Sie sind auf Seite 1von 17

ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau

1444107

Introduction
Suicide – the act of intentionally causing one’s own death – is a very complex issue in which
characteristics and reasons for such behavior is widespread and varied between demographic and
geographical populations. Suicide is usually associated with mental illness and stress related
attributes; “[several] individual risk factors have been documented well, including physical
illnesses and mental disorders, economic difficulties, social isolation, disaster, and physical
abuse” (Lin et al, 2016). However, until recently, it has been rarely been correlated with our
physical environment.

It is interesting to note the environment, and subsequently, environmental quality, are often
considered to be “luxury goods” in environmental economics. Poorer groups of people must feed
themselves and therefore cannot afford to care about the environment; in such extent, poorer
countries, due to their lack of technology, will have huge opportunity costs for pollution
abatement. It is only when people are wealthier, when they are able to afford leisure while also
affording day to day living, that they will start caring for the quality of air that they are
breathing.1 As such, developed countries with higher GDP should have better pollution control
than developing countries. It has been clear for a long time that economic activity has a large
effect on “consumer consumption” of environmental quality. But what are the effects of such a
“luxury good” on socio-economic issues and the social wellbeing of the people? “Because
suicide has such economical and psychological burdens on the community, preventing suicide is
a global imperative”; if pollution does indeed affect suicide rates and has such an adverse
externality, then environmental quality can no longer be considered as just a “luxury good.

The quality of one’s surroundings have been documented to be detrimental towards physical
health and mental health; air pollution, in particular to issues with the brain, is linked to physical
illness such as headaches and depression, where long term exposure can lead to neurological
damage. Several health literatures noted by Lin et al suggest that air particulates, when inhaled,
could increase the release of stress related hormones. Short term memory and behavioral changes
were noted to be an effect of exposure to these contaminants. Multiple experimental articles 2 also

1 Note that environmental quality is considered to be a good for middle-class only. As people get extravagantly rich,
they can simply afford to move elsewhere if the quality of their surroundings is not to their liking.
2 Done on animals

1
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

present the increase in deterioration of neuronal cells and synaptic deficiencies, resulting in
“impairment of neuronal behavior”; mitochondrial abnormalities, “which are involved in the
pathogenesis of neurological disorders” have also been observed to be an effect of these
pollutants.

Aggravation caused by pollutants, already noted to inhibit ability to make rational decisions as
well as cause physical and mental distress, may subsequently induce suicide. I propose that
environmental quality has an inverse effect on suicide rate; as quality decrease, suicide rate
increases.

Data Description
This data comes from the World Bank, an organization made up of nearly 200 countries set on
ending extreme poverty. The data itself is cross panel data, taken from various countries all over
the world over the years 2000, 2005, and 20153. The dependent variable, suicide rate, is
measured in terms of suicide mortality rate per 100000 people. For the purposes of this
regression, since environmental quality is an open-access good and thus shared by everyone, the
suicide rate does not need to be standardized to a certain group like age or gender. It should be
noted that because some cultures consider suicide and mental illness a stigma, this statistic may
not be completely accurate.

The regressors also come from the World Bank. Environmental quality is proxied to one air
quality categories – CO2 metric ton per capita. While environmental quality is a big umbrella
term which encompasses water pollution as well as soil pollution, air pollution is a good enough
proxy because it is such an immediate and constant factor in someone’s life; soil pollution (toxic
chemicals and heavy metal pollutants) take a long time to bio-accumulate and affect the body;
water pollution can be somewhat mitigated by a water filter. Pollution may take some time to
start affecting the body so interaction variables are created for the log of CO2 emissions.

