Content uploaded by James Jozefowicz
Author content
All content in this area was uploaded by James Jozefowicz on Feb 24, 2021
Content may be subject to copyright.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
92
The Effects of Environmental Factors on Cancer Prevalence
Rates and Specific Cancer Mortality Rates in a Sample of
OECD Developed Countries
Shannon M. Stare and James J. Jozefowicz*
Indiana University of Pennsylvania
Abstract The effects of environmental factors on cancer prevalence and mortality rates are
analyzed empirically. Using data from 30 OECD developed countries for the year 2002, this
study uses environmental factors that have been suggested by other studies to have significant
effects on cancer risk. A control variable for economic growth is also included. The dependent
variables include cancer prevalence rates as well as mortality rates for cancers of the breast,
cervix, colon, lung, and prostate. Independent variables are lagged to account for the long
latency period of cancer. The independent variables can be categorized as follows: air pollutants,
nutrition, lifestyle (all of which are considered to be environmental factors), and economic. The
OLS and WLS results indicate a strong association between cancer rates and total fat intake and
fruit and vegetable consumption. Smoking was also found to be a statistically significant factor
for cancers of the lung, breast, and colon.
Keywords: cancer, environment, cancer prevention
JEL Classification: I10, I18, Q50, Q53
1. Introduction
1.1 Background
Incidence rates of malignant neoplasm—more commonly referred to as cancer—have increased
in prevalence in the past several decades, as people are living longer due to public health
improvements in controlling infectious diseases. Most cancers result from the interaction of
genetic host factors and exposures to environmental health hazards. There are particular genes
known as “oncogenes” which are present in normal cells. DNA damage caused by
environmental factors, such as tobacco smoke, can trigger abnormalities or mutations in these
genes, resulting in increased and abnormal activity of the gene. This can then cause the gene to
become cancerous. Cancer is a group of more than 100 different diseases that are due to
abnormal growth of body cells. A variety of genetic syndromes illustrate how inherited diseases
are component causes of cancers and how environmental exposures may produce unfortunate
reactions, such as cancers, in large populations. Environmental factors by themselves are
believed to explain approximately 80 percent of all cancers, while genetic host factors alone are
believed to explain only 5 percent of all cancers (Morton, 1982).
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
93
Cancer is the second most common cause of death in the First World. According to the World
Health Organization (WHO), an estimated 6.7 million people died of cancer in 2002, which is
approximately 12 percent of total global deaths. In the following year, an estimated 10 million
more individuals were diagnosed with cancer. Of these 10 million people, over half lived in
developed countries (Dauvergne, 2005). Deaths from cancer are projected to continue rising,
with an estimated 9 million people dying from cancer in 2015 and 11.4 million dying in 2030.
These rising global figures reflect better diagnosis, longer average life expectancy, and increased
population size. Furthermore, the financial costs of cancer treatment are a burden to people
diagnosed with cancer, their families, and society as a whole. Between 1995 and 2004, the
overall costs of treating cancer increased by 75 percent. In the United States alone, the bill for
direct medical costs and lost worker productivity totaled nearly $190 billion in 2003 (American
Cancer Society, 2006). With cancer prevalence rates on the rise, cancer centers need to focus
more of their efforts on researching prevention measures and educating the public in order to
reduce the diagnosis of cancer.
The existing body of knowledge about the causes of cancer and about interventions to prevent
cancer is extensive. Cancer control is defined as public health actions that are aimed at
translating this knowledge into practice. It includes the systematic and equitable implementation
of evidence-based strategies for cancer prevention (WHO, 2006). However, research on
prevention tends to focus relatively little on the effects of systematic environmental factors.
Public health programs tend to focus more on the health effects rather than on the causes of ill
health, specifically the environment. Environmental factors are generally medically defined as
“natural and anthropogenic chemical and physical hazards in air, water, soil, foods, consumer
products, and climate; which are usually involuntary due to the need to eat, drink, and breathe in
order to survive” (Kreiger, et al., 2003). As far back as 1843, cancer was coined the “disease of
civilization.” Well over a century later, the WHO in effect still agrees, calling cancer “largely
avoidable and preventable” on account of environmental factors (Dauvergne, 2005).
Human health and environmental health are intimately intertwined. Despite the surge in
international recognition of the link between the environment and health, the burden of disease in
developing countries—including cancer—is increasing. Environmental threats to health are
aggravated by persistent poverty and social inequity (Gopalan, 2003). A medical approach alone
is not sufficient for a holistic understanding of the factors affecting human health; economic,
social, and environmental components may play important roles as well. There is a need to
introduce an economic approach to environmental health through better identification of the
quantitative links between environment, health, and economic growth.
1.2 Most Common Cancers
Cancers of the prostate, lung and colon are the most common types diagnosed among adult
males. Breast, lung, and cervical cancers are the most common among females (Department of
Health, NY, 2003).
Breast cancer is the most common type of cancer in females worldwide. Based on recent studies,
social and environmental factors may be playing a more important role in increasing rates of
breast cancer than genetic factors, contrary to previous thinking. The number of cases has
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
94
significantly increased since the 1970s, a phenomenon partly blamed on modern lifestyles in the
Western world. Dietary influences have been proposed and examined, and recent research
suggests that low fat diets may significantly decrease the risk of breast cancer. Also, recent
epidemiological studies have suggested passive smoking as a possible risk factor in breast cancer
etiology (Khuder, 2000). Although many epidemiological risk factors exist, the majority of
breast cancer prevalence remains unattributable; therefore, the primary cause is unknown.
Lung cancer is the most lethal of all cancers, responsible for approximately 1.2 million deaths
annually. Current research indicates that the environmental factor with the greatest impact on
the risk of lung cancer is long term exposure to inhaled carcinogens, mainly tobacco smoke.
Passive smoking has shown a consistent, significant increase in the risk of being diagnosed with
lung cancer. In the developed world, almost 90 percent of lung cancer deaths are caused by
smoking (Friberg and Cederlof, 1978).
