Population censuses in developing countries



|FRENCH MINISTRY |

|OF FOREIGN AFFAIRS |

| |

|Directorate-General for Development and |

|International Cooperation |

|Development and Technical Cooperation |

|Directorate |

|N° /DCT/F |

|Communication for the |

|International Expert Group Meeting on Strategies for Reducing Census Costs |

|Pretoria, 26 - 29 November 2001 |

| |

Population censuses in developing countries: In support of an alternative strategy combining surveys and censuses

|Executive Summary |

|This paper, which sums up the orientations adopted by French cooperation services in the field of statistics, is a contribution to the |

|studies conducted by the PARIS21 consortium on the elaboration of national and regional strategies for statistical capacity building. |

|It establishes that implementing PRSPs calls for a global review of the demographic and social information system. |

|The paper suggests reversing the current approach to this matter by giving priority to the development of a coherent system of sample |

|surveys, completed if necessary by a small-scale and rolling population census. |

|It shows that such a new approach is technically operational, given that it is possible to set up and update a sampling frame for all |

|household surveys without conducting censuses. |

|The paper suggests, in conclusion, forming a group of countries and institutions which favour this approach, with a view to setting up pilot |

|projects. |

1 finding an alternative to 10-year population censuses is essential

1 Observations

For over fifty years now, international recommendations on the collection of statistical data have advocated as a minimum for countries unable to do more the conduct of

a 5-year census on productive activities,

a 10-year population census.

These recommendations can be interpreted in different ways:

a) First, that data collection based on surveys requires a proper sampling frame. And that international recommendations for population censuses are to be regarded as the minimum standard of periodicity for updating sampling frames, whatever method is used to do so.

In the area of productive units statistics, the first interpretation has prevailed for a long time in a great many countries. The importance of managing and updating a directory of companies and their units was acknowledged, with data collection being carried out by means of periodic surveys by stratified random sampling. For large units, surveys are exhaustive and conducted on a yearly basis; medium-sized units are also surveyed on a yearly basis but sampling fractions are adapted to the number of units per stratum. Very small units are often surveyed on a multi-year basis (every 3 to 5 years). It was observed that countries having adopted that approach generally succeeded in setting up an effective information system. It was further observed that countries which kept to 5-year censuses failed to set up such a system since preparing and exploiting these censuses absorbs all their resources, except for the richer countries, obviously.

b) The second and more comprehensive interpretation holds that the exhaustive collection of vital data on all observation units is required to measure a structure properly, since sample surveys only make it possible to supplement information and assess changes.

In the area of demographic and social statistics, the second interpretation has always prevailed, as setting up and updating sampling frames was for a long time regarded as an obstacle that cannot be overcome without conducting censuses. The consequences of this are well known:

Population census costs are exorbitant. Between 1 and 1.5 times the overall annual budget for statistical systems in European countries; between three and five times in intermediate Latin American countries and between ten and twelve times in sub-Saharan African countries. These estimates provided by Statistical Institutes Directors chiefly highlight how low budgets devoted to statistics are. They also reflect the fact that census budgets are in part artificially inflated since they must make it possible to periodically re-equip Statistical Institutes. They also show that decennial population censuses are no longer simply population censuses, but have gradually become, often under the pressure of irresponsible donors, tools for large-scale surveys that claim to collect all social data. This means of course that censuses cannot be conducted in developing countries without external funding which is increasingly difficult to obtain.

Aside from budgetary aspects, what is more serious is the fact that the burden of preparing, conducting and exploiting a census greatly exceeds the capacities of National Statistical Offices (NSOs). The impact of this is twofold:

- on the one hand, many censuses are still poorly or not fully exploited and therefore wasteful of resources,

- on the other hand, Statistical Institutes now lack the capacity to develop and conduct vital and permanent statistical work.

On the basis of this observation, French cooperation services decided over ten years ago now to stop promoting and funding the conduct of population censuses and has been pressing statisticians in developing countries to study alternative solutions. It is necessary to rethink the social and demographic data collection system as a whole, both in terms of surveys and censuses, by seeking the optimum combination designed to collect all priority demographic and social data.

2 A new environment

For a few months now, it would appear that reactions to our suggestions are changing: officials are becoming aware of the rapidly changing situation stemming from the debt reduction initiatives for heavily indebted poor countries and the implementation of PRSPs.

