3.3 The quality of statistics

In order to ensure the accuracy of interpretation, it is important to assess the quality of statistics. Statistical representations of the world around are always indirect, and therefore the results are bound to be influenced by various kinds of errors and methodological solutions.

In Finland there are established quality criteria for official statistics which require that a quality description is attached to all statistical publications. This allows users to determine whether the statistics concerned are suited to their purposes.

1. The information must be suited to its purpose

Statistics are used for various different purposes and therefore it is important to give as detailed an account as possible of any restrictions with regard to their utility. For example, statistics based on data about taxation do not give a complete picture of people's income in Finland because not all income is taxable. The utility of statistics is also influenced by the concepts and classifications used, and therefore their definitions must be given.

2. The information must be as accurate as possible

No information is completely error-free because it would be hugely expensive to eliminate all errors. It is important therefore to inform users about any possible sources of error and the magnitude of the error.

Possible sources of error in statistics include the use of samples and the selection of the research material (e.g. non-response); the instrument or index used (e.g. questions are misunderstood); and data processing. Errors cause either random variation or bias, i.e. systematic distortion in the results.

The quality of measurement results

The quality of measurement results can be described by three different marksmen's hits on the target. In the target on the left, the hits are heavily scattered, giving a poor overall result. In the middle target, there is little scatter but all hits are slightly wide of the mark. In the target on the right hand side, the marksman has consistently hit the middle and there is not much scatter. The validity of measurement results can similarly be undermined by high scatter due to the failure of the instrument to produce very accurate results. There may also be systematic bias in the measurement results. For example, results from interview surveys aimed at determining levels of alcohol consumption in the population are systematically lower than the true figures.

3. The information must be as fresh as possible

Statistics always describe things that have already happened. When a set of statistics is based on administrative data registers, it may sometimes take a very long time before the figures are actually released. For example, a major source for income statistics is provided by information on taxation, and these figures are not completed until the autumn of the following year in question. Usually the aim is to publish annual statistics during the course of the following year; quarterly statistics during the following quarter; and monthly statistics during the following month.

It is also important that there is advance information on the publication of statistics. In Finland the publication dates of the most important statistics are usually given one year ahead of time (see the Release calendar for statistics at www.stat.fi).

4. The information must be readily accessible

To make sure that key statistical information is accessible to everyone, large amounts of data are made available free of charge (see www.stat.fi). It is not enough simply to publish statistical tables, but it is also necessary to have a description of the methods of data collection and the most important findings.

5. The information must be comparable internationally and over time

It is by no means easy to achieve comparability in every instance. Societies change with time so that old concepts are no longer applicable to describing them. Similarly, concepts that are used to describe our society today are not necessarily applicable to societies of days gone by. The same problems apply to international comparisons. For example, it is difficult to compare housing costs in different countries because there are very different systems in different countries for the financing of housing: in some countries rented accommodation is favoured over owner-occupied housing, in other countries vice versa. In some societies there are strong mechanisms of housing subsidies, in others there are not.

To allow for maximum comparability, it is crucial that statistics are based on universally accepted concepts and classifications. Statistics Finland works constantly to keep up to date its statistical classifications (occupational, industrial, educational, etc.). Key concepts and classifications are often agreed upon internationally. For example, national accounting concepts have been approved by the United Nations. Work is also ongoing in the EU to develop a common set of concepts and definitions for European statistics.

6. Statistics shall be as consistent and uniform as possible

Because they have different end-uses, statistics inevitably differ from one another. One example that has received much attention in Finland is the difference between the unemployment figures published by Statistics Finland and the Ministry of Labour (see the Labour Market Statistics topic). Statistics Finland's aim is to produce an internationally comparable unemployment figure, whereas the Ministry of Labour keeps records on the clients of employment offices. Whenever there are differences between statistics, their reasons must be brought to the end-users' attention. The same concept must not be used in different statistics in different meanings.