The causes of prostate cancer remain poorly understood. With the exception of age, race, and a
familial predisposition to prostate cancer, there are no well-established risk factors. Striking
international variations in prostate cancer mortality rates suggest that environmental factors play
an important etiologic role (Villeneuve, 1999). One consistent finding concerning prostate
cancer rates is the strong positive correlations with fat intake. However, the primary cause of
prostate cancer is unknown.
Most scientific studies have found that virtually all cases of cervical cancer were due to the
human papillomavirus (HPV) infection, which is a common sexually transmitted disease. In
addition, incidence of cervical cancer has been shown to increase with socioeconomic
deprivation. This suggests that cervical cancer is not necessarily an environmentally caused
disease; socioeconomic factors may be more influential because they directly relate to a person’s
level of exposure to HPV. This is because people’s risk of contracting HPV depends on how
much their lifestyles, behaviors, and environments expose them to the virus and whether they
have the knowledge or ability to reduce their risk of contracting it (Plummer, 2003).
Like most other cancers, colorectal cancers—including colon—should be preventable through an
improved lifestyle because medical research has shown these to be major factors in risk for
contracting them. A comparison of colorectal cancer prevalence rates in various countries
strongly suggests that high caloric intake and a diet high in meat could increase the risk of colon
cancer (Willett, 1995). In contrast, physical exercise and eating fruit and vegetables is expected
to decrease cancer risk.
Although the risk of contracting some of the most prevalent types of cancer are believed to be
influenced by lifestyle and environmental factors, there are few studies that examine this
influence on a large scale or consider the influence of economic factors. In response to the lack
of empirical research investigating which environmental factors significantly influence cancer
rates, the goal of this paper is to examine developed countries and interpret the impact of specific
incremental environmental exposures on prevalence rates of all cancers and mortality rates of
five specific, common cancers. By determining which environmental factors significantly
influence cancer rates, this paper will help cancer prevention programs better allocate research
funding. A variable will also be added to control for socioeconomic differences in the various
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
95
developed countries. The findings of this paper may also encourage additional research and
advocacy into environmental causes.
This paper is organized as follows: the second section will review literature that investigates the
specific causes of cancer. Data and variables used in this study will then be discussed in the
third section. The fourth section will explain the methodology and models used for the study.
The results will be discussed in the fifth section. Finally, conclusions and extensions will be
discussed in the last section.
2. Literature Survey
Despite much discussion concerning the relationship between environmental factors and cancer
rates, research lacks empirical studies using data analysis to document results. One study,
however, conducted a comparative analysis by pooling registry data on identical and fraternal
twins in Sweden, Denmark, and Finland, where large population-based twin registries are
available. When all types of cancer are considered, the study found that even identical twins
seldom develop the same kind of cancer. Lichtenstein, et al. (2000) found that environmental
factors (as opposed to inherited genes) account for 100 percent of cervical and uterine cancer, 78
percent of leukemia and ovarian cancer, 74 percent of lung cancer, 73 percent of breast cancer
and 58 percent of prostate cancer. Because identical twins start out with identical genes at
conception, Lichenstein, et al. (2000) conclude, "that the overwhelming contributor to the
causation of cancer in the populations of twins that we studied was the environment" (p. 85).
Another study conducted by Kneese and Schulze (1974) focused on the impact of socioeconomic
factors, air quality, water quality, and lifestyle factors on cancer mortality rates for six specific
cancers in 60 cities located in the United States. The regressions included numerous suspected
explanatory variables simultaneously in an effort both to achieve a fully specified equation and
to include some measure of the total body burden. It also included lagged explanatory variables
to account for the long latency period of cancer. However, the authors had trouble finding
historical data for some variables. The authors concluded that beef consumption, pork
consumption, cigarette smoking, and ammonium in the atmosphere, which were all lagged
variables, exhibit strong positive correlations with several cancers including digestive, breast,
respiratory, genital, and urinary cancers.
In a similar study, Robertson (1980) examined the “urban factors” of cancer—which include
motor vehicle emissions, industrial pollution, factors in the water supply, and climate on cancer
mortality rates. This study analyzed 98 cities in the United States in the year 1970. The data
was analyzed using linear correlation and stepwise multiple regression techniques. Robertson
found that cancer mortality rates were higher in cities that had more motor vehicles per square
mile and had higher concentrations of barium. Lower cancer mortality rates were found in cities
with more bicarbonate and sodium in water supplies and with warmer climates. Usage of lagged
variables within this study was not an option because the environmental data needed was not
available prior to 1970. Also, at the time of the study, it was difficult for the author to find
sufficient data for alcohol consumption, smoking tobacco consumption, and dietary
consumption. Therefore, these factors were not included in the study.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
96
Willett (1995) provides an overview of statistical findings from various studies on the impact of
diet and nutrition on mainly colon and prostate cancer prevalence rates. His analysis shows that
total energy intake and the consumption of beef, pork, and lamb increase the risk of colon and
prostate cancer. Since researchers did not find a strong association between fat composition and
the risk of colon cancer, there is more evidence that some component of red meat is related to the
risk of colon cancer. Perhaps the most compelling evidence found by the author was the
significant importance of protective factors in fruits and vegetables, which decreased the risk of
cancer. From Willett’s findings, it is roughly estimated that about 32 percent of cancers may be
avoidable by changes in diet.
Consistent with the empirical studies and findings mentioned in the literature, a study on the
environmental factors influencing cancer prevalence rates which uses satisfactory, current data
concerning socioeconomic factors, air quality, water quality, nutritional factors, and lifestyle
factors is the most effective approach to determine significance. Also, lagged variables that were
not accessible for Robertson’s study will be included in this study to account for the delayed
effect of some environmental factors, such as air pollutants. By incorporating 30 countries into
this study, results will show how cancer rates vary by country based on differences in
environmental factors and which factors are statistically significant. By identifying the
significant environmental factors, the results of this study will be useful in helping cancer
prevention programs focus their research and utilize their resources more effectively.
3. Data
This study uses data provided by the Organization for Economic Co-operation and Development
(OECD). All data obtained from OECD were for the years 2002 and 1990. An appropriate
latency period for cancer ranges from 10 to 40 years as suggested by Kneese and Schulze (1974).