The impact of these changes on statistical systems and on the external support they may receive from international - bi- or multilateral - donors is analysed jointly by the donors and their partners within a consortium created in the framework of the PARIS21 - Partnerships in Statistics for development in the 21st century - initiative. Through PARIS21 all partners have been made aware of three major changes:

1 PRSPs should include a programme for strengthening their information systems and their statistical systems in particular, since implementing development and poverty reduction policies calls for effective information systems.

2 The funding of statistical systems should be borne by States' budgets from the margins released by the debt reduction initiative and from the overall budgetary aid provided by the donors.

3 The funding of statistical data collection and processing by external donors will soon come to an end. The remainder of "project" aid, part of which will have been transformed into overall budgetary aid, will focus on support for statistical capacity building in the form of equipment, training, provision of expertise and trials of methods suited to data collection and processing.

We should now take a closer look at the impact of these rapid changes on demographic and social statistics.

During the first phase of a PRSP's elaboration, the results of a recent population census and of a large-scale survey of personal consumption, where available, would doubtless be very useful. However, when implementing development and poverty reduction policies and assessing their impact, this data soon proves highly inadequate. It appears essential, therefore, to put in place a permanent system for conducting household sample surveys. This is one of the objectives of the strategic programmes to be elaborated by countries for strengthening statistical systems. But this elaboration is subject to constraints both in terms of resources and funding. It will soon be apparent that, except in the rich countries, the setting-up and implementation of survey programmes is incompatible with the conduct of large-scale periodic censuses, whether population, economic or agricultural censuses. Comprehensive statistical systems must be designed to prevent any peaks in funding requirements and the use of staff and computer resources.

2 pragmatism lies in an alternative solution

There is now a pressing need to find an alternative to large-scale periodic population censuses so as to set up the sampling frames that are essential for household surveys.

This would make it possible to break the current vicious circle of necessarily large-scale and costly censuses as these are required to pursue too many objectives, given the lack of a coherent sample survey system. A lack which is due to the fact that all available resources are absorbed by censuses.

If it were possible to set up sampling frames independently of censuses, it would then be possible to build a coherent survey system as a matter of priority. This could, possibly, be supplemented by a small-scale population census should it prove essential for meeting highly specific requirements.

1 Ways of creating sampling frames without censuses

The solutions adopted by the few developed countries like Sweden which have succeeded in doing away with large-scale periodic population censuses or, as undertaken by France, in transforming them into continuing censuses, rely on the extensive use of administrative files. Clearly, these solutions are utterly inapplicable to developing countries, in particular to African countries.

A solution could be sought by exploring a completely different avenue. Starting to seek for a solution would then necessarily involve analysing techniques for selecting representative samples for household surveys.

Random sampling for household surveys is in practice always an area sampling. To start with, a random sampling of settlements is carried out; a second random sampling of the chosen observation units, such as dwellings, households and invididuals, is then carried out.

The solution must thus contain two elements:

• Way of setting up and updating a sampling frame for settlements

• Way of extracting a master sample and identifying the observation units in the settlements chosen in the master sample.

1 First stage: way of setting up a sampling frame in settlements

A sampling frame is a list which is properly updated. Three key elements are to be defined: listed units, stratification variates and weighting variates.

The list to be drawn up is a list of settlements. What is a settlement? It is a partition of the national territory centred on a village in rural areas and on a district in urban areas.

What stratification variates should be chosen? Variables that are easily observable and characterise the settlement independently of its inhabitants. They are the sort of data which figures in France in the inventory drawn up by communes, namely whether there is a school, a medical centre, a company registered with the tax authorities, a water hole, how far is the settlement from a tarred road, whether it is connected to the power network, has a market, etc.

What weighting variate should be chosen? At least one is needed. The simpler one to obtain is probably the number of dwellings.

Itemising the list is insufficient. It must be drawn up and updated.

Initially, settlements are to be recorded and identified (by giving them a non-significant identifier).

Then, it is necessary to identify and include in the list the data corresponding to the stratification and weighting variates. To do this, the easiest is to send one or two investigators to the settlement to interview the local authorities, the administrative and traditional authorities, teachers, religious representatives, etc.

Updating the list should be a continuous process, with 10% of settlements updated on average every year. Updating should take place more frequently (every five years?) for high-density migration areas, and in particular for urban areas and urban peripheries, than for rural areas (every twelve years?).

Countries in the process of conducting population censuses should take advantage of this operation to draw up such a list.

