Alternative approaches to census design

As part of their preparation for the 2010 global round of population and housing censuses, some countries are developing, testing, and implementing alternative techniques for collecting, processing and disseminating key statistics that used to be generated by the classic approach to population and housing censuses. Even so, the crucial principle of providing detailed statistics at the lowest geographical level remains of paramount importance.

On this website you can read descriptions of each of the alternative approaches, as well as the census plans of specific countries utilizing each approach. When a country has contributed information about its census plans, that information is part of a menu selection following the general design description. Countries have been asked to describe the census design they are using, identify the extent to which this national exercise meets the four essential features of a population census, discuss the necessary conditions for implementing the method (including legal, policy, and technical issues), and provide Internet links providing additional information and other parameters.

If you would like to submit your country’s census depiction to this website, please send it by email to Jay Keller at jay.k.keller@census.gov.

The classic approach

During the 2000 round of censuses, over 190 countries conducted a population census and an overwhelming majority utilized the classic, or traditional, approach to a census. It is a complex operation of actively collecting information from individuals and households on a range of topics at a specified time, accompanied by the compilation, evaluation, analysis and dissemination of demographic, economic, and social data pertaining to a country or a well-delimited part of the country. Members of the public respond to a census questionnaire, or interviewers are deployed to collect information from respondents. For interviewer-based censuses, enumerators assigned to different enumeration areas cover all households and persons in the enumeration area during a specified and usually short period of time to meet the requirements of universality and simultaneity. Both short and long forms may be used within the context of traditional censuses. The short form contains only questions intended for universal coverage, while the long form is used to collect information only from a sample of households and population. This form usually contains detailed questions on a particular topic in addition to covering complex topics such as fertility. Both are utilized during the same time frame of the census, with no content data collected outside of that time frame. While the long form estimates are not based on full coverage they are regarded as census output.

Because various methods can be used for collecting the data, including a mailed or dropped off questionnaire, the telephone, the Internet, personal visit follow-up, or a combination of such methods, countries employing the classic design may utilize very different methodologies in doing so.

The traditional census has unrivalled merit in providing a snap shot of the entire population at a specified period and the availability of data for relatively small geographic domains. In that sense the traditional census is perhaps unique in nature. This approach is particularly suitable for the countries having a federal structure and having the requirement of producing population numbers by various social and economic characteristics simultaneously for all geographical levels to meet the needs of planning and allocation of funds. The delimitation of electoral boundaries also demands simultaneity, and for that reason also the classic approach may be better. But at the same time, traditional censuses have been singled out as the most elaborate, complex and costly data collection activity that national census offices undertake. In addition to costs, this complex task requires full awareness and agreement of the public to participate in it. Because of their complexity and expense, such censuses are usually mounted only once every five or ten years, so that census data is often several years out of date.

Countries depicting their use of the classic approach:

The register-based approach

The concept of producing census-like results based on registers emerged in the 2000 round of censuses, although it has been debated and tested to various degrees since the 1970’s, and several countries succeeded in using this approach to generate census data in the 1990 round of censuses. The underlying philosophy in this concept is to take advantage of the existing administrative sources, namely different kinds of registers of which the following are of primary importance: registers of households, dwellings and individuals. In the next iteration these are linked at the individual level with information in business, tax, education, employment and other relevant registers. While it is theoretically possible to link the records on the basis of the name of the individuals, the existence of a unique identification number for each individual, household and dwelling is of crucial importance as it allows much more effective and reliable linking of records from different registers.

One of the essential preconditions of this approach is that the country should have an established central population register of high quality and good coverage linked with a system of continuous updating. In the case of local registers, continuous updating along with communication between the register systems must be good. Quality assessments should be conducted. If these conditions are not met, the country should rely on the population census as the primary source of benchmark population statistics.

The primary advantages of a register approach are reduced cost for the census process and greater frequency of data. However, establishing and conducting administrative registers involve higher costs than the census itself may justify. It is a more useful and effective administration that must prove the need of a register, not the statistics alone. The use of administrative data sources also involves certain drawbacks that need to be taken into account. One such drawback is the fact that register-based descriptions have to rely exclusively on the information contents that can be formed on the basis of the registers available. In addition, in not a few countries, registers are legally restricted to use for another purpose such as making statistics. This imposes some restrictions with respect to characteristics that are available for description, and may also undermine international comparability.

Combination of classic and register-based approaches

Countries whose registers on different topics are not deemed equally developed or reliable enough to produce all required census results, may however consider the use of the available information contained in them as a basis for a complete census enumeration. For the 2000 round of censuses some countries produced partially pre-filled forms showing address of dwelling and names of occupants, obtained from a population register, as well as other recorded information. Such information was then ratified or corrected at the time of the interview. Required but not recorded information was simultaneously gathered by the interviewer.

In this fashion, average interview time is substantially reduced which has a positive impact on costs. Respondent burden is kept at a minimum thus improving on data quality. Also, coverage can be assessed through comparisons between the population register and field-work information.

Furthermore, the quality of the registers can in turn be assessed through similar comparisons and, for a limited number of variables, improved by appropriate updates based on information collected in the field. This may make possible the production of more frequent inter-censal results based solely on registers considered to be of good enough quality. The periodic collection of this type of census may be considered as a stock taking activity which allows for correction in the population numbers that would be obtained directly from the registers and which are necessary due to the occurrence of un-registered events as, for instance, those related to migration.

