioned above. The current draft of the report is available on the UNECE Census Wiki at: https://statswiki.unece.org/display/censuses/Future+censuses+beyond+2020
We hope that the content of the report can provide useful inputs for the discussion on this platform, and look forward to receving comments on the report itself.