However, estimation of appropriate latency periods is filled with uncertainty in existing
literature. Misidentification of the latency period for a given disease can lead to mistaken
estimates of the risk associated with a particular substance. If the actual latency period is shorter
than it is assumed to be in the study, then exposures will be included in the analysis which could
not possibly have contributed (Heinzerling, 1999). In addition, sufficient data was available for
every variable for the year of 1990. Therefore, a latency period of twelve years was chosen for
this analysis.
Dauvergne (2005) mentions that cancer prevalence and mortality rates are significantly larger in
highly industrialized regions because of their increased exposure to environmental factors. In
addition, data necessary to conduct this study are typically only available for developed nations;
therefore, this study includes data for 30 developed countries. Also, by examining international
data as opposed to using United States data only—as most studies have done—one is able to see
the drastic variety of environmental and socioeconomic differences from one region to another
(Samet, 1995). A list of the 30 countries used in this study can be found in Table 1.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
97
3.1 Dependent Variable
Ideally, the measurement of specific cancer rates would be of prevalence. However, due to
limited data, the specific cancer rates are measured by total mortality. Since most of the cancers
used in this study are considered to be more fatal than most, the results of this study are still
relevant. Overall cancer prevalence rates were available. This gives a better estimate of the
number of individuals diagnosed with cancer. The variable does take into account cancers of all
type, including cancers such as skin cancer that are highly less fatal than the specific cancers
used in this study. This study uses the prevalence of cancer per 100,000 people (CANCER) as
the dependent variable in the equation. The specific cancer variables are measured by total
deaths per 100,000 people. The variables for breast and cervical cancer only include females,
and the variable for prostate cancer only includes males.
3.2 Independent Variables
The pollutant variables nitrogen oxide, sulfur oxide and carbon oxide (NOX, SOX, and COX)
are used to measure the amount of chemical compounds in the environment. Recent
epidemiologic studies have suggested consistently that ambient air pollution may be responsible
for increased rates of lung cancer. Relative to cigarette smoking, the excess lung cancer risk
associated with air pollution is lower. However, given the ubiquity of outdoor pollution, the
contribution of this exposure across the general population may be relevant (Friberg and
Cederlof, 1978). Nitrogen oxides (NOX) are emitted in very large amounts from the stacks of
power plants and automobile exhaust pipes. Sulfur and certain sulfur compounds (SOX) are
produced in various industrial processes. Carbon monoxide (COX) is the main poisonous gas in
car exhaust and is present in all cigarette smoke. All three of these pollutants are of concern
because of their negative effects both on human health and on the environment. Due to their
negative impact on health, the pollutant variables are expected to have positive coefficients,
implying that an increase in air pollutants will result in an increase in cancer rates.
The nutritional variables (FAT, FRUITVEG, and CAL) are intended to estimate the effect of
dietary trends on cancer rates. The nutrition variables are also lagged twelve years to account for
the delayed effect of cancer. Unhealthy diets have been possibly linked to several different types
of cancers including cancer of the cervix, breast, colon, and prostate. “Diet has been estimated to
be responsible for between a quarter and one-third of all cancers that occur in economically
developed countries” (Willett, 1995). Fat intake per capita (FAT) and calorie intake per capita
(CAL) are believed to increase cancer rates partly due to their correlation to obesity rates;
therefore, the expected signs for these variables are positive. Fruit and vegetable intake per capita
(FRUITVEG) should have a negative relationship with CANCER because of the protective
factors in fruits and vegetables that have been shown to reduce the risk of cancer. These
protective factors, however, are largely unidentified (Willett, 1995).
The lifestyle variables included in the equation are alcohol consumption (ALCOHOL), tobacco
smoke (SMOKE), and obesity (OBESE). These variables are considered lifestyle factors since
individuals seemingly have the most control of these variables. Tobacco consumption (SMOKE)
is measured by the percentage of the population smoking daily in the year 1990. Not only do
smokers cause damage to their bodies, but they also distribute second hand smoke to the
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
98
environment, which also has shown evidence to increase the risk of lung cancer in non-smokers.
Alcohol consumption (ALCOHOL) is the measurement of annual consumption of pure alcohol
in liters, per person, aged 15 years and over for the year 1990. Alcohol and smoking were the
only lifestyle variables that had sufficient historical data to be lagged. Consumption of large
quantities of alcoholic beverages in conjunction with tobacco smoking sharply increase the risk
of cancer specifically in the upper respiratory and digestive tract, which includes cancers of the
lung, colon, and prostate. (Willett and Trichopoulos, 1996); therefore, SMOKE and ALCOHOL
are expected to both have a positive sign. Lastly, OBESE measures the percentage of the
population with a body mass index (BMI) greater than 30 in the year 1990. Obesity rates
generally imply a lack of physical activity and poor diets which both may increase the risk of
cancer. Higher levels of physical activity have been shown to reduce the risk of cancer,
specifically colon cancer, breast cancer, and prostate cancer (Willett and Trichopoulos, 1996).
Therefore, OBESE is expected to yield a positive relationship with CANCER.
The persistently wide gaps in life expectancy between wealthy and relatively poor citizens in
industrialized countries are well documented. The material deprivation and various kinds of
behavior prevalent in the lower social classes are the most severe threats to a long and healthy
life. There is little doubt that cancer deserves attention from a social equality perspective. To
measure social equality among the developed countries, a composite index measuring average
achievement in three basic dimensions of human development—a long and healthy life,
knowledge, and a decent standard of living—is used. This is known as the human development
index (HUMDEV). HUMDEV is expected to have an ambiguous relationship with cancer rates
since it is not necessarily clear if a healthier life will prevent or increase the likelihood of cancer.
A healthier country may imply that individuals are living longer; since cancer does increase with
age, a longer life could imply a higher risk of cancer. Definitions of the independent and
dependent variables are presented in further detail in Table 2.