Finally, since the approach is based on universal coverage, it shares the advantages of the traditional approach: providing a snap shot of the entire population at a specified period and the availability of data for relatively small administrative domains. Countries with experience in this type of census claim that some of the drawbacks of more traditional exercises are reduced.

Register-based censuses with sample surveys

In some countries register-based censuses are complemented by sample surveys. This is the case when registers do not contain all the information for a full-fledged register-based census. This approach combines records from registers with the data collected via sample surveys and links them at the micro-level, that is, the record level. In the next iteration, depending on results, data are tabulated or extrapolated. This approach appears to be akin to the population census’s short form/long form paradigm: a set of basic topic is enumerated from each individual and households and the set of more detailed topics on the sample of individuals and households and then the results are extrapolated. Thus, the recommendations on the use of sampling as they are now presented in the Principles and Recommendations for Population and Housing Censuses would apply for this model, with some modification, since there is a need to elaborate on the weighting procedures.

The main advantages and disadvantages of this approach are similar to the purely register-based approach. . There are almost no objections to a virtual census and the non-response problem only plays a role in the surveys of which the data are used. If non-response can be corrected in a survey, it will certainly be possible to correct for the selectivity of that survey in the census where it is used. The disadvantage is that it involves more work to produce the tables from the microdata as weighting problems may arise. Moreover, it may be more difficult to get attention for the census results when people are no longer interviewed.

Countries depicting their use of register-based censuses with sample surveys:

-

-

The rolling census approach

A “rolling census” represents an alternative to the classic model of the census by means of a continuous cumulative survey, covering the whole country over a long period of time (generally years), rather than a particular day or short period of enumeration. The two main parameters of a rolling census are the length of the period of enumeration (which is linked to the frequency of updates required) and the sampling rate (which depends on the available budget and the geographic levels required for dissemination purposes). For example, it is possible to build a sample framework in order to produce national results with one annual survey, regional results by cumulating three annual surveys, and small areas results by cumulating five years. Annual surveys may be conducted over the full course of the year or in a particular month or other shorter timeframe.

Implementation of such an approach requires highly complex sampling and modeling techniques; a high quality sampling frame in order to allow sampling at very low levels of geography (a master address file updated annually is indispensable); and successful consultation about the approach with major stakeholders, including national and local governments and the user community. The main advantage of this approach is the higher frequency for updating data: a traditional census provides an update every five or ten years, whereas a rolling census provides annual updates. Another advantage is in smoothing the burden of the census, instead of the high cost and labor requirement of a traditional census. Further, it is possible to improve the process year after year, and test new technologies. The central disadvantage is that this approach no longer provides a simultaneous snapshot of the whole population, complicating comparisons between areas due to different enumeration times, even if data collected at different dates are adjusted to have the same reference period.

Countries depicting their use of the rolling census approach:

-

Traditional enumeration with yearly updates of characteristics

This design is a variation on the traditional census design and focuses on counting the population and collecting only the basic demographic data in the census year. A very large household survey collects and tabulates detailed demographic, social, economic, and housing data every year throughout the decade, replacing a census-year long form to collect this detailed data from a sample of the population. It may not be necessary to collect data on all topics every year, since requirements of such data may vary from country to country. The survey samples a percentage of addresses each year to approximate a long form sampling rate over a certain period of the census cycle, such as five years. To improve the reliability of the estimates for small governmental units, a larger proportion of addresses are sampled. The sample is cumulated over time to produce the lowest levels of geographic detail similar to the long form sample in the traditional census. Survey data are weighted to reflect the sample design, to adjust for the effects of nonresponse, and to correct for survey undercoverage or overcoverage. This final weighting adjustment helps to ensure that estimates of the characteristics are comparable to the standard, which is the periodic census. Once the final weights are applied, the statistics are generated, including population estimates, proportions, means, medians, and ratios.

The primary impetus for this approach is twofold - to provide more frequent and relevant data on the population than is available when a census is conducted only once a decade and to reduce the operational risks associated with the census. However, such a program is costly and technically difficult to mount, and requires a multi-year program of comprehensive planning, development, and testing. Particularly in countries with legal requirements for complete counts of the population at intervals, the complete count component of the census design is crucial.

Countries depicting their use of the traditional enumeration with yearly updates approach:

-

-

Modeling-based approaches under development

It is evident that the suggested census alternatives address, with varying degrees of success, one or more of the following concerns for the production of information, dealing with (a) small areas, (b) a wider range of topics, (c) each studied in greater detail and from different perspectives, (d) more frequently but (e) at ever decreasing production costs. Should it prove impossible, for the time being, to develop a full system of automated registers, NSO’s may find useful to consider modeling and estimation techniques.

Over a relatively short period of time information on a select group of questions is collected for each and every individual, household or dwelling. Sample surveys, taken during the intercensal period and possibly also simultaneously to the universal enumeration, are used for the collection of that information and on a number of additional questions required to adequately characterize both already touched upon topics as well as additional ones. Sampling results are used to model the relationship between information universally collected and sample questions. The models so developed are used to fill the information gap that usually results from sample surveys, particularly at the small area level, using census data as explanatory variables. Experience related with this proposal may be found in the field of poverty studies.

Among the disadvantages of this approach one can count the more complex field operation when simultaneous collection takes place.