Você está na página 1de 13

Annual Natural Population Increase – An Empirical Study

1. Introduction
2. Empirical Natural Increase Model
3. Data information
4. Empirical results: Model estimations and Hypotheses testing
5. Summary and Conclusions
6. References

Empirische Arbeit von


Eigner Franz, a0301345
Sagerschnig Sophie, a9951023
Tirnitz Benjamin, a
bei Frau Prof. Kaufmann,
PR Empirische Wirtschaftsforschung,
SS05: Juni 2005
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

1. Introduction
In this paper we are interested in creating an empirical model, which is able to explain natural population
increase for 105 countries worldwide and in identifying the various responsible factors and discovering
their relative impact. Besides we want to check, whether it is meaningful to create separate models for rich
and poor countries to get better results and to find out, if the same variables are important in explaining
nat. population increase, or whether there are different significant variables.
We do not use a formulated model based on a theoretical framework but we create a regression model by
ourselves based on considerations after dealing with literature about this problem.
We generate our regression model with Eviews using the ordinary least squares (OLS) procedure which
produces the most efficient estimations.

Population Growth Facts:1 2


The world has experienced a rapid population growth since Industrial Revolution. The annual growth rate
increased in the last half of the 19th century from almost zero to 0.5 percent. It reached the 2.0 percent
mark in the 1960s, and declined to about 1.4 percent by 2000.
Nowadays almost all of the world population growth takes place in developing countries. In contrast to
more developed countries, developing countries which we find mainly in Asia, Africa and Latin America
had no distinctive mortality decline before World War II. The mortality revolution in developing countries
afterwards was not based on economic growth, but was a consequence of international foreign aid. In the
second stage, death rates decreased rapidly, whereas birth rates kept high or even increased on account of
better health conditions. In the 1960s, these countries had a mean population growth of 2.5 percent.
Around 1970, also birth rates began to fall and population growth reached the 1.9 percent mark in 2000.
In the future ‘world population is projected to increase to 7.8 billion by 2025 and to reach 8.9 billion by
2050’.

We can classify the countries after their population growth in 4 groups ( Table 1)

Table 1: Tendency in population growth3


Low fertility - Decreasing fertility – Increasing mortality (Aids) - Fast growing population
Population decreases Small population growth Decreasing population growth
Developing countries (except for among others: China, Among others: Zimbabwe, Botswana Israel, various developing
USA, GBR, FRA), some developing other countries in East- and South Africa, and other countries of countries like Palestine,
countries (Cuba) Asia, USA Sub-Saharan Africa Ethiopian

Basic information:
Population change is composed of 3 important variables: births, deaths and migration. If we subtract
deaths from births, we get the natural increase of a population, which generally accounts for the greatest
amount of population growth. Population growth itself arises from the natural increase rate added to the
net migration rate.

1
Pay attention: for some introducing parts of this paper we are dealing with Population growth, the model itself works with Natural growth.
2
http://www.prb.org/Content/NavigationMenu/PRB/Educators/Human_Population/Population_Growth/Population_Growth.htm
3
See Globale Trends 2002. Fakten Analysen Prognosen. Chapter: Weltbevölkerung und Verstädterung
Page 2 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

Population growth is based on a range of direct and indirect factors.4

 The biggest direct factor is the young population structure.


Approximately half of today’s world population is younger than 25 years. which is more than ever before.
This generation will be the ‘parents of the future’. As a consequence, world population will continue to
grow, even if birth rates would commence to decline. Experts claim, that this factor is responsible for half
of the population growth.
 The second biggest direct factor is the high number of unwanted births.
About a quarter of population growth can be attributed to this factor. Abortion and prevention can lower this
number. Furthermore it is one of the main factors responsible for high maternal mortality in developing
countries.
 Culture, religion, deficiency in education and health and discrimination of women are indirect
factors. Influences of such indirect factors are hardly to quantify.
 The impact of HIV/Aids on world population growth is still small. ‘Only’ 3 million people died in 2000
on account of Aids. But it will gradually rise because approximately 6 million people are newly infected
each year - with upward tendency. In the next 2 decades it could lead in many African countries to an
enormous decrease of population growth, an increase of child mortality of more than 100 % and a fall of the
life expectancy rate to less than 50 years.

For our model, we will pick out some of these factors.