The other regressors are GDP per capita, tertiary education enrollment, and unemployment. GDP
and unemployment, or financial stress, are assumed to be a major reason as to why people take

3 For some reason, suicide rate is rarely reported annually.

2
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

their own lives. Education should have a negative relationship with suicide rates – though, from
my previous paper, this is not the case in South Korea. As such, it would be safer to include this
variable than to omit it. We also assume that mental illness is a constant factor throughout the
countries – the WHO reports that about 25% of people in the world suffers from some kind of
mental illness – and can be included in our constant factor4. Because the data is macro-data, age
will not be included in the regression – though it would be interesting to see if the environment
affects differently aged people differently. Any shocks or unforeseen events that can possibly
affect the data, such as rape victims or death of a loved one, will be included in our error term.

The final sample size is 484. A lin-log relationship regression will be used to test the hypothesis.

Findings
Our regression (see Table 1) yielded some interesting results; the R^2 was ~0.29, meaning that
about 30% of the variance was explained by the model.
- The constant term is 13.42, meaning that if all our other terms had zero effect, the suicide
rate would be 13.42 per 100000 persons. This suggests that there are indeed other factors
not considered by the model that affects suicide rate.
- The coefficient of “lnco2”, or the log of CO2 metric tons emitted per capita, came out to
be ~1.39. This means that a 1% increase in CO2 emissions per capita results in an
average of 1.39 unit increase in the global suicide rate per 100000 persons. The p-value is
less than 0.05; this is statistically significant, and as a result, we can reject the null
hypothesis that this control has zero effect on the dependent variable. Because there is an
interaction term between lnco2 and year, this coefficient means that a 1% in CO2
emissions generates a 1.39 unit increase in suicide rate only in the current year.
- The two GDP per capita terms both have a p-value below 0.05, meaning that they are
statistically significant. However, both are pretty close to zero, suggesting that their
effects on suicide rate are pretty insignificant. For a unit increase in GDP per capita, the
suicide rates decrease by 0.0004, but at the same time, the model also increases by
GDP(_GDP2), meaning that, with respect to a one unit increase in GDP per capita,
suicide rates will increase with the following model: suicide rate = 3.06x10^-9 (GDP) –
4 Though one may argue that mental illness (like depression, etc) and the will to act on it can be a middle class
good, like environmental quality, because poorer people are more focused on surviving. However, economic factors
do tie into the suicide rate, so this area becomes muddy.

3
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

0.0004. Even though GDP^2 is statistically significant, suggesting that there is indeed a
curvature in the data, the value is so small that it may be sampling error and that it
actually is just linear.
- Tertiary enrollment rate has a coefficient of ~0.21. Because of enrollment rate is
measured in percentages, a 1% increase in tertiary enrollment rate results in a 0.21 unit
increase in suicide rates per 100000 persons. The p-value is less than 0.05; it is
statistically significant.
- Similar to the GDP terms, both p-values present unemployment rate as statistically
significant terms. Unemployment has a coefficient of ~-0.62, and unemployment^2 has a
coefficient of ~0.02, meaning that a one percent increase in unemployment rate will result
in a 0.02 (unemployment) – 0.62 unit change in suicide rate per 100000 persons.
- The year control, which is a dummy variable for year consisting of 1 for 2000, 2 for
2005, and 3 for 2010, is also statistically significant and has a coefficient of -1.56,
meaning that there is a general trend over each 5 year gap that suicides around the world
are decreasing at ~1.56 persons per 100000 persons.
- The interaction term between year and lnco2, “yearlnco2”, is statistically significant. The
coefficient is -0.58, meaning that a 1% increase in CO2 generated today will result in a
0.57 unit decrease in suicide rate every 5 years. This suggests that CO2 generated today
will have a decreasing effect over time. With this interaction term, the total effect of CO2
emissions on suicide rates become the following: suiciderate = b1 + b8(year) = 1.39 –
0.58(year).
- The interaction term between year and GDP is statistically insignificant, and has a pretty
small coefficient value, suggesting that year and GDP do not affect each other in terms of
effect on suicide rate.