3.3 Descriptive Statistics
The mean prevalence rate of cancer in the year 2002 is 266.2167. The United States yielded the
highest value of prevalence rates at 357.7. The two countries with the smallest prevalence rates
of cancer are Turkey and Mexico, with respective values of 114.30 and 147.30. The low cancer
rates of these countries may be attributed to their less industrialized standards as compared to the
other countries used for this study. This is evident because the variable for human development
ranges from 0.682 to 0.929, which implies the economic differences in the countries. The total
emission of carbon monoxide has a range of 489 kilograms per capita. The United States has the
highest amount of total carbon monoxide with 533 kilograms while Japan has the lowest amount
with only 33 kilograms. The United States also has the highest percentage of obesity rates with
30.6 percent of the population having a BMI greater than 30, while Japan and Korea both only
have 3.2 percent of their population considered obese. Descriptive statistics for the dependent
and all of the independent variables are presented in further detail in Table 3.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
99
4. Econometric Model
This study utilizes ordinary least squares (OLS) and weight least squares (WLS) regression
analyses to measure the impact of environmental and economic indicators on all cancer
prevalence rates and specific cancer mortality rates. The original hypothesized equation was
constructed with explanatory variables that had been either shown or predicted to be significant
determinants of prevalence rates of all cancers in previous studies. It was also expected that a
variable controlling for population age structure would be appropriate within the model because
of varying cancer prevalence rates by age group. Data on the percentage of each country’s
population 65 years old and over was collected for the year 2002. This age group was chosen
because cancer prevalence rates tend to be higher for older individuals, and 65 years old and over
was the only age range available for all the countries in this sample. However, investigation of
the data revealed several strong influential points that when removed, caused the slope of the
trend line between the data and cancer prevalence rates to switch from significant positive to
significant negative (see Graphs 1 and 2).1 Because it is unclear why these points are so
influential or whether removing them from the data is justified, this variable was not included in
the model. This conclusion was also reached because an age structure variable was not used in
any previous studies.
Both OLS and WLS are used in the study. The models are estimated with WLS in order to check
the robustness of the OLS results, and to address any concerns over potential heteroskedasticity.
WLS is implemented via the two-step procedure outlined by Maddala (2001, 210).
4.1 Overall Cancer Prevalence Rate Model
Thus, the original hypothesized equation was as follows:
CANCER = β1 + β2COX + β3NOX + β4SOX + β5FRUITVEG +β6CAL +β7FAT +
β8ALCOHOL + β9SMOKE + β10OBESE +β11HUMDEV + ∈ (1)
The step-wise multiple regression method was used to remove insignificant variables and obtain
a model that contained only variables with significance minimally at the 10 percent level. This
method is often used when there is a lack of theory or previous empirical work. For this study,
there is a lack of previous econometric analysis; therefore, the step-wise method was chosen like
Robertson (1980) used for his analysis. This method resulted in one final equation, which
produced all significant explanatory variables:
CANCER = β1 + β2COX + β3FAT + β4FRUITVEG + β5ALCOHOL + β6HUMDEV + ∈ (2)
The White test was used to test for heteroskedasticity with OLS in equation (2). The results of
the White test indicated homoskedasticity was present in the model.
One argument against using the step-wise method is that it may result in omitted variable bias.
This would be true in a case in which an insignificant variable is removed from the model even
though theory or strong past empirical evidence indicates that it should remain in the model.
However, in the case of this study, there is not a strong foundation of theory or empirical
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
100
research dictating that any specific variables be included in a model for cancer prevalence rates.
Therefore, there is not sufficient evidence to justify keeping insignificant variables in the model
simply because literature has suggested they may be significant.
4.2 Specific Cancers’ Models
The models for each specific cancer were originally developed depending upon what previous
literature theorized for each cancer. The final models for the five specific cancers were
hypothesized as follows:
LUNG = β1 + β2SOX + β3SMOKE+ β4HUMDEV+ ∈ (3)
COLON = β1 + β2FAT + β3FRUITVEG + β4HUMDEV + β5SMOKE+ ∈ (4)
PROSTATE = β1 + β2ALCOHOL+ β3FAT+ β4FRUITVEG +∈ (5)
CERVICAL = β1 + β2FRUITVEG + β3SMOKE + β4HUMDEV + ∈ (6)
BREAST = β1 + β2FAT+ β3FRUITVEG+ β4SMOKE + ∈ (7)
5. Results
5.1 Overall Cancer Prevalence Rates
The OLS and WLS results for the first model that contains all of the environmental variables are
reported in Tables 4 and 5, respectively. OLS and WLS regression analysis of Model 1 revealed
two statistically significant coefficients. According to the results, total fruit and vegetable intake
(FRUITVEG) and carbon monoxide (COX) both have a statistically significant relationship at
the 5 percent and 10 percent levels to cancer prevalence rates, respectively. Willett (1995) also
found fruit and vegetable intake to be statistically significant in relationship to cancer rates.
FRUITVEG and COX carried their expected signs, which were negative and positive,
respectively. The OLS regression demonstrated an overall fit of 0.657 based on the adjusted R-
squared while the WLS regression demonstrated a fit of 0.874.
After implementing the step-wise method, the OLS model in Table 6 revealed five of the original
hypothesized explanatory variables to be statistically significant with a higher adjusted R-
squared value of 0.728. The WLS model in Table 7 revealed four of the original hypothesized
variables to be significant with an adjusted R-squared value of 0.860. For both regressions, the t-
statistic of FRUITVEG improves slightly, still remaining statistically significant at the 5 percent
level in this model. Significance of COX improves greatly from 10 percent to 1 percent level in
Model 2. Alcohol consumption in liters per capita (ALCOHOL) and total fat intake (FAT) also
play positive roles in cancer prevalence rates. Human development index (HUMDEV) is also
significant at the 10 percent level with a very large, positive coefficient. A reason for the
positive relationship between cancer rates and human development index may be due to the fact
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
101
that countries that are more developed generally have higher life expectancies. Therefore,
individuals will live longer, which highly increases the likelihood of cancer.
