The advantage of shifting projections
Demographic projections are usually reviewed periodically -every five years. This was not much of a problem if demographics changed only progressively.
But it is a problem if changes are substantial even within short periods – a phenomenon often_ observed at local level where mobility patterns are most unstable over time.
Therefore, GeoLabour introduced the principle of SHIFTING PROJECTIONS, yearly updated whenever possible. The advantages are considerable. First of all, it gives a good monitoring of the changing mobility patterns over time, and allows not to build projections on mobility patterns that are outdated.
Example 1 : The County of Yakima (WA) displays an obviously – although moderately - declining attractiveness:
based on the inward mobility rates per gender and age of the 1990-95 period, the projection suggested that age group 25-54 would have reached the 100.000 threshold by 2002 or 2003. In fact, the observed data for 2002 only reached 89.000. Using the 1997-2002 mobility profiles (and of course also the fertility and mortality rates of this latest period ) brings the population in this age group to the 100.000 threshold only in the last years
of the 2010s.
(Data source for data 1990-1992-1995-1997-2000-2002: US Census Bureau, Population Estimates, County Population Datasets; Projections by GeoLabour Projection)
In this case, the attractiveness that was still positive in the early 1990s became slightly negative in the late 1990s (see the chart right), and remained hardly positive around the turn of the century. All those elements must be taken into account in population forecast.
Example 2 : In the County of Payne (OK), it is the opposite development: while the stability of the projection based on the 1990-95 mobility profiles was leading to an almost stable number of people in the 25-54 age group, the combined shifts in mobility profiles afterwards suggest an improved attractiveness – in fact a reduced repulsiveness, due to the declining share of people in their late twenties leaving the place.
The Geolabour Projection engine allows developing a similar analysis for any county (or set of counties) and any gender or age group. From a methodological point of view, the question is the following: why should one build the projection on the average mobility rates on a long past period, as the attractiveness of places and their capacity to retain people is changing over time? We can understand the reasons why the US Census Bureau, when making State-level demographic projections, keeps a cautious attitude, extrapolating into the future the average mobility rates of the last quarter of century: doing so, all changes are smoothed down, and the role of a public institution is indeed, and expectedly, to stick to smooth projections – and even more so when making projections up to 2050.
The “shifting projection” that was preferred by Geolabour has of course both advantages and disadvantages. The advantages are that they are more reactive to recent changes in the patterns of differential attractiveness – and this is important in a country where mobility is as high as in the USA compared to most European countries. There is in fact another major reason for which a shorter reference period from the past should be preferred: we observe an unprecedented instability in the changes of age distributionsand the mobility behaviour itself is changing. For example, when the number of young people is declining, they may have less reasons to move away just because they are given more opportunities to stay where they are. So, above the argument that a shorter reference period makes the projections more reactive to changing local factors comes the argument that in times of profound demographic changes the average behaviour per age does itself change.
The disadvantage of the “shifting-projection” approach extrapolating mobility rates from a shorter reference period is that it can make the projections hyper-reactive to the recent local trends – i.e. depending on shorter term parameters such as large-scale but exceptional local developments or incidents. For example the rush towards some “burb” may fade at short notice, and this means that the projection extrapolating the rush may reveal unrealistic. This seems obviously the case for example in the Douglas county (CO), where the multiplication by eight between 2002 and 2022 can be considered as nothing else than just suggesting a very high growth rate. Similarly, based on the 1992-1995-2002 mobility profiles, who would take for granted that the population of Maricopa County (AZ) will really double between 2002 and 2022?
An additional disadvantage – that has in fact a limited impact – is that the Geolabour engine is not ensuring a permanent re-calibration of projections - i.e. ensuring that for each moment of time the sum of the projections at say the county level equals exactly the projections at the corresponding State level (or the sum of State-level projections equals the national projection).
However, there are good reasons why Geolabour can well stick to the “shifting projection” approach despite missing intrinsic re-calibration.
One reason – one may consider this one a weak reason from a pure scientific point of view – is that a systematic recalibration would make the computing process much more complicated – given a quantity of 3.140 counties adding up to 52 States which in turn make the national figure.