Because it is very unlikely that the effect of CO2 emissions on suicide rates decreases over time,
I repeated the model with two lag variables (see Table 2). The results are the following:
- For the two lag variables, “lnco2lag1” and “lnco2lag2”: both p-values are less than 0.05,
meaning that they are statistically insignificant. We do not have enough evidence to say
that these regressors have any effect on suicide rate. Lag1’s coefficient is ~0.17, meaning
that the number of suicides is expected to increase by 0.17 per 100000 persons 5 years
after the 1% increase of CO2 emissions this year. Lag2 suggests that there is an expected
decrease of 0.10 per 100000 persons of suicides ten years after the 1% increase in CO2

4
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

emissions. The total CO2 effect can be given by adding up the three CO2 variables (lnco2
+ lnco2lag1 + lnco2lag2); we can say from these three variables that a 1% increase in
CO2 emissions will result in a total of ~2.2 suicides per 100000 persons. It is noted that
each lag has a lesser and lesser effect, suggesting that CO2 has a diminishing effect over
time.

When plotting the observed values against the fitted values (see Figure 1), the scatterplot
presents a general upward sloping trend, but has a lot of variability. This suggests that the model
can only predict suicide rate mediocrely.

To test for heteroscedasticity in the model, I created a correlation table between the dependent
variables and the residuals (see Table 5). There was some correlation so heteroscedasticity might
be present; I then regressed the squared residuals on the regressors (see Table 4). There were
some variables that were statistically significant, meaning that heteroscedasticity was indeed
present. To make sure that these were caused by the interaction terms or the polynomial terms, I
regressed the squared residuals on just the variables in the original data set. GDP per capita,
tertiary enrollment rate, and the year dummy all exhibited statistically significant correlated with
the residuals. Performing a Breusch Pagan test and a White general heteroscedasticity test, both
resulting p-values were 0.000, confirming that there was heteroscedasticity present in the model.
As such, I re-regressed the model using White robust errors (see Table 6). The results were the
same, albeit with mostly smaller confidence intervals.

The results from performing the STATA command “vif” to test for multicollinearity were
alarming, as multiple regressors had a result of over 10 (see Table 7). However, this could be
explained by the inclusion of polynomial and interaction terms; by removing them and reusing
the command (see Table 8), all variables did not exhibit multicollinearity.

In order to test for autocorrelation, I predicted the residual errors and created a lag term in which
error lagged over one period of time. I then regressed the errors on the error lag and the rest of
the regressors (see Table 9). The resulting error had a p-value of 0.19, meaning that it is
statistically insignificant and therefore we have no violated out autocorrelation assumption.

5
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

Literature Review
In this study, it was found that a 1% increase in CO2 emissions causes a 1.39 increase in suicide
rate per 100000 persons. It was also found that a 1% increase of emissions today will have
diminishing positive effects on suicide lasting over the next five years.

While scientific and economic literature have extensively explore the physical health and
economical costs of environmental quality or the lack thereof, papers on the connection between
the externality of pollution and moral cost has been scarce; association between distinct decision
making and air quality has only recently begun increasing in popularity.

Pertaining to the moral costs in suicide rates, a study was conducted on the effect of air quality
on suicide mortality in Guangzhou, China. Lin et al. found that “significant air pollution [was
correlated with] violent suicide mortality in the cool season but not on non-violent suicide
mortality or in warm season” (Lin, 2016). There was an average of a 14% increase in mortality
rate over an increase in their three pollutants – PM10, NO2 and SO2 – concluding that there was
a significant increase in suicide risk with an increase in ambient pollutant levels. Another study
was done on the same topic in Korea by Kim et al. While they did not find enough evidence to
support a correlation between gaseous pollutants and suicide, they were consistent in that both
found that an increasing concentration of air particulates had detrimental significance on suicide
rate. A third study was done on short term air pollution on suicide risk in Utah by Balkan et al. It
produced similar results to the study done by Lin et al. All three studies had found that suicide
rates and pollution levels followed a similar pattern of seasonal rates; the difference in results
could be explained by how “seasonality of air pollution effects may be different in subtropical
regions” (Lin et al, 2016) and the fact that certain geographical regions could have differing
levels and variability in ambient levels of contaminants.