Consistent with the results of Kneese and Schulz (1974), all statistically significant coefficients
were lagged variables. Robertson (1980) found motor vehicle emissions per square mile to be
statistically significant in relationship to cancer rates; since carbon monoxide is one of the main
carcinogenic substances in motor vehicle emissions, the finding of this study is consistent with
Robertson. As hypothesized, FRUITVEG was the only variable to have a negative sign. The
findings of Willett (1995) also showed a negative and significant coefficient on his fruit and
vegetable variable. ALCOHOL, FAT and COX yielded positive signs as expected.
5.2 Lung Cancer
The only two environmental variables that have consistently shown in previous literature to have
effects on lung cancer rates are sulfur oxide and smoking. These two variables were included in
a model on their own; however, after including the variable HUMDEV, both variables’
significance improved and the adjusted R-squared increased. Therefore, HUMDEV was
included in the model even though it did not have significance. In Tables 8 and 9, respectively,
the OLS and WLS results each show that both percentage of total population smoking daily
(SMOKE) and total emissions of sulfur oxide (SOX) were statistically significant at the 1 percent
level. The large, positive coefficient for smoking is consistent with the National Cancer
Institute’s estimate that 87 percent of all lung cancer is caused by tobacco use. The OLS model
has an adjusted R-squared of 0.497 while the WLS model has a value of 0.524.
5.3 Colon Cancer
The OLS regression analysis of colon cancer revealed four statistically significant variables in
Table 10. Consistent with most of the literature concerning colorectal cancers, FAT and
FRUITVEG were both significant, at the 5 percent level. Smoking was also slightly statistically
significant at the 10 percent level. Unlike the model for lung cancer, HUMDEV was statistically
significant in the equation at the 5 percent level with a negative sign. The OLS model has an
adjusted R-squared of 0.531. In Table 11, the WLS regression analysis revealed only two
significant variables, which were FAT and FRUITVEG. The adjusted R-squared for the WLS
analysis has a value of 0.333.
5.4 Prostate Cancer
The best hypothesized model for prostate cancer revealed three statistically significant variables.
In the OLS regression in Table 12, FAT and FRUITVEG proved to be significant again both at
the 1 percent level while ALCOHOL was slightly significant at the 10 percent level. The WLS
regression in Table 13 had very similar results except FAT was no longer significant. Unlike the
previous two models, HUMDEV did not improve the results of the model and was not included
in the final regression for prostate cancer. The adjusted R-squared for the OLS model is 0.425
while the adjusted R-squared for the WLS model is 0.541.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
102
5.5 Cervical Cancer
Previous research has shown that the main contributor to the prevalence of cervical cancer is the
HPV virus. However, other environmental variables have been theorized to play possible roles
as well. For this reason, dietary variables and the smoking variable were analyzed to decide
whether to include such variables in the final regression. HUMDEV was included in the model
in attempt to account for the differences in the stage of development across the varying countries.
A difference in the human development index value may have something to do with contracting
the HPV virus. The OLS regression in Table 14 showed that FRUITVEG, SMOKE, and
HUMDEV were all statistically significant at the 1 percent level. The same three variables were
significant in the WLS regression in Table 15 as well. FRUITVEG and SMOKE had a negative
and positive relationship with cervical cancer rates, respectively, which were their expected
signs. Like the results for colon cancer, HUMDEV carried a negative sign in the equation for
cervical cancer. The adjusted R-squared for the OLS model is 0.847, and the adjusted R-squared
for the WLS model is 0.541.
5.6 Breast Cancer
Although many epidemiological risk factors exist, the primary cause of breast cancer is
unknown. Therefore, many environmental factors were examined to include in the model. The
best model that was used for the final regression of breast cancer included FAT, FRUITVEG,
and SMOKE. In Table 16, all three variables were statistically significant in the OLS regression
analysis at the 1 percent, 5 percent, and 10 percent levels, respectively. In Table 17, only FAT
was statistically significant with WLS. Due to possible omitted variable bias, it is legitimate to
say that these results may be inconclusive concerning the rates of breast cancer. The adjusted R-
squared for the OLS model is 0.697, and the adjusted R-squared for the WLS model is 0.901.
6. Summary and Conclusions
Chosen based on previous literature, environmental and lifestyle variables that can be possibly
manipulated by cancer prevention programs are utilized to estimate cancer prevalence rates and
specific cancer mortality rates in 30 developed countries for the year 2002. Several important
findings are evident. The results show the importance of including lagged variables in the
equation to account for the delayed effect of environmental factors on the prevalence of cancer.
Fruit and vegetable consumption had a negative effect on cancer prevalence rates as well as on
most of the specific cancer mortality rates. Total fat intake also plays a significant role as well.
FAT had a positive relationship with both cancer prevalence rates and most of the specific cancer
mortality rates. Both variables were significant in colon cancer, prostate cancer, and breast
cancer rates. Fruit and vegetables were also significant in cervical cancer rates. While much
remains to be learned, evidence is sufficient to claim that changes in diets—specifically by
decreasing total fat intake and increasing the consumption of fruits and vegetables—can reduce
the likelihood of many cancers.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
103
Smoking was highly significant to lung cancer mortality rates. This finding is not surprising
since most studies have concluded that cigarette consumption increases the likelihood of lung
cancer. More surprisingly, the results show that smoking may have some causation to cancers of
the breast, colon and cervix. Other studies have found weak but significant results concerning
second-hand smoke and breast cancer (Khuder, 2000). Similarly, the epidemiological evidence
on the relationship between cigarette smoking and the risk of colon cancer is inconclusive,
although higher risks for colon cancer precursors have been consistently found among smokers
(Tavani, 1998). As for cervical cancer, smoking decreases the ability to absorb folic acid, and
taking folic acid is a respected way of treating cervical dysplasia, an extremely common
symptom of HPV—which is the primary cause of cervical cancer (Plummer, 2003). Therefore,
this result is consistent with previous literature. More studies are needed to establish whether
the observed associations are causal.
Alcohol only had a positive relationship with prostate cancer mortality rates. It has been
hypothesized that factors capable of modulating the performance of the endocrine system, which
includes alcohol, may be related to prostate cancer. Studies which have examined these factors
have yielded equivocal results (Villeneuve, 1999). Therefore, there is a need for further studies
to investigate the relationship between prostate cancer and alcohol consumption.
