Determinants of Dying
Coronary Artery Disease
Dr. Jackie Khorassani
Instructor of Econ 421
January 5, 2012
In 2001, 700,000 people died from coronary artery disease. Coronary artery disease is the
hardening of the arteries near the heart. This hardening can lead to reduced blood flow, heart
attacks, and death (“What is Coronary...” 2003). Why are so many people dying from coronary
artery disease? With the use of OLS regression analysis of 49 observations in the year 2000, I
will examine thirteen variables that may affect coronary artery disease. The thirteen variables are
education level, income level, lack of health insurance, state health expenditures, alcohol
consumption, inactivity, state stress levels, tobacco consumption, average age of the population,
diabetes, high blood cholesterol, high blood pressure, and obesity.
This study is organized in seven sections. Section one is the introduction. Section two is
a description of each variable and the reasons for each variable’s inclusion in my model. Section
three discusses the meaning of the raw data. Section four and five test my regression analysis for
common errors. Section six discusses the significance of each variable and the effects of each
significant variable. Section seven concludes with a brief wrap up of this study.
For the purpose of measuring the effects of thirteen factors on the death rate caused by
coronary artery disease among the US population in 2001, Equation 1 is estimated with a cross
sectional data set that consists of 49 observations from 49 US states1. The method of estimation
is OLS, and the estimation software is EViews:
Equation 1: LCAD = F (EDU, INC, LHI, SHE, ALC, INY, SSL, TOB, AAP, DIB, HBC, HBP, OBY) + error term
Florida was excluded for lack of state health expenditure data.
The dependent variable, LCAD, is the population per 100,000 that dies from coronary
artery disease. Table 1 includes the definition of the independent variables, and their expected
effect on the dependent variable.
Table 1: Independent Variables
Variables Definitions Expected Sign of the
EDU percentage of the 25 and older population that has a college degree negative
INC per capita disposable personal income in current dollars negative
LHI percent of the population without health insurance positive
SHE per capita dollars spent by the state on health care negative
ALC average gallons of beer consumed per person ambiguous
INY percentage of adults with no leisure-time physical activity positive
SSL percent of the population of each state living in metropolitan areas ambiguous
TOB percentage of the population that has reported having smoked 100 positive
or more cigarettes during their lifetime and who currently smoke
every day or some days
AAP average age of the population positive
DIB percent of the population that has been diagnosed with diabetes positive
HBC percent of the population with high cholesterol positive
HBP percentage of adults with high blood pressure positive
OBY percentage of adults who were obese positive
Coronary artery disease is the hardening of the arteries near the heart that leads to reduced blood
flow, heart attacks, and can lead to death (“What is Coronary...” 2003).
The nature of my thirteen independent variables allows me to summarize them in three
categories: economic independent variables, lifestyle independent variables, and medical/genetic
independent variables. The economic independent variables are EDU (education), INC (income),
LHI (lack of health insurance), and SHE (state health expenditures). The lifestyle independent
variables are ALC (alcohol), INY (inactivity), SSL (state stress levels), and TOB (tobacco). The
medical/genetic independent variables are AAP (average age of the population), DIB (diabetes),
HBC (high blood cholesterol), HBP (high blood pressure), and OBY (obesity).
Economic Independent Variables:
The first economic variable is EDU, or education. EDU measures the percentage of the
25 and older population that has a college degree in each state in the year 2000. The effect of
education on the number of cases of lethal coronary artery disease overlaps with the effect of
many of the other variables in my study. The Tromso Heart Study (1988) found that more
educated people are less likely to be overweight, seem to smoke less, are more physically active,
and have better diets (“Risk factors for...” 1988). That is why I expect the sign of the coefficient
of EDU to be negative.
The second economic variable is INC, or income. More specifically, INC measures the
per capita disposable personal income in current dollars in each state in the year 2000. A health
study (2001) out of Canada shows that in general the higher the income level, the more likely a
person is to have an active lifestyle, a healthy weight, not smoke, and not drink dangerous
amounts of alcohol. (“Health and Wealth...” 2001). Due to this study, I expect the sign of the
coefficient for INC to be negative.
The third economic variable is LHI, or lack of health insurance. LHI measures the
percentage of the population who did not have health insurance in each state in 2002. According
to a 2003 report by the Robert Wood Johnson Foundation, the uninsured are more likely than
those who have health coverage to receive second-rate care and to die from health-related
problems (Anil Kumar. 2004). Due to this study’s findings, I expect the sign of the coefficient
of LHI to be positive.