This study provides a general overview on the association between environmental quality and
global suicide rates, while most other studies focus on a more specific area. Exploring this
relationship will provide incentive for additional support of increasing ecological conditions. The
fact that our immediate physical surrounding have such a significant effect on mental health

6
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

presents the need for increased environmental controls; it can no longer be considered as just a
“leisure good” as some environmental economics texts suggest. This relationship also presents
mental health as a physical condition, which will hopefully reduce some of the stigma
surrounding mental health and suicide.

While this paper provides the advantage of having a global study, it does have many major
drawbacks. The first and foremost of these flaws is that, by having a global study with such a
widespread panel of time, the data is very general. As noted by all three mentioned papers,
seasonal changes incur a large effect on the effects of ambience quality on suicide rates.
However, because this study uses time data measured in periods of five years, including these
seasonal controls are completely ineffective as they fluctuate in much smaller intervals. Having
data on a macro-scale like this data set prevents the exploration of certain controls that can
definitely be included in micro-data. By studying the “suicide rate per 100000 persons” at a
certain year, we cannot include certain variables such as gender, age, race, or if the suicide
happened in an urban area or not. For example, the regression analysis presented that tertiary
enrollment rates had a positive effect on suicide rates while the predicted effect should have been
negative. In relation to my previous paper 5, because East Asians make up such a large portion of
the world’s population, it suggests that there may be some locational or racial bias in this model.
Balkan et al. noted that different age groups were affected by air particulates differently, as
“suicide risk increases linearly with age”, and all three studies mentioned gender as a major
associate. As such, there is a large probability of the presence of omitted variable bias caused by
the fundamental nature of macro-data.

Another factor that limits this study is its pollution measurement as well as its limited inclusion
of different pollution forms. Whereas other studies use ambient levels of pollution to study the
effect, this study uses emission levels. This may not necessarily be an accurate assessment of
“surrounding environmental quality” as air pollutant emissions are transboundary – meaning that
they are not limited to one area – and are dispersed easily by weather conditions. Furthermore,
by only studying the effect of CO2 emissions, only a small perspective of pollution and
environmental quality is captured. Not only are there are many other forms of air toxins that are

5 Where participation in education and academia increase the suicide rates of Korean youth

7
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

known to have a major effect on the human body – such as PM2.5 (air particulates with a
diameter of 2.5 micrometers or less), nitrogen oxides, sulfur oxides, etc. – there are water and
soil toxins as well. However, it can be argued that each different form of pollutant must be
regressed separately as inclusion of multiple contaminant controls may introduce
multicollinearity; as such, even though the conclusion that can be drawn from this regression
may be limited, this defect does not introduce bias to the model itself.

An important remark to make is that this suicide variable only includes data on suicides itself,
and not suicide attempts. While these two are similar in nature, they may differ in risk factors,
and as such, the effect of air quality on suicide attempts may differ from actual suicide rates.

Conclusion
In summary, CO2 emissions per capita have a somewhat large positive effect on suicide rates.
Significantly, this effect has a positive effect that lasts over five years on average. Even though
macro data prevents the study of environmental detriments on different groups such as age and
gender, a correlation has been established between pollution and mental health and, more
importantly, decision making6. From this model alone, we can conclude that pollution does have
significant moral costs that must be addressed in addition to its economical and health costs
already associated with it.

Tables and Figures

Variable Obs Mean Std. Dev. Min Max


Country Name 0
Year 693 2005 4.085432 2000 2010

6 Which paves the road for other studies on effects of environmental quality on other decision making issues, such
as crime.