Also, another element that needs to be addressed in cancer prevention is the avoidance of
emitting carcinogens into the environment, specifically carbon monoxide and sulfur oxide. Since
the results of this study indicate that carbon monoxide is positively related to the prevalence of
all cancer rates, exposure to it should be controlled. In addition, although sulfur oxide plays a
significantly smaller role in lung cancer than smoking, the effects of the chemical compound
need to be examined further. However, controlling the exposure to pollutants can be difficult
especially in highly industrialized countries. Government policy must evaluate the situation and
determine how to control such exposures.
The human development index was able to combine several aspects of a society to determine
economic differences between the countries. The relationship between HUMDEV and cancer
prevalence rates was positive. This implies that a higher standard of living increases the
likelihood of being diagnosed with all different types of cancers. This could be because more
developed nations may be exposed to more harmful environmental exposures than less
developed nations. The relationship between HUMDEV with both colon and cervical cancer
mortality rates was negative. Since HUMDEV does partly measure a long and healthy life, it
would also make sense that healthier countries with a higher human development value would
have less mortality rates of certain cancers.
By identifying significant environmental factors that are caused by chemical and physical
hazards in air, water, and foods, environmental policy can be potentially manipulated to reduce
prevalence rates of cancer. By determining the significance of these factors, this analysis
suggests that cancer prevention may be possible through controlling environmental factors and
points out which factors are the most important to target. Fortunately, cancer centers are
focusing more of their efforts on researching preventative measures and educating the public.
Many cancer centers are increasing their cancer prevention and control budgets. Econometric
techniques can make an important contribution, but the effectiveness of their application is
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
104
limited when data is insufficient. For studies such as this to be effective in explaining cancer
prevalence rates, countries must have a commitment to consistently and effectively collecting
data related to cancer prevalence rates and the factors influencing them.
6.1 Extensions of Research
Further research should focus primarily on lagged variables and determining appropriate lag
periods. These variables may also need to be measured over time rather than just from a lagged
time period because of the cumulative effect they have on cancer. Also, some alternate measures
of pollution variables may be appropriate. As some past studies have done, further research
should try using variables that measure the level of industrialization as proxies for the pollution
factors. For example, the number of power plants or amount of car exhaust per capita may be
better measures of pollution that people come in contact with on a regular basis. The direct
measures of pollutants used in this study do not indicate how effective a country is at keeping
pollutants contained or at least away from people. Lastly, the effect of population age structure
on cancer prevalence rates should be further investigated.
Endnotes
* Department of Economics, Indiana University of Pennsylvania, Indiana, PA 15705. Email:
james.jozefowicz@iup.edu. Phone: (724)357-4774; Fax: (724)357-6485.
The authors gratefully acknowledge helpful comments and suggestions from Chris Krahe, Pat
Litzinger, Brian O’Roark, and participants at the 2007 Eastern Economics Association Annual
Conference. Special thanks go to Debbie Bacco, Stephanie Bearjar, Stephanie Brewer
Jozefowicz, Elizabeth Hall, and Yaya Sissoko.
1. The data points were determined to be influential because the absolute values of their DfBeta
statistics were greater than 1 (Norusis, 2005). The countries removed for this test were
Turkey, Mexico, and Korea.
References
American Cancer Society. 2006. Cancer Facts & Figures. Atlanta, GA.
Dauvergne, P. 2005. “Cancer and Global Environmental Politics: Proposing A New Research
Agenda,” Global Environmental Politics, 5, 6-13.
Friberg, L. and R. Cederlof. 1978. “Late Effects of Air Pollution with Special Reference to
Lung Cancer,” Environmental Health Perspectives, 22, 45-66.
Gopalan, H. N. B. 2003. “Environmental Health in Developing Countries: An Overview of the
Problems and Capacities,” Environmental Health Perspectives, 111, A446-A447.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
105
Heinzerling, L. 1999. “Environmental Law and the Present Future,” Georgetown Law Journal,
87, 2025-2078.
Khuder, S. A. 2000. “Is There an Association between Passive Smoking and Breast Cancer?”
European Journal of Epidemiology, 16, 1117-1121.
Kneese, A. V. and W. D. Schulze. 1977. “Environment, Health, and Economics – The Case of
Cancer,” American Economic Review, 67, 326-332.
Kreiger, N., F. Ashbury, M. Purdue, and L. Marrett. 2003. “Workshop Report:
Environmental Exposures and Cancer Prevention,” Environmental Health Perspective, 111, 105-
108.
Lichtenstein, P., N. V. Holm, P. K. Verkasalo, A. Iliadou, J. Kaprio, M. Koskenvuo, E.
Pukkala, A. Skytthe, and K. Hemminki. 2000. “Environmental and Heritable Factors in the
Causation of Cancer,” New England Journal of Medicine, 343, 78-85.
Maddala, G. S. 2001. Introduction to Econometrics. Chichester: John Wiley & Sons Ltd.
Morton, J. F. 1982. “Cancer and Environment,” Science, 218, 521.
New York Department of Health. 2003. “Cancer.” Retrieved October 15, 2006, from
http://www.health.state.ny.us/nysdoh/cancer/center/cancerhome.html.
Norusis, M. J. 2005. SPSS 14.0 Statistical Procedures Companion. Upper Saddle River, NJ:
Prentice Hall.
Organization for Economic and Co-Operation and Development (OECD). Retrieved
October 1, 2006, from http://www.oecd.org.
Plummer, M. 2003. “Smoking and Cervical Cancer: Pooled Analysis of the IARC Multi-
Centric Case-Control Study,” Cancer Causes and Control, 14, 805-814.
Robertson, L. S. 1980. “Environmental Correlates of Intercity Variation in Age-Adjusted
Cancer Mortality Rates,” Environmental Health Perspectives, 36, 197-203.
Samet, J. M. 1995. “Controlling the Avoidable Causes of Cancer: Needs and Opportunities
for Etiologic Research,” Environmental Health Perspectives, 103, 307-311.
