There is little doubt that the world population will grow for quite some time, as being projected by the most recent UN World Population Assessment. Of course we can imagine massive natural catastrophes such as the world being hit by a huge meteor; we can also speculate about the emergence of a highly contagious lethal virus for which no cure or immunization can be found (Garrett, 1994); or we might fear a worldwide nuclear war that would result in sudden, non-reversible climate change - but short of these highly unlikely events (see: Budiansky, 1995) nothing could stop the global population from increasing another few billion people. Why are we so sure about this?

First, there is a driving force concealed in the "young" age structure of the world population that just cannot be switched off (see Figure C2_1). Due to high fertility in the 1950s, 1960s and 1970s in many developing countries large numbers of women (and men) are currently entering reproductive age. The world is full of young adults that will have children. Even if each couple has a smaller number of children than their parents the total number of offspring will be substantial. This "echo effect" of a high-fertility period in the past creates a "population momentum" which works against changes in reproductive behavior that favor smaller families.

Second, it is highly unlikely that large populations will change their reproductive behavior instantly. Certain sections of a population, such as highly educated middle-class couples in urban areas, might adopt radical behavioral change almost overnight, but many developing countries still have large rural populations where fertility is linked to deep-rooted cultural values or social conditions and can decline only gradually over two or three generations. We must also take into account that the average fertility of a population is a composite measure which results from the reproductive behavior of several parent cohorts: these include couples which already have a certain number of children and can only reduce the number of additional offspring. Even in a country like China, where we have the most rigorous family planning program and a highly controlled society, it took 20 years to reduce average fertility from about 6 to 2.4 children. In India - according to UN projections - this process might take 60 years or more.

These two basic facts, which are well known among demographers, tend to slow-down demographic change. They can produce a considerable time-lag between the first signs of a fertility decline and a slow down of population growth. In fact, it is quite typical for developing countries that the total number of birth increases for one or even two decades, while the fertility (that is the average number of children per women) already declines.

This "paradox" of population growth during a time of fertility decline is simply a consequence of the fact that the increase in the number of parents outpaced the decline in fertility. In fact, this situation will continue for some time to come. According to the most recent UN projections we will have a stable annual increase of about 80 million people until 2015 - only then will this increase gradually decline to about 47 million in 2050. By the middle of the next century the world population will still grow by about the same number of people as in 1950 - only the total number of people on the planet will be more than three times larger.

Table C2_2

In table C2_1 (a and b) I have compiled results from various world population projections, including the most recent IIASA World Population Scenarios and the 1996 edition of the UN Population Estimates and Projections. There are several remarkable results:
The "medium" or "central" variants of all projections for the year 2000 are very close - no matter whether these projections were prepared in the early 1960s or mid-1990s. For instance, the UN projection from 1993 projected the 2000 world population at 6.13 billion. In 1996 the US Bureau of the Census (International Programmes Center) published its "World Population Profile" with an estimate of 6.09 for the year 2000. In 1981 Frejka projected a 6.2 billion world population for the year 2000 and in 1983 Keyfitz projected a 6.08 billion population.
For the year 2025 we can compare two UN projections, one from 1980 and the most recent from 1996: the first has projected the 2025 population at 8.2 billion, the second at 8.04 billion.
There is also remarkable little variation in the selected "medium" variant projections for the year 2050: In 1981 Frejka projected the 2050 world population at 9.89 billion; the most recent UN projection in 1996 was 9.37 billion. Only Keyfitz had a more "optimistic" estimate for the year 2050 world population: 8.68 billion.
Table C2_1 also shows the extremely wide range of "variants" or "scenarios" in some of the projections. The most recent IIASA World Population Scenarios, for instance, include scenarios that range from a projected population of 7.1 billion to 13.3 billion for the year 2050. The most recent UN projections, on the other hand, have a much lower range of output: the low variant projection for 2050 is 7.26; the high variant is 8.08 billion.