The fourth economic variable is SHE, or state healthcare expenditures. SHE measures
the per capita dollars spent by the state on health care in each state in the year 2000. The effect
of this variable on the percentage of the population that dies from coronary artery disease is
similar to the effects of income and health insurance. State healthcare expenditures can take
different forms such as funding clinics and hospitals, state-funded insurance, and funds given
directly to citizens to spend on healthcare. The money spent on clinics and hospitals improve the
quality of the care provided, which will decrease lethal coronary artery disease. State-funded
health insurance and funds given to citizens increase the quantity of health care a person can
afford, meaning fewer cases of lethal coronary artery disease. Due to these effects, I expect the
sign of the coefficient for SHE to be negative.
Lifestyle Independent Variables:
The first lifestyle variable is ALC, or alcohol consumption. More specifically, ALC
measures the average gallons of beer consumed per person in each state in the year 2000. The
effects of alcohol on the heart are a little questionable. One study by the Cleveland Clinic Heart
Center (2004) has found that, “moderate alcohol consumption (wine or beer) does offer some
protection against heart disease for some people (“Heart Disease: Alcohol...” 2004).” Alcohol’s
poisonous effects, however, may be dangerous to the heart. The same article warns that those
who already have heart disease should avoid alcohol, and it also warns to not start drinking
because the same benefits made by alcohol can be produced through healthy eating and exercise
(“Heart Disease: Alcohol...” 2004). Due to the uncertainty of the effects of alcohol, I expect the
sign of the coefficient of ALC to be ambiguous.
The second lifestyle variable is INY, or inactivity. This variable is measured as the
percentage of adults who reported no leisure-time physical activity in each state in 2000.
Inactivity prevents the heart from benefiting from exercise. There are numerous benefits to
exercise for the heart including strengthening the heart and cardiovascular system, improving
circulation and helping the body use oxygen better, improving heart failure symptoms, lowering
blood pressure and helping reduce stress, tension, anxiety and depression (“Heart Disease:
Exercise...” 2004). Being active obviously has good effects for the heart, so I predict that the
sign of the coefficient of INY will be positive.
The third lifestyle variable is SSL, or state stress levels. SSL is measured as the
percentage of the population of each state living in metropolitan areas in 2000. This is not a
perfect measure of stress. It doesn’t include other sources of stress outside of living in a city,
such as the number of children per couple, the nature of their jobs, or how well people respond to
stressful situations. Data for these and any other sources of stress are not available; therefore I
was unable to include them in my measure of stress. I also have to consider that people in
metropolitan areas most likely have better access to healthcare, which may affect the results for
this variable. The connection between stress and heart health has not been proven. That is
because, according to an article by the Texas Heart Institute (2004) on heart disease, people
define and respond to stress in different ways (“Causes of Heart Disease.” 2004). It is hard to
determine why stress may be damaging to the heart. In general, however, the article points to
three effects of stress that would have a damaging effect on the heart. Those three are: 1)
stressful situations increase heart rate and blood pressure, which makes the heart demand
additional oxygen, 2) during stress, extra hormones are released which causes blood pressure to
increase, 3) and stress also increases the amount of clotting agents that are flowing in the blood.
The need for additional oxygen can cause angina (pain of and around the heart) in persons with
preexisting heart disease. Angina can damage the heart and blood vessels further, leading to
hardening of the arteries. The increase in blood pressure can damage artery walls. When this
damage heals, the arteries may become hard and more prone to collect plaque. Additional
clotting agents in the blood make it more likely to form a clot in arteries that are already partially
blocked by plaque (“Causes of Heart Disease.” 2004). Considering these effects of stress on the
heart, but keeping in mind the issues of measuring stress in this manner, I am expecting the sign
of the coefficient of SSL to be ambiguous.
The fourth lifestyle variable is TOB, or tobacco smoke consumption. TOB is measured
as the percentage of the population that has reported having smoked 100 or more cigarettes
during their lifetime and who currently smoke every day or some days in each state as of the year
2000. There is a strong link between smoking and developing lethal coronary artery disease.
According to The Cleveland Clinic, smoking increases risk of coronary artery disease in four
ways: 1) decreased oxygen to the heart, 2) increased blood pressure and heart rate, 3) increased
blood clotting, and 4) damage to cells that line coronary arteries and other blood vessels (“Heart
Disease: Smoking...” 2004). I already established that these four effects are damaging and will
lead to more cases of lethal coronary artery disease. That is why I expect that the sign of the
coefficient of TOB is positive.