2. Empirical Natural Increase Model

The following empirical model of natural population increase is postulated:

RNI = const + β1*GDP + β2*GROWTH + β3*EDU + β4*URB +


β5*POP15 + β6*POP65 +β7*TFR + β8*HIV +β9*LIFE + β10*GEM

Observations: 105, especially larger countries, worldwide


Dependent variable: RNI = Rate of Annual Natural Population Increase 2003 (Birth rate minus
Death rate, expressed in %)
Explanatory Variables5:
Economy: GDP = GDP per capita (PPP US$, 2002)
GROWTH = GDP per capita annual growth rate 1975-2002
Education: EDU = Special index from the UNDP for education, 2002
Population: URB = Percentage of people living in urban areas, 2002
POP15 = Population younger than 15 (in % of total population, 2002)
POP65 = Population older than 65 (in % of total population, 2002)
TFR = Total fertility rate (births per woman, 2002)
Health: HIV = HIV/AIDS among adults, Ages 15-49, 2003/2004 (%)

4
See Globale Trends 2002. Fakten Analysen Prognosen. Chapter: Weltbevölkerung und Verstädterung
5
selected definitions from PRB:
Total Fertility Rate (TFR)
The average number of children a woman would have assuming that current age-specific birth rates remain constant throughout her childbearing years (usually
considered to be ages 15 to 49)
Life Expectancy at Birth (LIFE)
The average number of years a newborn infant can expect to live under current mortality levels.
Percent Urban (URB)
Percentage of the total population living in areas termed “urban” by that country. The population living in towns of 2,000 or more or in national and provincial
capitals is classified “urban.”
Rate of Natural Increase (RNI)
The birth rate minus the death rate, implying the annual rate of population growth without regard for migration. Expressed as a percentage.
Page 3 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

LIFE = Life expectancy (2002/2003)


Women: GEM = Emancipation rate of women (2002, UNDP)

This empirical model focuses on explaining the natural increase rate for 105 countries worldwide by
cross-section analyses with 11 different explanatory variables. 2 variables, these are POP15 and TFR are
direct factors for RNI. All the others can be classified into the indirect factors.

 Economic developments are assumed to be captured by the average growth rate of real GDP per capita
between 1975 – 2002 and by the real GDP for the year 2002. Higher economic production leads to an
increase in the GDP and to higher economic development standards.
 The level of Education of the population is measured by a special education index of the UNDP, which
‘measures a country's relative achievement in both adult literacy and combined primary, secondary and
tertiary gross enrolment’6, which in fact, delivers very similar results as the alphabet rate.
 Population structure flows into this model by the percentage of population younger than 15, which can also
be seen as an indicator for the amount of natural increase in the last 15 years, by the percentage of
population older than 65, by the urban percentage and by the TFR.
 Health conditions can be explained by the life expectancy and the amount of people who suffer from HIV.
 Women: Furthermore, the UNDP created the Gender Empowerment Measure index to describe the degree
of emancipation of women especially in rich countries (even today it is not calculable for many poor
countries)

All these factors are assumed to be responsible for annual natural increase. But we have to be aware of the
fact, that we used 2002 data for these factors, whereas the dependent variable consists of data from period
2003 on account of failing to find data for 2002. So we have to assume that the explanatory variables for
2002 are not significant different from the ones for 2003. Nearly all of our variables contain values, which
do not change much within one year (e.g. GDP or literacy rate). As a result of missing exact data for the
HIV variable, we have partly estimates. The time period does not correspond exactly to the dependent
variable, and data are changing pretty fast, so we should be careful when we interpret the influence of this
variable on the natural increase.

1. Our first principle hypothesis is that the existence of a high economic standard leads to a smaller
natural increase.
Especially in rural areas of poor countries children do offer a variety of material benefits for their parents.
In rich countries women represent a larger part of the labour force and have more influence on political,
social and economical aspects in rich countries. Therefore they have less time and interest in childcare and
giving birth.

2. Our second principle hypothesis is that better social indicators, like health and education should
be indicators for smaller natural increase.
In particular for women it is enormously important, to have the opportunity to access education.
‘The number of children that a couple will have is determined by many factors, including health, religion,
culture, economic status and the ability to decide on the number of children – many of them are related to
the status of women.’7 This status of women can be measured by the GEM of the UNDP. Countries with
high social security will have smaller families, because people are less in the need of getting many children
for economic benefit. Life expectancy in many poor countries is currently decreasing on account of a high
mortality rate, caused by HIV. These hypotheses correspond to the first one of course, for the reason that
higher living standards generally correlate with higher economic standards.