Another reason, to which there should be little to object – is that one cannot reduce the level of uncertainty attached to any county-level projection based on past parameters just by formally making sure that projected local level entities correctly add up to projected State-level entities. The two levels of uncertainties (local and State level) are rather complementary than substitutive. In other words: wherever the State combines a high number of county-level entities – as is the case everywhere, and as is also the case for the 50 States within the national frame – the calibration might well be theoretically and formally smarter, but its effective impact could only be marginal considering the local level. Higher uncertainty is unavoidably attached to local demographic projections. It is not a consequence of ignoring all determinants that will actually affect future changes. It is rather related to the non-deterministic nature of social systems. Omniscience would not help as it would also fail to discover and formally describe the way existing spatial configurations of human beings are giving way to intrinsically new configurations or to configurations that human beings still have to “invent”.
 
It must be added that the drift associated with the lack of recalibration usually keeps below 5% of total population over a 20-year period - being the difference between the sum of local projections and the projection of their sum at their starting point. Moreover, a drift of such a magnitude only appears where local changes are abrupt, due to irregular (high or low) inflows. The magnitude of that drift is much smaller in that majority of places where the changes in the mobility profiles are moderate and progressive.
For the decision maker, the important point remains that the Geolabour projection indicates, for the coming years, trends that take into account the recent changes – and still these trends are nothing else than … just trends.
Nevertheless, three additional points should be made – besides that common sense argument that projections only show global trends “all other things remaining equal”, which is never the case as everybody knows. The first point is that the projection horizon chosen by Geolabour never goes beyond a twenty year period: this is already far longer than what is important for most strategic decision making.
The second point is that, after all, the Geolabour projection engine based on a shorter past reference period keeps the projected changes much better in line with the observed (ex post) changes – even at the State-level. As illustrated below for two high-growth States (Arizona and Nevada), the Geolabour projection can obviously be awarded a label of fairly high probability compared to the USCB State-level projections (based on the mobility profiles of the last quarter of century). This holds true for both the 1990-1995-2000 as well as the 1992-1997-2002 mobility profile to be used for the projection.
The third point is the Geolabour projection also makes some kind of compromise: it is based on the average 5-year mobility rates of the last two 5-year periods – namely the 1992-1997 and 1997-2002 periods. This means that, to some extent, it is also smoothing down the projection, instead of just considering, say, the last year or the last three years. It thus also avoids exposing itself too much to the dependence of incidental factors. It is a compromise just as the fact of considering 5-year age groups instead of year-to-year cohorts.
So, Geolabour claims that its projection methodology, allowing annual updates based on the annual publication of “estimates for the year before” by the US Census Bureau, represents an “acceptable compromise”. It draws its main advantage from the fact that it effectively helps strategic decision making better than any other available method. And it offers no alternative as to local projections.
Under the very condition that all “caution warnings” are well understood, it aims to help to take more fully into account the profound demographic changes that are ahead.
Last but not least, it must be added that the “confidence” that one can have in the projections is directly proportional to the size of the entity that is considered, i.e. to the overall size of the local population. It is of course a simple question of inertia, whereby small communities may react to any given incidental factor (closing a plant for example) in such a way that it will have a high impact, while a similar factor would only change the decimals in the larger entity.
This is also why, both in the Atlas of Prospective Labour Supply and in the Demographic Audit® and Location Rating® tool, the clustering of counties into Metropolitan Statistical Areas (MSA) was systematically given as an alternative: the projections were made starting from the sum of the corresponding counties and they do not refer to any ex-post sum of county-level projections. The Demographic Audit® and Location Rating® tool allows the user to define tailored areas, made up of any number of counties, and projected values are given by a projection of ex ante sums, not as an ex post sum of projections – as a way to gain some leeway considering the limitations linked to small entities.
More technical details on Geolabour’s projection methodology in
ATLAS OF PROSPECTIVE LABOUR SUPPLY Or in the help file of DEMOGRAPHIC AUDIT® and LOCATION RATING®
|