As already mentioned, the United Nations Population Division has a tradition of fine-adjusting their assessments with each new round of world population projections. This should be seen as a strength rather than a weakness in their approach. In their most recent edition of the World Population Trends the UN has somewhat reduced their projections. It is interesting to see in which countries the UN thought it necessary to make the biggest adjustments. Table C2_2 presents only the differences in population estimates and projections (both as an increase or decline in the number of people and as a percentage of the 1994 assessment). The biggest adjustment was made for India: the 1996 edition has a projection for 2025 population which is almost 62 million smaller than in the 1994 edition. This is a 4.4% lower projection than in the 1994 edition. Obviously the UN is a little more "optimistic" that India will be able to reduce population growth than in the 1994 assessment.
As one can see in Table C2_2 the UN not only adjusts its projections according to new trends in fertility, but also revises the historical population estimates. The population of Russia in 1950 was estimated more than 1 million lower in the 1996 edition, as compared to the 1994 edition.
Finally, it is interesting that most of the differences between the 1994 and 1996 edition of the UN World Population Trends is due to adjustments in only 12 countries. The world population in 2050 is projected to be some 466 million less in the 1996 edition than in the 1994 edition. Almost 400 million (of this 466 million difference) is due to adjustments in the 12 countries listed in Table C2_2.

1. Replace population projections by population scenarios.
2. Develop more sophisticated projection techniques - including probabilistic projections.
3. Live with the problem, but regularly revise projections.

While the scenario approach is certainly useful for better understanding the consequences of certain demographic assumptions, it also has serious shortcomings:
In its most rigorous form it does not help the user to choose between the different scenarios. They are just presented as "possible alternative futures". The user of the projection has to decide which set of assumptions (scenario) is more likely. This is particularly frustrating when a large number of scenarios is presented - resulting in extremely divergent demographic trends. In essence, a strict scenario approach is an attempt to transfer responsibility for the critical decisions from the demographic expert to the user, who is usually a non-demographer.
This vagueness and ambiguity in demographic scenario modeling would hardly be tolerated in other disciplines. For instance, what would a pilot say, when the weather service would tell him that the destination airport could be either covered under 2 meters of snow or be flooded from heavy rain - depending on scenario. We expect that meteorologists make up their mind (after consulting the most sophisticated models) and deliver a clear prediction of the weather that can be proven wrong. That is why weather forecasts are so popular: the experts' predictions can be falsified by everyone.

Knowing about the difficulties of users to choose between a broad range of alternative scenarios some demographers have tried to define "most likely" or "medium variant scenarios. This, actually, is in no way different from the traditional approach of population projection. In practice, everyone will use the "most likely" scenario results as if they were some kind of "old-style" medium variant population projection.

In essence, the scenario approach in population projections is a "didactic tool" for educating the (often non-demographic) audience about the not-so-obvious consequences of certain demographic assumptions. However, it does not solve the basic problem of predicting human reproduction or migration - namely the possibility of radical and intentional changes in human behavior.

One can use information from past changes in fertility, mortality and migration to estimate the probability of certain changes in the future. For instance, one could use the distribution of inter-annual changes in fertility during the last 20 years to estimate the probability that the fertility rate will increase or decline by a certain margin in the future. If, for example, the (age specific) fertility rates only dropped by a small margin from year to year it is unlikely - according to this approach - that it will jump up by a huge margin in the future. There are several highly sophisticated variants in using time-series data to calculate probabilities for future changes in the demographic components (Lee, 1974 and 1992), but none of these methods does anything to solve the basic problem that the past does not necessarily give us hints about the future. Moreover there are numerous technical problems, such as where to begin the time-series for the calculation of the probability estimates: Should they go back before the "baby boom"? Should they be only for the most recent few years? If we take into account fertility data from the "baby boom" and "baby bust" years we automatically end up with a much broader probability distribution for our estimates of future fertility.

The second approach in probabilistic population projections is a variation of the scenario method. A panel of demographic experts is asked to estimate the probability of certain changes or levels in fertility, mortality or migration (Lutz / Sanderson / Scherbov, 1996). These probability estimates are then used to calculate a population projection with (various) confidence intervals. Unfortunately, it is by no means easier to estimate the probability of a certain future fertility range (or level) than just to predict a specific, most likely, level of these demographic component. How can one say, for instance, that with 95% probability Germany's TFR will be in the range between 1.1 and 1.9 in 2030? This "point-based" probability estimates completely depend on the ability of demographic experts to translate their intuitive "best guess" estimate of future fertility, mortality or migration levels into (quantitative) probabilities. Cognitive research tells us that humans are particularly bad in estimating the probability of certain events. For instance, many people are frightened to fly because they believe crashes are quite frequent - not realizing that driving their car to the airport is the real risk.

Probability projections use - sometimes excessively - methodological sophistication in order to solve the principal problems of predicting human behavior.Unfortunately, they either depend on information extracted from historical time-series data or they must rely on the cognitive capacity of experts to estimate the probability of future events. While they are certainly more complicated than conventional projections, it is rather questionable whether they are more accurate.