6
http://www.undplao.org/HD%20measurement.htm
7
http://www.prb.org/Content/NavigationMenu/PRB/Educators/Human_Population/Women/The_Status_of_Women1.htm
Page 4 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

3. Our third principle hypothesis is concerning the population structure.


It is clear that countries with a high percentage of young people, tend to have a higher birth rate and as a
result should have a higher natural increase. So POP15 should be pos. correlated with RNI, as opposed to
POP65, which indicates the living standard and should have a negative effect on RNI. The TFR is also an
important criterion, which depends partially on social habits. Women who tend to get married early in life
(mainly in rural areas), will on average bear more children. TFR should have a pos. effect on RNI. An
interesting factor is urban population, which adds up to 76 percent in developed countries and only to 40
percent in poor ones, although urbanization in poor countries took on a new quality in the last decades. So
URB is assumed to have a neg. effect on RNI.

“Education, urbanization and labour force participation […] have a strong correlation with levels of
fertility.”8 But we have to be careful not to make the mistake, confusing correlation with causation.

3. Data Information

Some parts of the data9 used are obtained from UNDP Statistics, a database which offers many social and
economic statistics and from the ‘UNDP Human Development Report 2004’. Another reference is the
PRB 2004 World Population Data Sheet, and the ‘datafinder’ of the PRB, which are both available on the
website of the PRB. For some data, which were missing, MS Encarta Professional Edition 2005 was used.
The data set, which is indexed after the GDP, consists of 105 countries, which are selected by availability
of the data and by the size of the country’s total population.
At first view, the ranges of the different values of the explanatory variables are very large, which is
apparent from Table 2.

Table 2: General information about the data


Lowest Highest Mean
LIT in %10 Burkina Faso: 17,1 Several countries up to: 99 83,85
LIFE in years Japan: 81,6 Zambia: 32,4 67,27
POP15 in % Italy: 14,1 Uganda: 50,1 30
RNI in % Ukraine: -0,8 Niger: 3,5 1,24
HIV in % Botswana: 37,3 Several states: 0,1 2,38
TFR in % Niger: 8 Hong Kong: 1 2,92
GDP in $ per capita Luxembourg: 61190 Niger: 520 10953
Pay attention: different population sizes are not considered in the calculated mean

We become aware of the idea, that the varieties of these countries may perhaps make the estimation of a
single model more difficult. So it could be useful to compare rich countries with poor ones and to separate
them in later calculations. Nonetheless we have to keep in mind that this could be an inadequate
arrangement too. ( Table 1)
Table 3: Comparison: rich – poor
Poorest 30 Richest 30
LIT in % 63,9 96,9
LIFE in years 52,4 78.1
POP15 in % 41,4 18.1
RNI in % 2,21 0,35
HIV in % 5,21 0,35
TFR in % 4,7 1,7
GDP in $ per capita 1541 26951

8
See 4
9
UNDP Human Development Report 2004 / UNDP Statistics: GDP, GROWTH, LIT, EDU, URB, POP15, POP65, TFR, LIFE, GEM
PRB 2004 World Population Data Sheet / Datafinder of the PRB: RNI, HIV
10
Adult literacy rate (>15 years, 2002 UNDP)

Page 5 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

Austria itself is a typical rich country with a high life expectancy of 78.5 years, a Natural Increase of 0 %,
a TFR of 1.3 % and a GDP of 29220 USD. The population under the age of 15 amounts to 16.2 %.
The most important outliers are the United Arab Emirates, which have a high GDP of 22420 USD, but
only 77% of the inhabitants are able to read and the natural increase and the TFR are unusually high.
Botswana has a RNI of 0.1 %, population growth itself (with migration) amounts to -0.8 %, but has a
LIFE value of only 39.7 years and belongs to the poorest developing countries in the world.

As you can see in Table 4 RNI correlates strongly with POP15, POP65 and TFR, but very low with
GROWTH and moderately with all other variables.
The explanatory variables are also correlated among each other. Relatively high correlations can naturally
be seen between the variables that can be integrated in one umbrella term (cp. Empirical natural growth
model). For example LIFE and HIV show a high negative correlation (r=-0.75) as well as POP15, POP65
and TFR (r= - 0.81, r= 0.87, -0.74). However we have to keep in mind that a high correlation between two
independent variables doesn’t have to be desirable, because it might cause Multicollinearity.

