In the study and research of popula2on growth dynamics, selec2ng a mathema2cal model

to extrapolate, predict and model data is crucial for represen2ng growth pa9erns in various
contexts. Popula2ons will flu9er due to an array of factors: resource availability, biological
interac2ons, physical space etc and thus needs a model that reflects and takes into account
all those factors. This IA will a9empt to inves2gate, evaluate and judge different
mathema2cal models for growth, par2cularly logis2cs and exponen2al models. By analyzing
advantages and limita2ons, of each model and how they build upon each other, this
inves2ga2on aims to iden2fy which model best captures the complexi2es of real-world
popula2on growth pa9erns and trends. It’ll use UAE as the focus data set for the models and
inves2ga2ons, as UAE will serve as an excellent backdrop for popula2on analyzing due to its
standard popula2on growth rate.

Popula2on of UAE:
Years since 1970(t) Popula6on(millions) (p)
0 0.24
5 0.55
10 1.0
15 1.4
20 1.9
25 2.4
30 3.2
33 3.7
35 4.6
36 5.2
38 6.9
40 8.3

Sca9erplot of given data points:

If we were given just the data set above, some may assume that the points take an
exponen2al pa9ern, following an exponen2al line of best fit. Hence, we can take our given
data points to extract it in exponen2al form:

𝑃 (𝑡) = 𝑎 × 𝑏 !
By trying to subs2tute values t=0 and t=10 we can use simultaneous equa2ons to find a and
b approximately:

1. 𝑎 × 𝑏 ! = 0.24
2. 𝑎 × 𝑏"! = 1.0

è 𝑎 × 1 = 0.24
è 0.24 × 𝑏"! = 1.0
!" ".!
è 𝑏 = *!.$%

∴ 𝑎 = 0.24, 𝑏 ≈ 1.15 , 𝑃" (𝑡) = 0.24 × 1.15&

This specific value of b, 1.15, indicates this exponen2al func2on’s constant mul2plier. The
constant mul2plier in an exponen2al func2on can dictate whether the graph indicates
exponen2al growth or decay, or how the output values behave as 𝑡 → +∞. When b is
greater than 1 like in our case here, the func2on models exponen2al growth, meaning the
output values increase rapidly as 𝑡 → +∞, hence the b value represents a growth factor.
Conversely if 0 < 𝑏 < 1 the func2on would model exponen2al decay causing the outputs to
decrease rapidly as 𝑡 → +∞, and the b value would indicate factor of decay.

Let’s test the model’s accuracy by predic2ng the popula2on with the model and comparing it
to the actual value, using the year 2000 (n=30) as reference.

0.24 × 1.15'!
≈ 15.9

Absolute error: |3.2 − 15.9| = 12.7

"$.(
Percentage error: '.)
× 100% = 326%

Instead of subs2tu2ng values to find an exponen2al, we can instead use a built-in

exponen2al regression model to take into account all points rather than just 2 points. This
makes the graph significantly more accurate, as you are comparing equa2ons derived from 2
points compared to infinitely many points.

𝑃$ (𝑡) = 0.355 × 1.08&

Here are the sta2s2c calcula2ons for the regression above:

𝑎 = 0.355
𝑏 = 1.08
𝑟 $ = 0.970
(𝑥, 𝑟𝑒𝑠𝑖𝑑𝑢𝑎𝑙) = (−0.115,0.0295)
(𝑥, 𝑇𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚𝑒𝑑 𝑟𝑒𝑠𝑖𝑑𝑢𝑎𝑙) = (−0.392,0.0551)

The values in the above indicate that the model is a good fit, in that it is closely correlated to
the points given. We know this not only by seeing how close 𝑟 $ is to 1, but we can also see
how small the residual and transformed residual values are. Because we are looking at non-
linear values, we will look at the transformed residual value (0.0551).

The transformed residual comes from:

𝑅𝑒𝑠𝑖𝑑 𝑡𝑟𝑎𝑛𝑠 = ln(𝑝*+,-./-0 ) − ln (𝑝1.-023&-0 )

When the data is linearized as a rela2onship for easier calcula2on, the original residual
values must be transformed, hence why we have to use the transformed residual values in
our case. The value of 0.0551 is considered low, and since it is the difference between
certain observed and predicted output values that are natural logged, it means that they are
very closely related, and the model fits the data set well.