8
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

Suicide Rate per 100000 Persons 693 10.76835 7.025163 0.2 50.1
CO2 metric ton emissions per
capita 680 4.540325 6.198808 0.017293 58.91873
NO2 equivalent of CO2 metric ton
emissions per capita 682 133659.2 370297.5 0 3084900
GDP per capita 678 9292.985 14456.17 123.8762 104965.3
Tertiary enrollment rate 497 31.18671 24.9228 0.268 102.7309
Unemployment rate 666 8.739605 5.54467 0.4 36
Table 1. Summary of statistics; the controls presented are year, suicide rate per 100000persons,
CO2 metric ton emissions per capita, NO2 equivalent of CO2 metric ton emissions per capita,
GDP per capita, Tertiary enrollment rates, and unemployment rates. The statistics shown are
number of observations, mean, standard deviation, and the min and max.

Number of obs
Source SS df MS = 488
F(9, 478) = 21.64
Model 7522.598 9 835.8442 Prob > F = 0
R-squared
Residual 18460.96 478 38.62127 = 0.2895
Adj R-squared
= 0.2761
Root MSE
Total 25983.56 487 53.35434 = 6.2146
Suicide Rate per 100000 Persons Coef. Std. Err. t P>t [95% Conf.Interval]
Ln of CO2 emissions, metric ton
per capita 1.39308 0.61909 2.250 0.025 0.17660 2.60955
GDP per capita -0.00044 0.00009 -5.050 0.000 -0.00061 -0.00027
GDP per capita squared 0.00000 0.00000 3.010 0.000 0.00000 0.00000
Tertiary enrollment rate 0.21265 0.01901 11.190 0.000 0.17529 0.25000
unemployment rate -0.61527 0.17301 -3.560 0.000 -0.95524 -0.27531
unemployment rate squared 0.01844 0.00668 2.760 0.006 0.00532 0.03156
year dummy variable -1.56250 0.43777 -3.570 0.000 -2.42268 -0.70231
year;lnCO2 emissions interaction
variable -0.57623 0.28786 -2.000 0.046 -1.14186 -0.01061
year;GDP per capita interaction
variable 0.00005 0.00004 1.310 0.192 -0.00003 0.00012
Constant 13.41650 1.32853 10.100 0.000 10.80602 16.02697
Table 2. Regression table; the regression of suicide rates per 100000 persons on logged values of
CO2 metric ton emissions per capita, GDP per capita, GDP per capita squared, tertiary
enrollment rate, unemployment rate, unemployment rate squared, a year dummy variable where
1 = 2000, 2 = 2005, 3 = 2010, an interaction variable between year and logged CO2 emissions,
an interaction variable between year and GDP per capita, and a constant value.

9
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

Figure 1. Plotted values of suicide rate on its predicted fitted values.

Number of
Source SS df MS obs 481
F(11, 469) 17.44
7527.68 684.335
Model 8 11 3 Prob > F 0
18405.2 39.2435
Residual 2 469 4 R-squared 0.2903
Adj R-squared 0.2736
25932.9 54.0268
Total 1 480 9 Root MSE 6.2645
Suicide Rate per 100000
Persons Coef. Std. Err. t P>t [95% Conf. Interval]
Ln of CO2 emissions, metric ton 0.64821 2.71723
per capita 1.44346 9 2.230 0.026 0.1696872 2
0.17092 0.34875 0.85624
1st lagged variable of ln CO2 1 7 0.490 0.624 -0.514399 2
0.25279 0.39587
2nd lagged variable of ln CO2 -0.10087 3 -0.400 0.690 -0.5976162 9
GDP per capita -0.00044 8.79E-05 -5.060 0.000 -0.0006172 -0.00027
GDP per capita squared 3.03E-09 1.03E-09 2.960 0.003 1.02E-09 5.05E-09
0.21342 0.01923 0.25122
Tertiary enrollment rate 1 9 11.090 0.000 0.1756151 6