Table 4: Complete Correlation Matrix - General Model


EDU GDP GEM GROWTH HIV LIFE POP15 POP65 RNI TFR URB
EDU 1.000000 0.622135 0.738999 -0.036310 -0.646040 0.868367 -0.785985 0.703723 -0.672268 -0.778563 0.593525
GDP 0.622135 1.000000 0.832173 -0.188010 -0.284417 0.644472 -0.667809 0.716462 -0.531158 -0.491397 0.596159
GEM 0.738999 0.832173 1.000000 -0.059106 -0.343677 0.696804 -0.602292 0.643468 -0.505868 -0.500015 0.582351
GROWTH -0.036310 -0.188010 -0.059106 1.000000 0.101357 -0.065051 -0.009974 0.038999 -0.057323 -0.021065 -0.047489
HIV -0.646040 -0.284417 -0.343677 0.101357 1.000000 -0.754571 0.398881 -0.323546 0.330239 0.508010 -0.266503
LIFE 0.868367 0.644472 0.696804 -0.065051 -0.754571 1.000000 -0.631165 0.565291 -0.477688 -0.664783 0.600966
POP15 -0.785985 -0.667809 -0.602292 -0.009974 0.398881 -0.631165 1.000000 -0.813390 0.872977 0.872291 -0.472072
POP65 0.703723 0.716462 0.643468 0.038999 -0.323546 0.565291 -0.813390 1.000000 -0.857838 -0.741279 0.457561
RNI -0.672268 -0.531158 -0.505868 -0.057323 0.330239 -0.477688 0.872977 -0.857838 1.000000 0.914094 -0.367022
TFR -0.778563 -0.491397 -0.500015 -0.021065 0.508010 -0.664783 0.872291 -0.741279 0.914094 1.000000 -0.415457
URB 0.593525 0.596159 0.582351 -0.047489 -0.266503 0.600966 -0.472072 0.457561 -0.367022 -0.415457 1.000000

Scatterplots11: EDU, GDP, GEM, HIV, LIFE, POP15, POP65, TFR, URB
4 4 4

3 3 3

2 2 2
RNI

RNI
RNI

1 1 1

0 0 0

-1 -1 -1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 20000 40000 60000 80000 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

EDU GDP GEM

4 4 4

3 3 3

2 2 2
RNI
RNI

RNI

1 1 1

0 0 0

-1 -1 -1
0 10 20 30 40 30 40 50 60 70 80 90 10 20 30 40 50 60

HIV LIFE POP15

11
without Cuba, Maldives, Nigeria
Page 6 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

4 4 2.0

1.6
3 3
1.2

2 2 0.8

RNI
RNI
RNI

0.4
1 1
0.0
0 0
-0.4

-1 -1 -0.8
0 4 8 12 16 20 0 1 2 3 4 5 6 7 8 9 50 60 70 80 90 100

POP65 TFR URB

4. Empirical Results: Model estimations and Hypotheses testing

4.1 Basic Model

Table 5 shows us the results of Equation 1 for the basic model which takes all the variables for all
countries.

Table 5: Equation 1 – Basic Model (all variables)


Dependent Variable: RNI
Method: Least Squares
Date: 07/31/05 Time: 16:24
Sample(adjusted): 2 99
Included observations: 48
Excluded observations: 50 after adjusting endpoints
Variable Coefficient Std. Error t-Statistic Prob.
EDU 0.519146 0.751422 0.690884 0.4940
GDP 3.50E-06 7.56E-06 0.463114 0.6460
GEM -0.614249 0.379371 -1.619127 0.1139
GROWTH -0.002468 0.013595 -0.181542 0.8569
HIV 0.053487 0.039657 1.348760 0.1856
LIFE 0.048083 0.012108 3.971039 0.0003
POP15 0.011743 0.010061 1.167177 0.2506
POP65 -0.072498 0.012399 -5.847327 0.0000
TFR 0.804613 0.088842 9.056647 0.0000
URB -0.001397 0.002112 -0.661241 0.5126
C -4.032997 0.817835 -4.931311 0.0000
R-squared 0.959983 Mean dependent var 0.795833
Adjusted R-squared 0.949167 S.D. dependent var 0.892214
S.E. of regression 0.201160 Akaike info criterion -0.171386
Sum squared resid 1.497213 Schwarz criterion 0.257431
Log likelihood 15.11325 F-statistic 88.76010
Durbin-Watson stat 2.007713 Prob(F-statistic) 0.000000

The very low number of included observations in Equation 1, which is even lower than the excluded ones,
is mainly caused by the variable GEM, which is not available for many countries. Besides this, we notice,
that primarily direct and population factors are significant in this model. Adjusted R-squared is extremely
high, but we have to take into account, that this unrestricted model leads to a lost of grades of freedom and
so to a reduction of the efficiency of the estimations. As we saw in Table 3, growth does not correlate with
any of the variables. This is represented in the equation. GROWTH seems not to be of significant
relevance for natural increase. Also URB and GDP are clearly not significant.

