Population forecasting methods

SUDHA GOEL, PH.D. (ENV. ENG.)

CIVIL ENG. DEPT., IITKGP
KHARAGPUR 721 302
 Population forecasting methods

1. Arithmetic progression
2. Geometric progression
3. Exponential increase (best for developing countries and urban areas)
4. Incremental increase
5. Changing rate of increase method
6. Ratio method
7. Graphical method (curve-fitting and extrapolation)
8. Logistic curve method
2

Goel, 2019; Garg, 2001

 Arithmetic Growth Model

Pn = P0 + nX

Where

P0 = population at time t = 0

X = fixed increment in population, usually annual or decadal

Pn = Population at the end of n increments of time; these can be months, years,

decades, etc.
Best application for controlled growth of campuses, towns or cities.
 Geometric growth model

Pn = P0 [1 + r/100]n

Where P0 = population at time t = 0

Pn = Population at the end of n increments (can be months, years, decades, etc.)

r = constant rate of growth, usually annual or decadal

Discrete function; no limits to growth; applicable to money in a bank account or
population growth in developing countries
 Exponential growth model

P0 = population at time t = 0

Pn = Population at the end of n increments (can be months, years, decades, etc.)

• r = constant rate of growth, usually annual or decadal.

• Continuous function; no limits to growth – runaway growth!; applicable to money in a

bank account or population growth in developing countries
Comparison of the three basic models
 Incremental increase method
This method is based on the arithmetic progression method and gives a higher value
than the simple arithmetic growth model due to the additional term
In this method, two terms are added to the last known population:
a) Average increase in population for the entire data period (X = ∆Pavg); and
b) Average of the increments (Y = ∆(∆Pavg)avg) during the period of data available.
All methods are illustrated with a single set of data shown in Table 2.7 (Goel,
2019).
The equation used to forecast future population is
P = P0 + nX + {n(n+1)/2}*Y
 Changing rate of increase method
• This method is similar to the geometrical progression method but the rate of increase is
assumed to change instead of remaining constant.
• Unlike the logistic growth curve, the changing rate of increase method is used for non-
monotonic changes in growth rates.
• This method may be useful for forecasting populations of large cities when their population
growth rate starts decreasing or fluctuating over time.
• The equation for forecasting population using this method is the same as for geometric
growth with only one difference, i.e., r is not a constant:
• P =P0 (1+r/100)n
• r = X – nY where
• X = last percent increase in population; Y = average percent increase in population
 Ratio method
• In this method, the local population and national population data for the last four to five
decades of the town or city for which population is to be forecast are taken into
consideration.
• Ratios of local population to national population are then calculated and a graph of
these ratios with time is plotted. The graph is extrapolated to get the ratio for a future
point in time and then it is multiplied with the expected national population at that time.
• Thus, the town’s or city’s expected population at any time can be calculated. This method
can be used to forecast relative population growth, e.g., urban versus total or rural
versus total.
 POPULATION GROWTH
• World Population = 7.9 billion (as on 8 Dec 2021*)
• India’s population = 1.21 billion (Census, 2011)
• West Bengal’s population = 91.3 million (Census, 2011)
• Kharagpur’s population = 3.0 lakhs (Census, 2011)

If Census data for India from 1911 to 2011 are used, we can assume exponential
growth (good assumption since R2 >0.95)
• Average annual total population growth rate = 1.75%
• Average annual urban population growth rate = 2.85%

10

Worldometers.info/world-population
http://www.census.gov/popclock/
11
 CURVE-FITTING EXERCISE
1. Fit the data to any equation - can be linear, exponential, polynomial, logistic growth or any
other equations.
2. In Excel, use the trendline option and choose to get R2 and equation. Do not force the curve
to go through zero or origin.
3. Coefficient of determination, R2 is used to test goodness-of-fit.
4. Coefficient of determination, R2 = 1 – RSS/TSS where RSS = sum of squared residuals
(residual = predicted – observed values) and TSS = sum of squared residuals with respect to
the mean of observed values (observed value – mean value).
5. R2 values can range from 0 to 1; closer to 1 implies best-fit.
6. Thumbrule: if R2 >0.9, it is a reasonably good fit and >0.95 is a very good fit. Values of R2
<0.5 show that there is no correlation between the two independent variables.
12
 LOGISTIC GROWTH CURVE
• = 𝑟 𝑁 1−