Tavani, A. 1998. “Cigarette Smoking and Risk of Cancers of the Colon and Rectum: A Case-
Control Study from Italy,” European Journal of Epidemiology, 14, 675-681.
Villeneuve, P. J. 1999. “Risk Factors for Prostate Cancer: Results from the Canadian National
Enhanced Cancer Surveillance System,” Cancer Causes and Control, 10, 355-367.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
106
Willett, W. C. 1995. “Diet, Nutrition, and Avoidable Cancer.” Environmental Health
Perspectives, 103, 165-170.
Willett, W. C. and D. Trichopoulos. 1996. “Nutrition and Cancer: A Summary of the
Evidence,” Cancer Causes and Control, 7, 178-180.
World Health Organization. 2006. The World Health Report 2006 – Working Together for
Health. Retrieved February 10, 2007, from http://www.who.int/whr/2006/en/.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
107
Table 1. Thirty Developed Countries Used for Study
Australia Finland Ireland Netherlands Spain
Austria France Italy New Zealand Sweden
Belgium Germany Japan Norway Switzerland
Canada Greece Korea Poland Turkey
Czech Republic Hungary Luxembourg Portugal United Kingdom
Denmark Iceland Mexico Slovak Republic United States
Table 2. Definitions of Variables
VARIABLE DEFINITION
DEPENDENT VARIABLES
CANCER Total cancer prevalence per 100,000 of population
LUNG Total lung cancer deaths per 100,000 of population
COLON Total colon cancer deaths per 100,000 of population
PROSTATE Total prostate cancer deaths per 100,000 of population (only
including males)
CERVICAL Total cervical cancer deaths per 100,000 of population (only
including females)
BREAST Total breast cancer deaths per 100,000 of population (only
including females)
INDEPENDENT VARIABLES
Pollutant Variables
NOX
Total nitrogen oxide emissions expressed in kilograms per capita
in 1990
SOX Total sulfur oxide emissions expressed in kilograms per capita
in 1990
COX Total carbon monoxide emissions expressed in kilograms per
capita in 1990
Nutritional Variables
CAL Calorie intake per capita per day for 1990
FAT Fat intake in grams per capita per day for 1990
FRUITVEG Total kilograms per capita for 1990
Lifestyle Variables
ALCOHOL
Alcohol consumption in liters per capita (age 15+) in 1990
SMOKE Percentage of total population smoking daily in 1990
OBESE Percent of total population with a BMI>30 for 2002
HUMDEV Calculated on the basis of data on life expectancy, adult literacy
rates, combined gross enrollment ratios, and GDP per capita
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
108
Table 3. Descriptive Statistics
Dependent Variables N Minimum Maximum Mean Std. Deviation
CANCER 30 114.30 357.70 266.21 49.69742
COLON 26 12.00 34.50 20.50 5.74874
LUNG 26 21.40 59.00 36.496 9.05496
BREAST 25 5.50 39.60 22.740 6.57502
CERVICAL 26 .80 7.50 2.9346 1.79754
PROSTATE 26 7.30 43.50 25.338 7.86979
Valid N (listwise) 25
Independent Variables N Minimum Maximum Mean Std. Deviation
NOX 30 11.00 104.00 43.2667 23.71808
COX 30 33.00 522.00 166.5000 122.29099
SOX 29 7.00 181.00 51.3793 40.46375
FAT 27 57.70 161.30 128.0926 25.58456
FRUITVEG 27 112.50 424.50 209.1444 70.25177
CAL 27 2823.00 3709.00 3302.963 226.54962
ALCOHOL 30 1.40 16.10 10.3233 3.47411
SMOKE 27 2.9000000 3.80 3.444 .2025478734
HUMDEV 29 .682 .929 .8747 .053432
Valid N (listwise) 25
Graph 1. Age Structure Variable with Influential Points
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
20.0
0 100 200 300 400
Cancer Incidence Rates
Percentage of Population over 65
Years Old
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
109
Graph 2. Age Structure Variable without Influential Points
With 3 Influential Points Removed
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
20.0
0 100 200 300 400
Cancer Incidence Rates
Percentage of Population
over 65 Years Old
Table 4. OLS Regression Results for Overall Cancer Prevalence Rates, Equation (1)
Dependent Variable : CANCER
Observations : 30
Method : Least Squares
R-Squared = 0.786; Adjusted R-Squared = 0.657
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
1 (Constant) 49.633 359.282 .138 .892
ALCOHOL 3.810 3.151 .272 1.209 .245
FAT .511 .537 .253 .951 .357
FRUITVEG -.255** .095 -.380 -2.690 .017
COX .193** .098 .475 1.977 .067
HUMDEV 148.357 263.699 .161 .563 .582
NOX .052 .526 .026 .098 .923
SOX -.106 .307 -.069 -.344 .736
CAL .007 .058 .031 .113 .911
SMOKE -3.594 35.517 -.015 -.101 .921
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
110
Table 5. WLS Regression Results for Overall Cancer Prevalence Rates, Equation (1)
Dependent Variable : CANCER
Observations : 30
Method : Weighted Least Squares Analysis
R-Squared = 0. 921; Adjusted R-Squared = 0.874
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) -115.412 271.277 -.425 .677
ALCOHOL 4.210 2.685 .275 .175 1.568 .138
FAT .414 .410 .169 .167 1.010 .328
FRUITVEG -.241*** .079 -.288 .095 -3.048 .008
COX .198* .096 .291 .141 2.068 .056
HUMDEV 310.522 202.244 .336 .219 1.535 .146
NOX .006 .500 .002 .175 .011 .991
SOX -.048 .287 -.020 .116 -.168 .869
CAL .004 .046 .014 .159 .090 .930
SMOKE .124 1.085 .013 .112 .114 .911
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Table 6. OLS Regression Results for Overall Cancer Prevalence Rates, Equation (2)
Dependent Variable : CANCER
Observations : 30
Method : Least Squares
R-Squared = 0.780; Adjusted R-Squared = 0.728
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
2 (Constant) -52.827 99.181 -.533 .600
ALCOHOL 3.382* 2.092 .227 1.617 .10
FAT .634** .299 .312 2.120 .046
FRUITVEG -.208** .080 -.281 -2.614 .016