Medical/Genetic Independent Variables:
The first medical/genetic variable is AAP, or average age of the population. AAP is
measured exactly how it sounds, the average age of the population in every state in 2000. It is
common sense that the older you are, the more health problems you are likely to have. Statistics
also show that about 80% of the deaths from coronary artery disease are people age 65 and older
(“Coronary.” 2005). Therefore, I predict that the sign of the coefficient of AAP is positive.
The second medical/genetic variable is DIB, or diabetes. This is measured as the
percentage of the population that has been diagnosed as having diabetes in each state in the year
2001. The reasons why diabetes increases cases of coronary artery disease are not completely
understood. However, according to The Cleveland Clinic, the high glucose levels in the blood
from diabetes may damage the small blood vessels of the heart and predispose a person to
atherosclerosis (hardening) of the large arteries (“Diabetes...” 2004). Since this is the definition
of coronary artery disease, I expect the sign of the coefficient of DIB to be positive.
The third medical/genetic variable is HBC, or high blood cholesterol. HBC is measured
as the percentage of the adult population in each state that reported having high cholesterol in
2001. According to the National Heart Lung and Blood Institute (2003), too much cholesterol in
your blood can build up in the walls of your arteries. This buildup of cholesterol is called plaque.
Over time, plaque can cause hardening of the arteries (“What is Coronary...” 2003). Given that
coronary artery disease is defined as hardening of the arteries high cholesterol obviously is a
cause of coronary artery disease (“What is Coronary...” 2003). Therefore, I expect the sign of the
coefficient of HBC to be positive.
The fourth medical/genetic variable is HBP, or high blood pressure. This is measured as
the percentage of adults who have ever been told by a health-care provider that they have high
blood pressure. High blood pressure may be caused by smoking, excessive alcohol consumption,
inactivity, and obesity, all of which are a part of my thirteen independent variables (“Heart
Disease: Risk ...” 2004). However, to a certain extent, getting high blood pressure seems to be
genetic, and may not be a good indicator of the overall health of the circulatory system.
Considering that the long list of possible causes of high blood pressure makes it the most
common coronary artery disease risk factor, it still should have a direct impact.
High blood pressure increases cases of lethal coronary artery disease for two reasons: 1)
it makes the heart work harder to supply the body with blood and 2) it contributes to the
hardening of the arteries.
Why does it make the heart work harder? First, for clarification, high blood pressure
causes the heart to work harder but the heart working harder does not necessarily cause high
blood pressure. Blood pressure is determined by two forces: 1) the pumping of the heart, and 2)
the force of the arteries resisting the blood flow (“Blood Pressure” 2004). In most cases, it is the
increase of resistance to the blood flow from the arteries that causes high blood pressure. As the
resistance to blood flow is increased, the heart must work harder to accomplish its job. A harder
working heart has a shorter life. If that isn’t bad enough, the increase in resistance to blood flow
from the arteries happens when arteries are damaged and harden. High blood pressure,
therefore, can be considered a sign that there may be some coronary artery disease present.
Therefore, I predict the sign of the coefficient of HBP is positive.
The fifth medical/genetic variable is OBY, or obesity. OBY is measured as the
percentage of adults in the US who were obese in each state in 2001. Since inactivity can cause
obesity, they have the same links to coronary artery disease, but there are additional effects from
obesity. A study by the American Heart Association (1997) finds that obesity is connected to
heart disease both indirectly (through other factors) and directly (“Obesity and Heart Disease”
1997). Considering this evidence, I expect the sign of the coefficient of OBY to be positive.
Table 2 shows the lowest values and corresponding states, the highest values and their
corresponding states, and the mean values for each variable.