Page 7 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

4.2 General Model

After removing redundant variables (see Table 6), these are URB, EDU, GEM, GDP and GROWTH, we
use Eviews for looking at the residuals from the regression in tabular and graphical form (Actual, Fitted,
Residual Table), in order to find outliers, which appear not explainable by our model and which could
lead to Heteroskedasticity. After removing these countries (Cuba, Maldives, Nigeria) from our data set we
obtain Equation 2.

Table 6: Redundant Variables Test


Redundant Variables: URB EDU GDP GEM GROWTH
F-statistic 0.744980 Probability 0.594904
Log likelihood ratio 4.604244 Probability 0.466058

RNI = const + β1*POP15 + β 2*POP65+ β3*TFR + β4*LIFE + β5*HIV

Table 7: Equation 2 – General Model


Dependent Variable: RNI
Method: Least Squares
Date: 06/28/05 Time: 22:57
Sample: 1 59 62 80 82 105
Included observations: 97
Excluded observations: 5
Variable Coefficient Std. Error t-Statistic Prob.
POP15 0.037825 0.009490 3.985786 0.0001
POP65 -0.081097 0.012195 -6.650099 0.0000
TFR 0.406058 0.052331 7.759470 0.0000
LIFE 0.039396 0.005904 6.672763 0.0000
HIV -0.018547 0.008037 -2.307725 0.0233
C -3.030037 0.574349 -5.275604 0.0000
R-squared 0.938562 Mean dependent var 1.153608
Adjusted R-squared 0.935187 S.D. dependent var 1.053843
S.E. of regression 0.268292 Akaike info criterion 0.266381
Sum squared resid 6.550254 Schwarz criterion 0.425642
Log likelihood -6.919492 F-statistic 278.0351
Durbin-Watson stat 2.014705 Prob(F-statistic) 0.000000

Testing the model:


Before we carry out regression analyses we check our data for Homoskedasticity, no autocorrelation and
normal distribution because these are assumptions which provide efficient estimations when testing
variables using the OLS procedure.
We use the “White Heteroskedasticity Test” in order to control for Homoskedasticity, the “Jarque-Bera
Test” in order to check for normal distribution and the “Durbin Watson Test”, which measures the serial
correlation in the residuals in order to check for no autocorrelation.

Homoskedasticity
As can be seen in Table 8 our data unfortunately feature Heteroskedasticity, which is shown by the
probability value of 0.000002. This value is smaller than 0.05, which marks the critical value for
Homoskedasticity. Heteroskedasticity means the variance of the residuals in the linear regression model is
different across the sample. Referring to our sample it could mean that either the poor or the rich countries
differ more strongly in the Rate of Annual Natural Population Increase.
As we use the OLS procedure to estimate the parameters, the estimators and forecasts are still unbiased
and consistent but on the other hand inefficient and no longer BLUE. Tests of hypotheses and
interpretation of the results must be handled with care.

Page 8 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

Table 8
White Heteroskedasticity Test: with cross terms
F-statistic 7.122745 Probability 0.000000
Obs*R-squared 63.25390 Probability 0.000002

Autocorrelation and normal distribution


Other assumptions of a regression analyses - no autocorrelation and normal distribution are given as you
can see in Table 7 and Table 9. The Test for normal distribution shows a probability value of 0.09, which
is higher than the critical value of 0.05 and means that normal distribution is given. The test for
autocorrelation shows a value of 2.015, which is very close to value 2 and means that no autocorrelation
exists.
Table 9: Test for normality of the error terms with Jarque-Bera
14
Series: Residuals
12 Sample 1 105
Observations 97
10
Mean -2.31E-15
8 Median -0.009486
Maximum 0.715757
Minimum -0.864046
6
Std. Dev. 0.261212
Skewness -0.201898
4
Kurtosis 3.999708

2 Jarque-Bera 4.698306
Probability 0.095450
0
-0.75 -0.50 -0.25 0.00 0.25 0.50 0.75

Interpretation of the results:


Adjusted R^2 for Equation 2 amounts to 0.93, which means that 93% of the variance of the independent
variable RNI can be explained by this model. Compared to the explanatory power of Equation 1, values
are a slightly worse. Of course we have to realize that these values can hardly be compared with each
other, because in Equation 2 more explanatory variables and much more observations are used than in
Equation 1. Furthermore R^2 and even adjusted R^2 always increase when adding explanatory variables.
Though for such cases12 it could be better to use the information criterions Akaike and Schwarz, which
furnish slightly better values for the new model. What is more the F-Statistic adds up to 278 and indicates
that our model as a whole is very significant. The remaining, significant variables are capturing population
characteristics, which could be easily explained by the high correlations with RNI, and also Health
characteristics. From the five significant variables, TFR, LIFE and POP65 have a strong influence on
RNI, whereas POP15 and HIV seem to be less important. POP15, TFR and LIFE correlate positively and
POP65 and HIV negatively with RNI.
More exactly: Provided that values of all the other variables keep the same, the following holds: When
TFR increases by one percent, RNI increases by 0.4 percent, when LIFE increases by one year, RNI
increases by 0.04 percent and so on. The results concerning TFR, POP15, POP65 and HIV are in harmony
with our correlation matrix and verify our assumptions; only the trend of the correlation of the variable
LIFE with RNI is surprising. We expected that RNI is less in countries where life expectancy is higher
and the correlation Matrix (Table 3) confirms this assumption but in this model the trend of the correlation
of the variable LIFE with RNI is in the opposite direction. This may be caused by East-European
countries, which ‘disturb’ our correlation.

Important insignificant variables: EDU, GDP, GROWTH, URB, GEM


Do we have a closer look at the variables, which are not significant. As we have seen before, GROWTH

12
With more explanatory variables, but not with more observations. In this case also the values of Akaike and Schwarz are not mmeaningful fir comparison.
Page 9 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

corresponds very little with RNI, and t-statistics verify its unimportance for our model. Surprisingly, EDU
and GDP are also not significant, although for both variables in the scatter plots as well as in the corr.
matrix (-0.67, -0.53) a negative correlation is discernible. It seems that either these correlations are not
plain enough (outlier) or they are not linear and so not valid for our model. The corr. matrix shows for
URB and RNI correlation with a value of only 0.36, and the scatter plot proves that its influence seems to
be of no importance. GEM is not available for many countries and probably for the same reasons as for
EDU and GDP it’s not significant in our model.

Comparison Equation 1 with Equation 2:


The Equation 2 delivers better results for our model, because all the variables are clearly significant and
furthermore it has a higher value of degrees of freedom, which indicates a higher flexibility for the model.
‘Artificial’ high explanatory power as in Equation 1, on account of too many expl. var. in the sample
could lead to a poor adjustment outside our sample (for forecasts). The variance of model 2 will be higher
than that without the irrelevant variables and hence the coefficients will be inefficient.

4.3 Model for rich countries

Now we desire to check, if separate models for the 30 richest / 35 poorest countries of our data give us a
better insight into some correlations and causalities. It is intelligible, that (alleged) more homogenous
countries can be compared with each other with less external effects, which distortion our results.

After sorting out the non significant explanatory variables for the 30 richest countries, which are HIV,
GDP, GROWTH, LIT, EDU and POP15, the model in Table 10 remains.

RNI = const + β1*LIFE + β 2*GEM + β3*POP65+ β4*URB + β5*TFR

Table 10: Equation 3 - Rich Countries (top 30)13


Dependent Variable: RNI
Method: Least Squares
Date: 06/16/05 Time: 00:19
Sample(adjusted): 2 30
Included observations: 25
Excluded observations: 4 after adjusting endpoints
Variable Coefficient Std. Error t-Statistic Prob.
LIFE 0.088754 0.010191 8.709370 0.0000
GEM -0.661161 0.166811 -3.963526 0.0008
POP65 -0.078053 0.007989 -9.769816 0.0000
URB -0.003980 0.001536 -2.590957 0.0179
TFR 0.729414 0.061881 11.78737 0.0000
C -5.915063 0.717432 -8.244773 0.0000
R-squared 0.971274 Mean dependent var 0.268000
Adjusted R-squared 0.963714 S.D. dependent var 0.423743
S.E. of regression 0.080718 Akaike info criterion -1.990150
Sum squared resid 0.123792 Schwarz criterion -1.697620
Log likelihood 30.87688 F-statistic 128.4840
Durbin-Watson stat 1.808970 Prob(F-statistic) 0.000000

Testing the model:


Both Homoskedasticity (p=0.46) and normal distribution (p=0.19, Table 11) are given and the Durbin-
Watson Test shows that there is no autocorrelation. Therefore there is no problem in using the linear
regression analyses.

13
Listed according to the size of the GDP
Page 10 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

Interpretation of the results:


The explanatory power of Equation 3 is a bit higher than the explanatory power of the general model.
TFR, LIFE, POP65 have a big influence on RNI, whereas the influence of GEM and URB is much
smaller. Again, with the exception of the trend of the correlation of the variable LIFE with RNI, all
variables are correlated with RNI as assumed.
In contrast to the general model, the variables GEM and URB get significant in this model, whereas
POP15 and HIV are no longer significant. That POP15 and HIV are no longer significant can possibly be
explained by the idea, that the “rich countries”, which are analysed in this model, do not differ much
among each other in these variables. Or this may be because they are closely associated with POP65 abbr.
LIFE and are captured by these factors. GEM and URB are variables, in which mainly “the rich countries”
differ, so they are probably more meaningful here.

