The smoothing method used in this Atlas followed the principle first used in the national cancer incidence atlas of Norway (Glattre et al. 1985) and further developed in the Finnish atlas (Pukkala et al. 1987). This method aims at deleting the random variation typical to observations based on small populations by showing their floating averages. Another way of cleaning a map from disturbing unreliable observations would be a Bayesian approach. This technique typically moves the SIRs or SMRs which are based on a few cases only towards 1.0. In the case of our Atlas this kind of correction would have led to a situation that colours indicating high or low incidence would not have been possible in areas with a low population density and/or small size of observational units.

The smoothing may hide some real high rates in areas with small populations. If there is some underlying knowledge of an association between an exposure and disease, it is better to pick up the incidence data for all areas where this exposure exists and analyse them by other methods than mapping.

If the population size of the areal unit is sufficient for a reliable observation there is no need to smooth. This has been the case in many published maps based on larger administrative units such as counties. Maps in the present small-area based Atlas indicate that there is often variation within the administrative areas, and showing just averages for those areas may hide aetiologically interesting observations. One such phenomenon, the variation between the main cities of an area and the surrounding less urbanised areas, was clearly demonstrated in this Atlas because we showed the rates for all cities with populations of 350,000 or bigger as such. Would it have been possible, it might have been wise to divide some of the largest cities into smaller parts because in has been demonstrated that the way of life may vary greatly within the cities as well (Pönkä et al 1993). The largest city (St. Petersburg) has a population of 5 million, i.e., more than many of the whole countries included in this Atlas.

We wanted to use a similar scaling principle throughout the book. Therefore the scale is a compromise: it should be broad enough to be used for cancers with an extreme variation, and at the same time the steps between different categories should be small enough to allow a moderate variation to become visible. Because the reference area was dominated by the Nordic countries, the scale failed in some cases to show intra-country variation in eastern parts of the maps, because the rates in east were either extremely high (stomach cancer) or extremely low (skin melanoma).

For most sites the scale based on relative differences in incidence and mortality rate worked well. Because the step between the categories was fixed, one step was equal to 15% increase in the rate, not every colour is needed for each cancer. The reader should be able to visualise at a glance, without looking at the rates in scale, whether there is variation in the rates or not.

One of the features of the smoothing technique is that the neighbouring colours in the map always exist in the order given in the scale, and there should be no problems observing to which direction the rate increases. If the colours could be in any order (like in the traditional maps in which the area of each administrative unit is coloured with one colour) it would not be possible to use so many tones.

The alternative scaling principle based on fixed absolute rates and a light-to-dark scale was also tested. The main idea of this principle first suggested by Nikolaus Becker (Becker 1994) is to allow readers an easy comparison of the magnitude of different diseases in certain areas. This principle worked well for common diseases. If one is interested to see in which parts of Northern Europe lung cancer mortality in men is higher than prostate cancer mortality, a map pair on absolute scale indicates it well. Maps using relative scale would also give a similar information, provided that the same scale would be used. The weak side of the absolute scale is on the rare diseases. If the scale is set as suggested by Becker, e.g., a comparison of mortality of cancers of the corpus and cervix uteri gives a map pair mainly indicating that cancer deaths due to both of these diseases are infrequent. It is difficult to figure out that there is large relative regional variation in these cancers, and that the geographical trends are opposite; this is obvious from maps on a relative scale. If the purpose of map presentation is also to arouse interest, a map including only one or two tones of yellow colour is not necessarily attractive enough.

We elected the light-to-dark colours to be used in maps indicating mortality/incidence ratios because the range of these ratios is (should be) restricted from 0 to 1, and therefore a fixed scale is easily usable in this case. Different colouring also makes it easy to separate these maps from the mortality or incidence rate maps. One of the good features of the light-to-dark scale is that it is also useful in black-and-white reproductions.