10
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

0.17487
unemployment rate -0.62534 4 -3.580 0.000 -0.9689779 -0.28171
0.01865 0.00675 0.03192
unemployment rate squared 2 3 2.760 0.006 0.0053815 2
0.45884
year dummy variable -1.53966 7 -3.360 0.001 -2.441305 -0.63801
year;lnCO2 emissions 0.33550 0.01062
interaction variable -0.64866 8 -1.930 0.054 -1.307948 2
year;GDP per capita interaction 0.00012
variable 5.32E-05 3.87E-05 1.370 0.170 -0.0000229 9
13.4180 1.37053 16.1111
Constant 3 1 9.790 0.000 10.72489 7
Table 3. Regression table; the regression of suicide rates per 100000 persons on logged values of
CO2 metric ton emissions per capita, 2 lagged variables of ln CO2, GDP per capita, GDP per
capita squared, tertiary enrollment rate, unemployment rate, unemployment rate squared, a year
dummy variable where 1 = 2000, 2 = 2005, 3 = 2010, an interaction variable between year and
logged CO2 emissions, an interaction variable between year and GDP per capita, and a constant
value.

Number of
Source SS df MS obs 481
F(9, 471) 8.24
448521.
Model 3 9 49835.7 Prob > F 0
6048.40
Residual 2848798 471 3 R-squared 0.136
Adj R-squared 0.1195
6869.41
Total 3297319 480 5 Root MSE 77.771
Residuals squared Coef. Std. Err. t P>t [95% Conf.Interval]
Ln of CO2 emissions, metric ton 22.2177 37.5097
per capita 1 7.78217 2.850 0.004 6.925637 8
0.00108
GDP per capita -0.00467 6 -4.300 0.000 -0.0068077 -0.00254
GDP per capita squared 3.34E-08 1.27E-08 2.620 0.009 8.41E-09 5.85E-08
1.45783 0.23884 1.92716
Tertiary enrollment rate 6 3 6.100 0.000 0.9885058 7
2.17021 4.07947
unemployment rate -0.18502 2 -0.090 0.932 -4.449521 2
0.08380 0.20291
unemployment rate squared 0.03824 5 0.460 0.648 -0.1264368 7
5.52938
year dummy variable -11.8287 7 -2.140 0.033 -22.69399 -0.96335
year;lnCO2 emissions
interaction variable -8.80817 3.63354 -2.420 0.016 -15.94812 -1.66821

11
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

year;GDP per capita interaction 0.00051 0.00047 0.00145


variable 4 9 1.070 0.284 -0.000428 6
35.9487 16.7284 68.8204
Constant 8 5 2.150 0.032 3.07715 1
Table 4. Regression table; testing for heteroskedasticity; the regression of squared residuals on
logged values of CO2 metric ton emissions per capita, GDP per capita, GDP per capita squared,
tertiary enrollment rate, unemployment rate, unemployment rate squared, a year dummy variable
where 1 = 2000, 2 = 2005, 3 = 2010, an interaction variable between year and logged CO2
emissions, an interaction variable between year and GDP per capita, and a constant value.

Ln of CO2 GDP
emissions per
Residual , metric GDP capita Tertiary
s ton per per square enrollment
squared capita capita d rate
Residuals squared 1
Ln of CO2 emissions, metric ton
per capita 0.1589 1
GDP per capita -0.0204 0.5838 1
GDP per capita squared -0.0336 0.3829 0.9041 1
Tertiary enrollment rate 0.2111 0.7041 0.6161 0.3835 1
unemployment rate 0.1099 0.0853 -0.1519 -0.1648 0.0229
unemployment rate squared 0.0869 0.011 -0.1439 -0.134 -0.039
year dummy variable -0.0992 0.0199 0.1631 0.1583 0.1751
year;lnCO2 emissions
interaction variable 0.0938 0.9105 0.6128 0.4426 0.6966
year;GDP per capita interaction
variable -0.0352 0.4967 0.9418 0.9122 0.5597

year;lnC
O2 year;GDP
year emission per
dumm s capita
unemployme y interactio interactio
unemployme nt rate variabl n n
(Continued table) nt rate squared e variable variable
Residuals squared
Ln of CO2 emissions, metric ton
per capita
GDP per capita
GDP per capita squared
Tertiary enrollment rate
unemployment rate 1