COX .172*** .053 .379 3.235 .004
HUMDEV 249.129* 123.907 .264 2.011 .057
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
111
Table 7. WLS Regression Results for Overall Cancer Prevalence Rates, Equation (2)
Dependent Variable : CANCER
Observations : 30
Method : Weighted Least Squares Analysis
R-Squared = 0.887; Adjusted R-Squared = 0.860
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) -209.180 95.903 -2.181 .041
ALCOHOL 3.014 1.891 .190 .119 1.594 .126
FAT .610** .256 .270 .113 2.385 .027
FRUITVEG -.166** .068 -.199 .081 -2.460 .023
COX .169*** .056 .257 .085 3.019 .007
HUMDEV 399.870*** 114.520 .408 .117 3.492 .002
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
Table 8. OLS Regression Results for Lung Cancer, Equation (3)
Dependent Variable : LUNG
Observations : 26
Method : Least Squares
R-Squared = 0.565; Adjusted R-Squared = 0.497
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
1 (Constant) 9.468 44.386 .213 .833
HUMDEV -7.728 45.962 -.027 -.168 .868
SOX .169*** .047 .573 3.614 .002
SMOKE .805*** .256 .494 3.150 .005
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
112
Table 9. WLS Regression Results for Lung Cancer, Equation (3)
Dependent Variable : LUNG
Observations : 26
Method : Weighted Least Squares Analysis
R-Squared = 0.589; Adjusted R-Squared = 0.524
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) 22.607 53.575 .422 .678
HUMDEV -20.897 56.872 -.056 .152 -.367 .717
SOX .137*** .043 .484 .151 3.196 .005
SMOKE .783*** .202 .573 .148 3.879 .001
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
Table 10. OLS Regression Results for Colon Cancer, Equation (4)
Dependent Variable : COLON
Observations : 26
Method : Least Squares
R-Squared = 0.595; Adjusted R-Squared = 0.531
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
4 (Constant) 59.136** 23.271 2.541 .021
FAT .099** .037 .471 2.696 .015
FRUITVEG -.030** .011 -.450 -2.686 .016
HUMDEV -61.19** 24.892 -.432 -2.458 .025
SMOKE .249* .141 .304 1.763 .096
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
113
Table 11. WLS Regression Results for Colon Cancer, Equation (4)
Dependent Variable : COLON
Observations : 26
Method : Weighted Least Squares Analysis
R-Squared = 0.379; Adjusted R-Squared = 0.333
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) 58.516 38.262 1.529 .145
FAT .059* .033 .381 .209 1.819 .087
FRUITVEG -.029*** .010 -.642 .215 -2.988 .008
HUMDEV -51.714 39.521 -.287 .220 -1.309 .208
SMOKE .205 .153 .288 .214 1.342 .197
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
Table 12. OLS Regression Results for Prostate Cancer, Equation (5)
Dependent Variable : PROSTATE
Observations : 26
Method : Least Squares
R-Squared = 0.535; Adjusted R-Squared = 0.425
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
5 (Constant) 17.412 7.340 2.372 .028
ALCOHOL .889* .469 .332 1.894 .074
FAT .222*** .058 .631 3.811 .001
FRUITVEG -.056*** .018 -.488 -3.129 .006
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Table 13. WLS Regression Results for Prostate Cancer, Equation (5)
Dependent Variable : PROSTATE
Observations : 26
Method : Weighted Least Squares Analysis
R-Squared = 0.603; Adjusted R-Squared = 0.541
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) 42.037 11.390 3.691 .002
ALCOHOL .862* .419 -.407 .198 -2.056 .054
FAT .063 .097 .116 .179 .646 .526
FRUITVEG -.065*** .020 -.545 .166 -3.272 .004
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
114
Table 14. OLS Regression Results for Cervical Cancer, Equation (6)
Dependent Variable : CERVICAL
Observations : 26
Method : Least Squares
R-Squared = 0.869; Adjusted R-Squared = 0.847
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std.
Error Beta
6 (Constant) 37.680 4.280 8.803 .000
FRUITVEG -.009*** .002 -.385 -4.500 .000
HUMDEV -39.709*** 4.454 -.786 -8.915 .000
SMOKE .068** .026 .231 2.632 .017
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Table 15. WLS Regression Results for Cervical Cancer, Equation (6)
Dependent Variable : CERVICAL
Observations : 26
Method : Weighted Least Squares Analysis
R-Squared = 0.603; Adjusted R-Squared = 0.541
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std.
Error Beta Std. Error B Std. Error
(Constant) 50.291 5.086 9.889 .000
FRUITVEG -.011*** .002 -.356 .076 -4.656 .000
HUMDEV -50.41*** 5.021 -.802 .080 -10.039 .000
SMOKE .059** .026 .184 .081 2.276 .035
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
Table 16. OLS Regression Results for Breast Cancer, Equation (7)
Dependent Variable : BREAST
Observations : 25
Method : Least Squares
R-Squared = 0.745; Adjusted R-Squared = 0.697
Model Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta
7 (Constant) -21.273 14.456 -1.472 .161
FAT .216*** .032 .850 6.648 .000
FRUITVEG -.021* .010 -.266 -2.092 .053
SMOKE 5.689 3.839 .188 1.482 .158
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level.
Stare and Jozefowicz, International Journal of Applied Economics, September 2008, 5(2), 92-115
115
Table 17. WLS Regression Results for Breast Cancer, Equation (7)
Dependent Variable : BREAST
Observations : 26
Method : Weighted Least Squares Analysis
R-Squared = 0.916; Adjusted R-Squared = 0.901
Unstandardized
Coefficients Standardized
Coefficients t Sig.
B Std. Error Beta Std. Error B Std. Error
(Constant) -5.450 6.054 -.900 .381
FAT .218*** .018 .949 .077 12.344 .000
FRUITVEG -.015 .010 -.108 .071 -1.520 .147
SMOKE .053 .138 .029 .076 .383 .707
*Significant at 10% level; **Significant at 5% level; ***Significant at 1% level