The given b value indicates this models growth factor, and it is less than the first model. This
means that although they follow along an exponen2al pa9ern, the first model increases or
grows at a faster rate than the second one. Hence why despite star2ng off at a lower y
intercept, post intersec2on at (6.19,0.57) or when 𝑡 > 6.19 , 𝑃" (𝑡) > 𝑃$ (𝑡).

Given that we have discussed that the model is quite reliable for the first few values, we can
try extrapolate for the t values 43, 45, 47.

Model given values: Actual values:

Years since Years since

Popula2on(P) Popula2on(P)
1970(t) 1970(t)
 43 9.52 43 9.0
45 11.1 45 9.2
47 12.9 47 9.4

Here is a sca9erplot aaer adding the new points to the data set:

Although at first it seemed as though the points in the data set fit the exponen2al models,
past a certain point the data points seem to plateau off heading towards what looks to be
around p=10. Exponen2al points are supposed to showing growing changes in output as x is
increases in the same increment, yet we see the opposite, we see a tapering off. This is likely
because popula2on realis2cally cannot con2nue to grow exponen2ally, as in each country
there are a finite number of resources, money, land and space. Resource constraints make
the exponen2al model unrealis2c. Furthermore, despite UAE’s very rapidly growing
economy in recent years, the popula2on cannot con2nue growing indefinitely, like the
exponen2al model suggests; it must have some sort of carrying capacity.

Due to this fact that popula2on must eventually plateau, I posit the logis2cs model.

16.3
𝑃(𝑡) =
1+ 𝑒 4!."!$(&4%$.%)

Here are the sta2s2c calcula2ons for the regression above:

𝐿, 𝑡ℎ𝑒 𝑐𝑎𝑟𝑟𝑦𝑖𝑛𝑔 𝑐𝑎𝑝𝑐𝑖𝑡𝑦 𝑜𝑟 𝑡ℎ𝑒 𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑠𝑖𝑧𝑒 𝑡ℎ𝑒 𝑒𝑛𝑣𝑖𝑟𝑜𝑛𝑒𝑚𝑛𝑡 𝑐𝑎𝑛 𝑠𝑢𝑠𝑡𝑎𝑖𝑛 = 16.3
𝑘, 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑔𝑟𝑜𝑤𝑡ℎ 𝑟𝑎𝑡𝑒 = 0.102
𝑡! , 𝑖𝑛𝑓𝑙𝑒𝑐𝑡𝑖𝑜𝑛 𝑝𝑜𝑖𝑛𝑡 𝑜𝑟 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑤ℎ𝑒𝑟𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑟𝑒𝑎𝑐ℎ𝑒𝑠 ℎ𝑎𝑙𝑓 𝑖𝑡𝑠 𝑐𝑎𝑝𝑖𝑐𝑡𝑦 = 42.4
𝑟 $ = 0.972

In contrast to the exponen2al model, the logis2cs model is highly suitable to model
popula2on growth. In par2cular it incorporates carrying capacity showing that growth slows
and prevents the unrealis2c projec2on of endless expansion. Furthermore, the model takes
a S shaped (sigmoid) curve and hence captures 3 dis2nct growing phases: ini2al slow
growth, rapid growth, and decelera2on. This pa9ern correlates very closely to the real
dynamics of popula2on growth, as naturally popula2ons grow fast when resources are
abundant, yet stabilize when resources gradually become more limited.

Moreover, the logis2cs model is clearly asympto2c as

𝐿
lim 𝑃(𝑡) =
&→8 1+0
=𝐿
Which explains that unlike the exponen2al model, the logis2cs one actually has a plateau or
carrying capacity and cannot physically handle growing popula2on indefinitely.

Lastly, the model shows a point of inflec2on or a certain point at which the concavity of the
01
graph will change, without being a maximum or minimum point, i.e. 0& = 0. It shows a clear
point at which growth begins to slow, and its insights certain cri2cal points of popula2on
growth, that can be analyzed to accommodate resources by governments, and make sure
they aren’t over or underproducing/ impor2ng.