4.4 Model for poor countries

Now we analyse the 35 “poor countries” in our sample separately. After removing the non significant
explanatory variables, which are TFR, URB, POP65, EDU, GROWTH and GDP, the model in Table 12
remains.

RNI = const + β1*HIV + β2*LIFE+ β3*POP15

Table 12: Equation 4 - Poor Countries (last 35)


Dependent Variable: RNI
Method: Least Squares
Date: 06/16/05 Time: 00:22
Sample: 71 105
Included observations: 34
Excluded observations: 1
Variable Coefficient Std. Error t-Statistic Prob.
HIV -0.042547 0.010648 -3.995726 0.0004
LIFE 0.019116 0.010338 1.849079 0.0743
POP15 0.118213 0.015861 7.453131 0.0000
C -3.483876 1.173389 -2.969072 0.0058
R-squared 0.810285 Mean dependent var 2.088235
Adjusted R-squared 0.791314 S.D. dependent var 0.742136
S.E. of regression 0.339024 Akaike info criterion 0.784641
Sum squared resid 3.448123 Schwarz criterion 0.964213
Log likelihood -9.338892 F-statistic 42.71069
Durbin-Watson stat 1.919063 Prob(F-statistic) 0.000000

Testing the model:


We check for normal distribution (p=0.19) and Homoskedasticity (p=0.96), which are both given. And
after ruling out autocorrelation (Durbin Watson =1.91), we carry out linear regression analyses.

Interpretation of the results:


In contrast to the general model and the model for rich countries, there is more unexplained variance left.
Akaike and Schwarz are also inferior to the other models. This may indicate that other important
variables, which are not included in the Empirical natural growth model (ct. S. 2) are important in order to
predict RNI for these countries.
In this model the variables HIV and POP15 are significant at an alpha level of 5% and the variable LIFE is
significant at an alpha level of 10 %. The coefficients of LIFE, POP15 and HIV are in agreement with our
assumptions.
In contrast to the general model, POP65 and TFR are not significant anymore, which is surprising,
Page 11 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

because in the scatter plot14 a tight correlation with RNI can be noticed. Also in this case it might be that
POP65 and also TFR are captured by POP15.

Comparison Equation 3 and 4


When we take a closer look at the 2 ‘smaller’ models for poor and rich countries, we see that different
variables show a significant contribution in explaining the population growth. We also recognize that
there is a large gap between the explanatory power of the 2 ‘smaller’ models. (rich: 0.96, poor: 0.79). This
can be due to the fact15, that differences between the group of developing countries are bigger than
between the whole developed and developing regions. Even within a country, like in India, fertility rate
and population growth rates are more varying within the country than between the totality of developed
and developing countries. Developed countries have similar governments (democracy, social market
economy), similar culture and economic data are quite the same within one country, so population growth
is much easier to rate here.’ This would also explain the small number of variables, which are significant
for the poor countries in our model. As a consequence it is thoroughly meaningful, calculating these
models, in order to confirm these statements and to get more insight into the differences between rich and
poor countries, concerning natural increase.

5. Summary and Conclusions

In our models, explaining Natural Population Increase, we could not find an influence of the economic
standard on RNI, which is surprising because in literature often a positive correlation between economic
standard or wealth and RNI is assumed.16 So our first hypothesis that the existence of a high economic
standard leads to a smaller natural increase cannot be fully affirmed by our outcomes. Nevertheless the
correlation matrix and its scatter plot indicate a positive correlation, which could imply that a high GDP
leads to a smaller RNI, but for other reasons GDP is not important in our model.
We also couldn’t find an influence of Education on RNI in any of the models, and of GEM in the general
model. So our assumptions that high education leads to a positive effect in our model cannot be affirmed,
either. However we see in the scatter plots, that a high education and a high GEM often correspond with a
smaller natural increase. The same as for GDP applies to GEM and EDU.
Regarding the “health variables” we could find an influence of HIV and LIFE. As expected, HIV
correlates negatively with RNI in the model for “poor countries” and in the general model. The influence
of life expectancy on RNI is in the opposite direction as expected. As life expectancy is often low in
countries with high population growth, and high in countries with little population growth, we thought that
high life expectancy would indicate little population growth.17 This can be due largely to some countries
in East-Europe, which distort our correlation. Our hypothesis regarding health can be affirmed partially,
namely the assumption that HIV influences the population increase in our model negatively.
Our hypothesis concerning population structure could be confirmed for all variables. A high amount of
young people and a high fertility rate lead to a high population increase, whereas a high amount of old
people and urbanisation is correlated with a population decrease.

