In this interview, Steffen Seibel from the Federal Statistical Office in Wiesbaden explains the changes to the sampling methodology for the next census and the differences to the 2011 census. How will the 2022 census differ from the 2011 census? What are the challenges? And why are municipalities with under 10,000 inhabitants now also being surveyed?

Mr Seibel, how was the population calculated in 2011?

In the previous census, the number of inhabitants was determined based on information from our registers – and we are going to use the same methodology for the next census. In other words, we will start by obtaining information from the population register. However, our population registers are not completely accurate, as they contain over-coverage and under-coverage. The term “over-coverage” is used to refer to people in the population register who no longer live at the address recorded there. By contrast, “under-coverage” means that people who live at a certain address are not recorded as living at that address in the population register. In the 2011 census, we used two different methods to correct those “nominal records” and “missing records”. In large municipalities with 10,000 or more inhabitants, we used a random sample to detect any nominal records and missing records and extrapolated the results to generate data for the municipality as a whole. In smaller municipalities with under 10,000 inhabitants, we only interviewed the people living at addresses where the information from the population registers did not match the results of the census of buildings and housing.

Why is a new method being used for the 2022 census?

The method I have just described is based on planning data from the census test carried out in 2001. However, the results of the 2011 census obviously give us a much better planning basis. When evaluating the data, we found that the need for adjustments in smaller municipalities was greater than we had initially expected after carrying out the census test. That is why we are planning to use random sampling in municipalities of all sizes – both large and small – to correct any misalignments during the upcoming census.

What will the samples be like?

Whenever you decide how to draw a sample – and the most interesting question is how big your sample size should be – you have to ask yourself how accurate you want your results to be. And as a rule of thumb, the more accurate you want to be, the greater your sample size has to be. From a scientific perspective, however, it would not be reasonable to aim for the same level of accuracy in all municipalities if we want to minimise the strain on the people concerned.

Why do you not aim to achieve the same level of accuracy in all municipalities?

Every sample is subject to a certain degree of uncertainty. A measure of this uncertainty or inaccuracy is something known as the “standard error”. During the past census, we aimed to achieve a simple relative standard error of 0.5% in the large municipalities. If we were to apply this relative uncertainty 1:1 to the smaller municipalities, that would mean we would aim to achieve a (simple absolute) standard error of no more than five people in municipalities with 1,000 inhabitants and a (simple absolute) standard error of no more than 0.5 people in municipalities with 100 inhabitants. This example shows that we wanted to be accurate down to a fraction of an inhabitant which, of course, is not reasonable.

This method would also lead to very high sample sizes – or even a full survey – in very small municipalities.

What is the solution?

The solution is to aim for a (simple relative) standard error of 0.5% for large municipalities with 10,000 or more inhabitants and then to gradually relax this accuracy target for smaller municipalities, so that we do not end up with a (simple absolute) standard error of five people in municipalities with 1,000 inhabitants like in the example I just mentioned, but a (simple absolute) standard error of 15 people. And we would not aim to lower this standard error of 15 people for even smaller municipalities. This will save us a lot of time and effort during our surveys. And a relaxed accuracy target does not necessarily mean a worse number of inhabitants, but merely a higher risk of the calculated population differing more from the actual population. However, this increased risk is always identical in both directions, i.e. the residual risk of undercounting the population by over xy% is always the same as the residual risk of overcounting the population by xy%.

How many people are expected to be interviewed in 2022? What was the number in 2011?

We had 9.1 million respondents in 2011. During the upcoming census, in which we will have the sample to correct any misalignments between the population register and our results, we expect 11.4 million respondents. This figure is the result of simulations carried out Professor Münnich and his team at the University of Trier. Professor Münnich also provided his scientific assistance during the previous census. Incidentally, the number is only an upper limit. The legislation adopted for the upcoming census should allow the federal states to pursue the accuracy target for municipal associations not in terms of each individual municipality, but in terms of the municipal association as a whole or a remainder of very small individual municipalities. That being the case, the calculations show that we could get by with 9.9 million respondents.
So, in those cases, the number of inhabitants won’t be determined for individual municipalities, but for municipal associations?
If this option for municipal associations and remaining municipalities in an association comes to fruition, the number of inhabitants will indeed initially be determined for a municipal association as a whole or for the remaining municipalities in the association. The number of inhabitants will then have to be broken down into the individual municipalities in each association using the information from our sample and the population register.

What can our mayors expect in 2022?

1. Unlike in 2011, the same procedure will be used in all municipalities to determine the number of inhabitants, namely the sample used to correct any misalignments.
2. By gradually lowering the relative accuracy target for smaller municipalities, we will make sure this correction sample does not put too much strain on the people concerned.
3. As a result, we will have accurate population figures for large and small municipalities and, of course, as in the last census, we will also get reliable results for other census data that cannot be obtained from the registers, such as employment and education data, for all large municipalities with 10,000 or more inhabitants and for all districts in Germany.
4. We aim to publish the results as quickly as possible. They should be available 18 months after the census date.