• Where r = 𝑟 1− Environmental resistance

• K = carrying capacity of the environment

• N = population at time t
• r0 = exponential rate of growth when N is very small (close to 0)

When N<<K, N/K can be ignored and =𝑟 𝑁

When N approaches K, r approaches 0

𝑁 = ( )

Where b = and N0 = population at t = 0

13

Ricklefs and Miller, 2000

 LOGISTIC GROWTH CURVE

700

Yeast cell cultures (Kormandy from Pearl, 1925; p196) 600

time, h N, cells dn/dt r 500

0 9.6 0 0

Number, cells
2 29 9.7 0.5528 400
4 71.1 21.05 0.4484
300
6 174.6 51.75 0.4492
8 350.7 88.05 0.3487 200
10 513.3 81.3 0.1905
12 594.4 40.55 0.0733 100
14 640.8 23.2 0.0376
16 655.9 7.55 0.0116 0
18 661.8 2.95 0.0045 0 2 4 6 8 10 12 14 16 18 20
time, hours

Growth rate is maximum at the beginning and decreases as N increases

14
 100
90
rate of change, dN/dt

80
70
60
r0 = 0.5716
50 slope = -r0/K -0.0008
40
30
K 714.5 cells
20
10
0
0 200 400 600 800
N, cells
0.6000

 All growth is exponential, but r is not constant as

exponential rate of change, r

0.5000
y = -0.0008x + 0.5716
in the exponential growth model 0.4000
R² = 0.9688

 r for each time interval is calculated based on 0.3000

the exponential equation
0.2000
 As N approaches K, r starts decreasing and
approaches 0 0.1000

 When dN/dt is maximum, N = K/2; this point is 0.0000

0 100 200 300 400 500 600 700
the inflection point N, cells
15
 LOGISTIC GROWTH CURVE
• This model is applicable to several species like yeast cells, and water fleas (based on published literature)
but not all species (Kormondy, 1996). It may or may not be the best-fit model in all cases or even for the
same species.
• It applies only to those populations where the growth rate is maximum in the beginning and decreases
monotonically over an extended period of time, i.e., the population is growing exponentially at all times but
with a decreasing growth rate. There is no time lag or lag phase in these populations. Note: The initial part of
the ‘S’ curve is not indicative of a lag phase, it is the nature of the logistic growth curve. This is the fastest growth
phase…..
• For populations where the growth rate varies over time and non-monotonically, e.g., India, it cannot be
applied.
• Even for the US population where this model was used in a famous ‘first application’ paper (Pearl and Reed,
1920), it was not an accurate prediction since many other factors are not accounted for.
16
 PROBLEM
Year Population
Population data for the US are shown here from a paper by
1790 3.93E+06 Pearl and Reed, 1920 where the authors showed that US
1800 5.31E+06 population dynamics followed the logistic growth curve.
1810 7.24E+06 i. Determine the carrying capacity for the US based on this
1820 9.64E+06 data. Ans: 313 million
1830 1.29E+07 ii. In what years, will the population reach 200 million and
1840 1.71E+07 300 million?
Ans:
1850 2.32E+07
It was estimated to reach 200 million between 1947 and
1860 3.14E+07
1948, and 300 million in 2030.
1870 3.86E+07
1880 5.02E+07
1890 6.29E+07
1900 7.60E+07
17
1910 9.20E+07
 Problem
• Draw a logistic growth curve for the following data for a bacterial culture:
• Bacterial cell conc at t = 0 was 3.4 x 105 cells/mL
• The carrying capacity of the culture flask is 1.5 x 109 cells/mL
• The exponential growth rate at t = 0 was 0.5953 1/hour
At what time will the cell population stabilize, i.e., grow no further?
Ans: At 25 hours, the bacterial conc is 1.49 x 109 cells/mL and there is almost no change
after that. Note: the conc will never actually be 1.5 x 109 cells/mL since the curve is
asymptotic.

18
 References

• Garg, SK (2001). Water Supply Engineering. Delhi: Khanna Publishers. [SKG]

• Goel S (2019) Water and Wastewater Engineering, Cambridge University Press, Delhi. Available from
Amazon India.