Our study shows some limitations, which may diminish its explanatory power and could lead to poor
results for future data. It should be taken into consideration, that …

14
only with the data from the 35 poorest countries
15
see Globale Trends 2002. Fakten Analysen Prognosen. Chapter: Weltbevölkerung und Verstädterung
16
http://www.bmz.de/de/themen/armut/hintergrund/index.html#Die%20Ursachen%20von%20Armut
17
http://www.weltpolitik.net/Sachgebiete/Weltwirtschaft%20und%Globalisierung/Grundlagen/Grundlagen/Akteure%20der%20Weltwirtschaft.html
Page 12 of 13
Annual Natural Population Increase – An Empirical Study Eigner, Sagerschnig, Tirnitz SS05

 In specifying the model, we implicitly assumed that X causes Y. “Although R^2 measures the goodness of
fit, it cannot be used to identify causality. In other words, the fact that X and Y are highly correlated does
not indicate whether changes in X cause changes in Y or vice versa. “The situation can also go both ways, a
situation known as feedback.” For instance: Inadequacies in social services lead to rapid population
growth. However in most poor countries we see that overpopulation can reduce food production gains
resulting from modernizing.
 we haven’t included all countries of the world in our study
 we possibly should have used some other explanatory variables, for example the use of contraception18, we
could have used an econometrical model as basis for our estimations (macro/microeconomic19).
 it perhaps would have been better to use mean values of the explanatory variables (for instance 1975-2003)
instead of numbers from one year to obtain a more robust model.
 we should have weighted variables after population size to avoid distortions caused by smaller countries.
 it would have been meaningful to line up the models to 4 groups, as explained in Table 1 for getting
superior results. Because the biggest problem was certainly the heterogeneity of and within the countries.
By the proposed enhanced differentiation robustness and explanatory power of the model should raise.

6. References
Data and Statistics:
or http://cfapp2.undp.org/hdr/statistics/data/rc_select.cfm Zugriff: 04.06.2005
http://hdr.undp.org/statistics/
UNDP - Human Development Report 2004. Statistics.

http://www.prb.org/ Zugriff: 04.06.2005


Population Reference Bureau.

http://www.prb.org/datafind/datafinder5.htm Zugriff: 04.06.2005


Datafinder of the PRB.

General Information:
Stiftung Entwicklung und Frieden: Globale Trends 2002. Fakten Analysen Prognosen. Hg. von Hauchler, Messner,
Nuscheler. - Frankfurt a. M.: Fischer Taschenbuch Verlag 2001.

http://www.prb.org/Content/NavigationMenu/PRB/Educators/Human_Population/Human_Population__Fundamentals_of_Growth_and_
Change.htm Erstellungsdatum: November 2000, Zugriff: 04.06.2005

http://users.rcn.com/jkimball.ma.ultranet/BiologyPages/ Erstellungsdatum: 02.07.2005, Zugriff: 21.07.2005


Kimball's Biology Pages: Checks on Population Growth, Human Population Growth.
P > Populations > how regulated / human

http://www.ecopop.ch/A2BULLETINS/bulletin41-2004.htm Zugriff: 28.06.2005


Eco Pop – Union for environment and population. Bulletin ECOPOP. Nr. 41 February 2004

Macro- / Microeconomic Theories:


http://www.berlin-institut.org/pages/buehne/buehne_bevwiss_prskawetz_oekonomie.html
Erstellungsdatum: 4.06.2003, Zugriff: 21.07.2005
Alexia Prskawetz: Bevölkerungsökonomie. Max Planck Institute for Demographic Research. Rostock

http://www.payer.de/entwicklung/entw31.htm Erstellungsdatum: 16.02.2005, Zugriff: 21.07.2005


Bettina Eckl und David Prüm : Einführung in Entwicklungsländerstudien. published by Margarete Payer (Juni 00).
especially: 1.3 Ökonomisch-demographische Theorien

18 neg. effect on RNI shown in http://www.ecopop.ch/A2BULLETINS/bulletin41-2004.htm


19
http://www.berlin-institut.org/pages/buehne/buehne_bevwiss_prskawetz_oekonomie.html and http://www.payer.de/entwicklung/entw31.htm

Page 13 of 13

Você também pode gostar