12
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

unemployment rate squared 0.9323 1


year dummy variable -0.0403 -0.0465 1
year;lnCO2 emissions interaction
variable 0.0616 -0.008 0.1864 1
year;GDP per capita interaction
variable -0.1143 -0.114 0.3335 0.6187 1
Table 5. Correlation table; finding heteroscedasticity; correlation table between regressors of
original regression.

Suicide Rate per 100000


Persons Coef. Std. Err. t P>t [95% Conf.Interval]
Ln of CO2 emissions, metric 0.69513 0.07749 2.80942
ton per capita 1.44346 4 2.08 0.038 7 2
0.17092 0.33768 0.83448
1st lagged variable of ln CO2 1 5 0.51 0.613 -0.49264 5
0.20850 0.30884
2nd lagged variable of ln CO2 -0.10087 3 -0.48 0.629 -0.51058 7
GDP per capita -0.00044 0.00012 -3.71 0.000 -0.00068 -0.00021
GDP per capita squared 3.03E-09 9.07E-10 3.34 0.001 1.25E-09 4.82E-09
0.21342 0.02599 0.16233
Tertiary enrollment rate 1 9 8.21 0.000 1 0.26451
0.16624
unemployment rate -0.62534 7 -3.76 0.000 -0.95202 -0.29866
0.01865 0.00556 0.02959
unemployment rate squared 2 8 3.35 0.001 0.00771 4
0.47584
year dummy variable -1.53966 7 -3.24 0.001 -2.47471 -0.6046
year;lnCO2 emissions 0.36618 0.07090
interaction variable -0.64866 4 -1.77 0.077 -1.36823 1
year;GDP per capita
interaction variable 5.32E-05 4.42E-05 1.2 0.229 -3.4E-05 0.00014
13.4180 10.6698 16.1662
Constant 3 1.39856 9.59 0.000 1 5
Table 6. Regression table; robust errors; the regression of suicide rates per 100000 persons on
logged values of CO2 metric ton emissions per capita, 2 lagged variables of ln CO2, GDP per
capita, GDP per capita squared, tertiary enrollment rate, unemployment rate, unemployment rate
squared, a year dummy variable where 1 = 2000, 2 = 2005, 3 = 2010, an interaction variable
between year and logged CO2 emissions, an interaction variable between year and GDP per
capita, and a constant value.

Variable VIF 1/VIF


0.03643
yeargdp 27.45 3
0.04736
gdp 21.11 8

13
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

0.08568
yearlnco2 11.67 7
0.08902
lnco2 11.23 3
gdp2 10.67 0.09376
0.11043
unemp 9.05 8
unemp2 8.56 0.11687
0.35496
tertiary 2.82 9
0.62214
year 1.61 9
Mean VIF 11.57
Table 7. VIF table; finding multicollinearity; VIF table of original regression

Variable VIF 1/VIF


0.42457
tertiary 2.36 6
0.44025
lnco2 2.27 3
0.53280
gdp 1.88 3
0.92457
unemp 1.08 6
year 1.07 0.93332
Mean VIF 1.73
Table 8. VIF table; finding multicollinearity; VIF table of original regression without polynomial
term or interaction terms.