Table 2: Data Analysis: Maximum, Minimum, and Mean
Variable Minimum Maximum Mean
LCAD 171.0 - Minnesota 329.0 - Mississippi 238.12
EDU 15.3% - West Virginia 34.6% - Colorado 25.0%
INC $19,258 - West Virginia $32,556 - Connecticut $24,076
LHI 7.9% - Minnesota 21.1% - New Mexico 13.8%
SHE $504.09 – Nevada $2,001.49 - Alaska $970.55
ALC 12.59 gallons – Utah 33.09 gallons - Nevada 22.87 gallons
INY 15.5% - Utah 41.1% - Kentucky 26.8%
SSL 27.8% - Vermont 100.0% - New Jersey 67.2%
TOB 12.9% - Utah 29.1% - Nevada 22.9%
AAP 27.1 years – Utah 38.9 years - West Virginia 35.5 years
DIB 2.71% - Alaska 6.08% - West Virginia 4.37%
HBC 24.8% - New Mexico 37.7% - West Virginia 30.5%
HBP 14.0% - Arizona 31.6% - Alabama 24.6%
OBY 13.8% - Colorado 24.3% - Mississippi 19.5%
Note: LCAD = lethal coronary artery disease; EDU = education; INC = income; LHI = lack of health insurance; SHE = state health
expenditures; ALC = alcohol consumption; INY = inactivity; SSL = state stress levels; TOB = tobacco consumption; AAP = average
age of the population; DIB = diabetes; HBC = high blood cholesterol; HBP = high blood pressure; OBY = obesity
According to my data, the worst state to live in when worried about lethal coronary artery
disease is Mississippi, and the best is Minnesota. The difference between these two extremes is
158 per 100,000 people.
The biggest observation involves West Virginia. This state appears five times in table
two, and each time, it is not a good thing pertaining to lethal coronary artery disease. They have
the minimum in EDU (education), a variable expected to have a negative affect on LCAD (lethal
coronary artery disease). They also have the minimum in INC (income), which is closely related
to EDU (education). This too is expected to have a negative affect on LCAD (lethal coronary
artery disease). West Virginia next appears as the maximum for AAP (average age of the
population), a variable expected to affect LCAD (lethal coronary artery disease) positively. The
state also appears as the maximum for DIB (diabetes) and HBC (high blood cholesterol), which
also is expected to affect LCAD (lethal coronary artery disease) positively. Each appearance as
maximum or minimum shows that West Virginia is expected to be more likely to develop lethal
cases of coronary artery disease. It would seem that with this much working against it, West
Virginia would most likely be the maximum for LCAD (lethal coronary artery disease), but they
are not. However, they do come in second from the maximum at 296 per 100,000 people, only
33 per 100,000 people below Mississippi.
Another state that sticks out in Table 2 is Utah. Utah holds the minimum for four
variables. All four variables, ALC (alcohol), INY (inactivity), TOB (tobacco), and AAP (average
age of the population), are expected to have a negative effect on LCAD. With Utah holding this
many minimums for variables expected to have a negative coefficient, it is likely that Utah is
very low on the percentage of lethal coronary artery disease deaths. In fact, they are third from
the minimum at 185.2 per 100,000 people, just 14.2 per 100,000 people above Minnesota.
Alaska is also interesting. It holds the maximum in state healthcare expenditures and the
minimum in diabetes. New Mexico also appears twice in Table 2, first as the maximum for lack
of health insurance, and second, the minimum for high blood cholesterol. Nevada appears twice
as the minimum for state healthcare expenditures and the maximum for tobacco use. The next
state to appear twice is Colorado, who holds the minimum for obesity and the maximum for
Any equation must be tested for problems that may affect the results of the estimation.
One such problem is multicollinearity. A. H. Studenmund (2001) states that multicollinearity is
either perfect or imperfect (A. H. Studenmund. 2001). Perfect multicollinearity is a violation of
the classical assumption that no independent variable is a perfect linear function of any other
independent variable. With perfect multicollinearity the variable’s coefficient cannot be
determined, and the standard error for the coefficients is infinite. Imperfect multicollinearity is
when the linear function between two or more independent variables is strong enough to affect
the estimation results. Imperfect multicollinearity results in increased variance and standard
errors of the coefficients and decreased t-statistics. Multicollinearity, however, does not bias the
coefficients of the equation and the overall accuracy of the equation is not affected.
The test for multicollinearity involves examining the correlation coefficients. The
correlation coefficient is not considered a problem unless the absolute value of any correlation
coefficient is higher than 0.7 and is higher than the correlation between the dependent variable
and the corresponding independent variables.