Using available resources to draw up the list of settlements could be suggested to countries which have failed to marshal the resources needed to conduct population censuses.

It should be noted that this solution is not new. It was already successfully implemented in some countries where it was called village files[1], in Togo in particular as a substitute for a population census to conduct an agricultural census. (Agricultural censuses in Africa are not censuses as such but sample surveys of urban areas households as numbered and identified by population censuses.)

2 Second stage: extracting a master sample

Once the list is drawn up, it is necessary to extract a master sample by means of a stratified random sampling (independently of the list updating plan).

The master sample would be used for X number of years, 5 for instance, and renewed by fraction (fifth) each year.

At the time a settlement enters the master sample, the following important work must be carried out to:

make a fine mapping of the settlement

and identify dwellings (each one is to be given an identifier and located on the map).

2 Building a coherent sample survey system

Once an effective sampling frame has been set up in this way, it is possible to operate a coherent network of sample surveys.

A possible and, we believe, effective strategy consists in placing a periodic survey of employment at the core of the network. It would be conducted across a panel of dwellings.

During the first visit, the demographic data on the composition of the households living in the dwellings are gathered together with the data on the occupants' formal or informal employment.

During later visits, only employment data will be gathered. To measure changes, at least two visits should be scheduled with a year's interval between them. To observe current trends, five visits should be planned with a three months' interval between them and a quarterly renewal of the panel by a fifth.

This core survey would be supplemented, every four or five years, by surveys of the informal sector, consumption, etc. In addition, variable modules concerning living conditions, health, education, migrations, and so on, could be included in the core survey.

3 Should supplementary censuses be planned?

Sample survey systems allow the gathering of all the required data with an accuracy close to that of censuses, as long as one is not seeking information at the finest geographical level.

If the gathering of information at the finest local level is given priority, it will be necessary to supplement survey systems with a small-scale population census involving simply a sex and age group count[2].

Such a supplementary census can be conducted as a rolling census, with each settlement to be surveyed during a given month during the ten-year period.

The settlements that are to be integrated into the master and other samples will thus be processed differently.

For the settlements to be integrated into the master sample, the census should take the form of a complementary survey conducted when the settlement leaves the master sample concerning only those dwellings which were not selected for one of the surveys.

For those settlements which are not to be integrated into the master sample, the supplementary census could be conducted when the list of settlements is updated.

3 proposals

We find the solution outlined to be a realistic one. It enables the management of a coherent household survey programme by eliminating all discontinuity in workloads and funding requirements.

Several American countries are highly interested in it and have asked for it to be studied as a priority theme during the next regional seminar organised by PARIS21 in Bolivia. Colombia has already volunteered to try out the proposed system on behalf of the Andean States.

We are aware that these proposals, which run counter to a well-established tradition, will meet in Africa with a great deal of reluctance. Two difficulties must be overcome.

The solution proposed is worthwhile only if implemented with great continuity over a long period of time. Are national statistical offices in developing countries sound enough to ensure such continuity? Moreover, are donors prepared to commit themselves to long-term programmes? For the time being, they prefer to back projects with a well-defined time frame.

Furthermore, the solution proposed requires competence in the fields of statistics, methodology and organisation which is much greater than that needed to conduct censuses. Do all countries have that capacity?

Should we decide to embark together on that road, it seems that we should provide for:

1) The constitution at subregional level of expertise centres to help countries implement those techniques.

2) An experiment in a few countries, in a French-speaking and an English-speaking country in Africa especially as, quite rightly, many national statistical offices will not choose that path without proof that the solution is feasible in their environment.

French cooperation services propose to all its partners to fine-tune these proposals together.

A consortium of donors, including if possible UNDP, UNFPA, the World Bank and the IMF, could be formed to fund pilot projects in two or three African countries. The States which would volunteer to carry out these pilot projects would obtain two advantages from this:

Firstly, the assurance of a significant external contribution.

Secondly, they would gain a substantial methodological edge and be the nuclei of the future subregional centres of expertise.

France, for its part, is prepared to provide full backing for those countries which will choose that path, both financially and by creating a small expert group.

-----------------------

[1] The expression "village files" is sometimes given a very different meaning by researchers who use it to refer to civil status records kept at village level.

[2] Possibly to be supplemented with a few more simple variates. The important thing is to avoid the need for individual schedules.

-----------------------

Paris, 22 November 2001

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download