Number of
Source SS df MS obs 398
F(10, 387) 29.57
6262.01 626.201
Model 6 10 6 Prob > F 0
8196.30 21.1790
Residual 8 387 9 R-squared 0.4331
Adj R-squared 0.4185
14458.3 36.4189
Total 2 397 5 Root MSE 4.6021
Residuals squared Coef. Std. Err. t P>t [95% Conf.Interval]
0.64946 0.72425
Residuals squared lagged 3 0.03804 17.070 0.000 0.5746718 3
Ln of CO2 emissions, metric ton 0.22026 0.52055 1.24374
per capita 9 9 0.420 0.672 -0.8032094 7

14
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107

GDP per capita 5.49E-05 7.35E-05 0.750 0.455 -0.0000896 0.0002


-8.13E-
GDP per capita squared 10 1.01E-09 -0.810 0.420 -2.79E-09 1.17E-09
0.01112
Tertiary enrollment rate -0.01874 0.01519 -1.230 0.218 -0.0486077 4
0.19333 0.14652 0.48142
unemployment rate 5 7 1.320 0.188 -0.0947535 3
0.00578 0.00659
unemployment rate squared -0.00477 2 -0.830 0.409 -0.0161423 3
0.36431 0.63665
year dummy variable -0.07962 2 -0.220 0.827 -0.7958996 6
year;lnCO2 emissions 0.24285 0.44730
interaction variable -0.03018 6 -0.120 0.901 -0.5076588 6
year;GDP per capita interaction
variable 1.49E-06 3.02E-05 0.050 0.961 -0.0000579 6.09E-05
1.10650 1.48199
Constant -0.69352 6 -0.630 0.531 -2.869032 9
Table 9. Regression table; finding autocorrelation; the regression of residuals squared on
residuals squared lagged, logged values of CO2 metric ton emissions per capita, 2 lagged
variables of ln CO2, GDP per capita, GDP per capita squared, tertiary enrollment rate,
unemployment rate, unemployment rate squared, a year dummy variable where 1 = 2000, 2 =
2005, 3 = 2010, an interaction variable between year and logged CO2 emissions, an interaction
variable between year and GDP per capita, and a constant value.

References
- “CO2 Emissions (Metric Tons per Capita).” CO2 Emissions (Metric Tons per Capita), World Bank,
data.worldbank.org/indicator/EN.ATM.CO2E.PC.
- “Suicide Rates (per 100,000 People).” Suicide Rates (per 100,000 People), World Bank,
data.worldbank.org/indicator/ SH.STA.SUIC.P5?view=chart.
- “GDP per Capita (current US $).” GDP per Capita, World Bank,
data.worldbank.org/indicator/NY.GDP.PCAP.CD?view=chart.
- “Gross Enrollment Ratio, tertiary, both sexes (%).” Gross Tertiary Enrolment ratio,
data.worldbank.org/indicator/SE.TER.ENRR?view=chart.
- “Unemployment, total (%).” Unemployment, data.worldbank.org/indicator/SL.UEM.TOTL.ZS?
view=chart.
- Kim, Y., Myung, W., Won, H., Shim, S., Jeon, H. J., Choi, J., . . . Kim, D. K. (2015). Association between
air pollution and suicide in South Korea: A Nationwide Study (C. Lee, Ed.). PLoS Pne,10(2).
doi:10.1371/journal.pone.0117929
- Lin, G., Li, L., Song, Y., Zhou, Y., Shen, S., & Ou, C. (2016). The impact of ambient air pollution on
suicide mortality: A case-crossover study in Guangzhou, China. Environ Health,15(1), 90th ser.
doi:10.1186/s12940-016-0177-1

15
ECON 497 B1: Effect of air pollution on suicide rates Timothy Lau
1444107
- Balkan, A. V., Huber, R. S., Coon, H., Gray, D., Wilson, P., McMahon, W. M., & Renshaw, P. F. (2015).
Acute air pollution exposure and risk of suicide completion. American Journal of Epidemiology,181(5),
295-303. doi:10.1093/aje/kwu341

16
The effect of air pollution on suicide rates
An in-depth econometrics study on the effects of CO2 emissions and the related quality of the
environment on the suicide rate around the world.

Timothy Lau
1444107
ECON 497 B1

Das könnte Ihnen auch gefallen