Table 3: Correlation Coefficients
LCAD EDU INC LHI SHE ALC INY SSL TOB AAP DIB HBC HBP OBY
LCAD 1 -0.513 -0.260 0.150 0.092 -0.079 0.706 0.070 0.587 0.195 0.831 0.478 0.530 0.631
EDU 1 0.761 -0.271 0.127 -0.229 -0.373 0.444 -0.556 -0.018 -0.475 -0.270 -0.461 -0.597
INC 1 -0.260 0.219 -0.172 -0.281 0.682 -0.241 0.141 -0.256 0.042 -0.225 -0.512
LHI 1 -0.064 0.116 0.059 -0.002 0.050 -0.429 0.044 0.010 0.064 0.189
SHE 1 -0.153 0.069 0.181 0.122 0.193 0.069 -0.134 0.010 0.026
ALC 1 0.061 -0.318 0.348 0.193 -0.077 -0.002 0.072 -0.052
INY 1 -0.135 0.463 0.230 0.604 0.188 0.233 0.497
SSL 1 -0.144 -0.153 0.164 0.138 0.112 -0.164
TOB 1 0.386 0.442 0.368 0.583 0.482
AAP 1 0.318 0.254 0.101 -0.075
DIB 1 0.426 0.546 0.596
HBC 1 0.448 0.273
HBP 1 0.482
Note: Any correlation coefficients with an absolute value more than 0.7 is underlined by a thick line and italicized. Any correlation coefficient with an
absolute value that is almost 0.7 is underlined by a dotted line and italicized. Any correlation coefficient with an absolute value that is larger than the
correlation coefficients between the dependent variables and the independent variable are underlined by a thin line and italicized.
As you can see in Table 3, there is only one clear multicollinearity problem between my
thirteen independent variables. The high correlation coefficient of 0.761 is between INC
(income) and EDU (education). That is clearly higher than 0.7. This correlation coefficient is
also larger than the absolute value of the correlation coefficient between LCAD (lethal coronary
artery disease) and EDU (education) and larger than the absolute value of the correlation
coefficient between LCAD (lethal coronary artery disease) and INC (income). These results
show that there is a severe multicollinearity problem between INC (income) and EDU
The correlation coefficient between SSL (state stress levels) and INC (income) is too near
0.7 to ignore. The issue is amplified since the correlation coefficient between LCAD (lethal
coronary artery disease) and SSL (state stress levels), and the correlation coefficient between
LCAD (lethal coronary artery disease) and INC (income) are considerably smaller than the
correlation coefficient between SSL (state stress levels) and INC (income). These results show
there may be a severe multicollinearity problem between SSL (state stress levels) and INC
The correlation coefficients between ALC (alcohol) and SHE (state healthcare
expenditures), SSL (state stress levels) and SHE (state healthcare expenditures), SSL (state stress
levels) and ALC (alcohol), and AAP (average age of the population) and LHI (lack of health
insurance) all may have a multicollinearity problem. Each one is more than the correlation
coefficient between each of these independent variables and the dependent variable. These
results show there may be a multicollinearity problem between each of these pairs. Considering
the very small size of these correlation coefficients, the equations should not be affected
significantly; therefore I am not doing anything to fix these problems.
In order to limit the effects of multicollinearity, Equation 1 will be split into two
variations: Equation 1-A and Equation 1-B. Equation 1-A will exclude EDU (education) and
SSL (state stress levels) and equation 1-B will exclude INC (income).
Heteroskedasticity is another issue to deal with when estimating an equation. As defined
by A. H. Studenmund (2001), heteroskedasticity is a violation of the classical assumption that the
observations of the error terms are drawn from a distribution that has a constant variance (A. H.
Studenmund. 2001). There are two types of heteroskedasticity: pure and impure. Pure
heteroskedasticity occurs when the assumption is violated even though the equation is correctly
specified. Correctly specified means there are no irrelevant or omitted variables, the functional
form is correct (linear), and there are no sample errors. In the case of pure heteroskedasticity the
coefficients of the variables are not biased, but the t statistics are bigger than they should be,
which results in a bigger chance that a variable will be considered relevant. Impure
heteroskedasticity occurs when the equation is not correctly specified (i.e. irrelevant or omitted
variables, wrong functional form, sample errors). The results of impure heteroskedasticity are
biased variable coefficients and incorrect standard errors.
In order to test for heteroskedasticity, I am using the white test, named after its creator
Halbert White. The white test has three steps. The first step is to obtain the residuals of
Equation 1-A. The second step is to use these residuals squared as the dependent variable in a
second equation. The independent variables of the second equation are the independent variables
of Equation 1-A, the squares of the independent variables of Equation 1-A, and the products of
each two independent variables of Equation 1-A. However, the white test, although considered
the best for cross sectional equations, does have one flaw. It cannot be used if, in the second
equation, there are more variables than observations. I have 49 observations, but the second
equation for Equation 1-A has more than 49 variables. The only way for the white test to work
here is to use its other form, which drops the products of each two independent variables, and
only uses the independent variables and their squares. Here, I would have the same 49
observations, but only 22 variables. This test is also sufficient to determine if there is a
heteroskedasticity problem. The third step is to multiply the number of observations by the
unadjusted R2 (n*R2). The decision rule is that if n*R2 is greater than critical chi squared, then
there is a heteroskedasticity problem. For Equation 1-A, n*R2 = 24.26 and chi squared with
degrees of freedom 22 = 33.92. Repeat the three steps for Equation 1-B, which also will use the
simple version of the white test. For Equation 1-B, n*R2 = 27.24 and chi squared with degrees of
freedom 24 = 36.41. By the decision rule, I find that there is no serious problem with
heteroskedasticity in Equation 1-A or Equation 1-B.
Empirical Estimation Results:
Table 4 reports the results of the estimation of Equation 1-A, and Equation 1-B.
Table 4: Estimation Results for Equation 1-A and Equation 1-B
Independent Variables Variations of Equation 1 Expected Sign
Equation 1-A Equation 1-B
Intercepts 40.25261 (0.499160) 56.58221(0.657541)
EDU 40.05244 (0.379967) negative
INC 0.000110 (0.113182) negative
LHI 56.77456 (0.712506) 55.23382 (0.686585) positive
SHE 0.004282 (0.430282) 0.005972 (0.596954) negative
ALC -1.223578 (-1.595043) -1.307321 (-1.675342) ambiguous
INY 199.6210 (2.969297) 186.9634 (2.654275) positive
SSL -13.09042 (-0.653808) ambiguous
TOB 347.0400 (2.599307) 374.4734 (2.549625) positive
AAP -2.839518 (-1.438962) -3.610169 (-1.558149) positive
DIB 2676.302 (4.645088) 2953.804 (4.038012) positive
HBC 246.6707 (1.824513) 269.3550 (2.008087) positive
HBP -41.07500 (-0.334783) -38.39052 (-0.310557) positive
OBY -14.01551 (-0.085363) -63.05713 (-0.372228) positive
Adjusted R2 0.797766 0.794527
Note: t-statistics are in parenthesis ( )
thick underline = significant at 99.5% level of certainty,
thin underline = significant at 99% level of certainty,
double underline = significant at 95% level of certainty, and
dotted underline = significant at 90% level of certainty.
As observed from Table 4, the adjusted R2 for Equation 1-A is 0.797, and the adjusted R2
for Equation 1-B is 0.794. According to A. H. Studenmund (2001), the closer the adjusted R2 is
to 1, the closer the estimated equation fits the data (A. H. Studenmund. 2001). Therefore, the
adjusted R2 is quite strong for both equations and slightly stronger for Equation 1-A.
In order to test the hypothesis that the coefficients have a significant impact on LCAD
(lethal coronary artery disease), the t-test will be used. For the coefficients whose signs are
expected to be positive or negative, I will use a one-sided test. The decision rule for a one-sided
test is if the absolute value of the t-statistic (in Table 4) is greater than the absolute value of the
critical-t then that coefficient is significant. For coefficients with signs that are expected to be
ambiguous, I will use a two-sided test. The decision rule for a two-sided test is if the positive t-
statistic is greater than the positive critical-t, or the negative t-statistic is less than the negative
critical-t then that coefficient is significant.
Out of the thirteen variables, eight are not significant. These eight are LHI (lack of health
insurance), HBP (high blood pressure), OBY (obesity), EDU (education), INC (income), SHE
(state health expenditures), ALC (alcohol), and SSL (state stress levels). All eight are tested at
the 90% level of certainty, and are found to be insignificant.
There are five significant variables, all of which are expected to effect lethal coronary
artery disease positively. For 99.5% level of certainty, the critical-t is 2.704. Out of the five
significant independent variables only DIB (diabetes) for both equations is significant at this
level. This means with 99.5% certainty, every additional one percentage point of the population
that has been diagnosed with diabetes causes a 2,676.302 to 2,953.804 rise in the population per
100,000 dying from lethal coronary artery disease.
For 99% level of certainty, the critical-t is 2.423. At this level, INY (inactivity) and TOB
(tobacco) for both equations are significant. Specifically for INY, this means that with 99%
certainty, every additional one percentage point of the population with no leisure-time activity
causes a 186.9634 to 199.621 increase in the population per 100,000 dying from lethal coronary
artery disease. For TOB, this means that with 99% certainty, every additional one percentage
point of the population that has ever smoked 100 or more cigarettes and currently smoke causes a
347.04 to 374.4743 increase in the population per 100,000 dying from lethal coronary artery
For 95% level of certainty, the critical-t is 1.684. This level has HBC (high blood
cholesterol) for both equations as being significant. This means that with 95% certainty, for
every additional one percentage point of the population with high cholesterol there is a 246.6707
to 269.355 increase in population per 100,000 dying from lethal coronary artery disease.
For 90% level of certainty, the critical-t is 1.303. For this level, AAP (average age of the
population) for both equations is significant. This poses a problem. The sign for the coefficient
for AAP is expected to be positive, but the estimated coefficient turns out to be negative. One
explanation is an omitted variable. An omitted variable can bias the estimated coefficient of a
variable if one of the following is true: the correlation coefficient (r) between the omitted
variable and LCAD (lethal coronary artery disease) is negative, and the estimated coefficient (β)
for the omitted variable is positive, or r is positive, and the β is negative. One possible omitted
variable is a measure of the quality of health care which would have a positive r and a negative β.
Out of my thirteen variables, only SHE (state healthcare expenditures) measures any of the
quality of healthcare, and that is only a small part of that variable. That is why I believe the
incorrect sign is from the omitted variable of quality of health care.
In brief, this paper investigates the effects thirteen variables have on lethal coronary artery
disease through OLS regression analysis using 49 observations in the year 2000. Equation 1 was
tested for multicollinearity and heteroskedasticity. There was no problem with
heteroskedasticity, but there was a problem with multicollinearity. In order to fix it, Equation 1
was split into two variations: Equation 1-A and Equation 1-B. Each new equation was estimated
resulting in a very high accuracy (adjusted R2) with Equation 1-A being slightly more accurate.
Out of the thirteen variables, eight were insignificant. These variables are LHI (lack of
health insurance), HBP (high blood pressure), OBY (obesity), EDU (education), INC (income),
SHE (state health expenditures), ALC (alcohol), and SSL (state stress levels). That leaves five
variables that significantly affect lethal coronary artery disease. These five are DIB (diabetes),
INY (inactivity), TOB (tobacco), HBC (high blood cholesterol), and AAP (average age of the
population). AAP is the only variable that is significant in the opposite way that was expected. I
determined that this was due to an omitted variable, namely, the quality of health care. My
results show that in order to avoid coronary artery disease, the best strategy is to prevent getting
diabetes, be more active, smoke less, and control cholesterol levels.
If this study is to be done again, I have two recommendations. First, I recommend
including the omitted variable, quality of healthcare. This would avoid the biased coefficient for
average age and make the equation more accurate. Second, spend more time finding data from
the same year. Even though a couple of years off should not make a huge difference, the
equation would be more accurate if all of the data is from the same year.
The data for lethal coronary artery disease was obtained from the National Center for Chronic
Disease Prevention and Health Promotion at:
The data for education was obtained from the National Census Bureau at:
The data for income was obtained from the National Census Bureau at:
The data for lack of health insurance was obtained from the United Health Foundation at:
The data for state health expenditures was obtained from the Milbank Memorial Fund at:
The data for alcohol consumption was obtained from the Brewers Association at:
The data for inactivity was obtained from the National Center for Chronic Disease Prevention
and Health Promotion at:
The data for state stress levels was obtained from the National Census Bureau at:
The data for tobacco consumption was obtained from the National Census Bureau at:
The data for average age of the population was obtained from the National Census Bureau at:
The data for diabetes was obtained from the National Center for Chronic Disease Prevention
and Health Promotion at:
The data for high blood cholesterol was obtained from the National Center for Chronic
Disease Prevention and Health Promotion at:
The data for high blood pressure was obtained from the National Center for Chronic Disease
Prevention and Health Promotion at:
The data for obesity was obtained from the American Obesity Association at:
The data for state populations was obtained from the National Census Bureau at:
A. H. Studenmund. “Using Econometrics: A Practical Guide.” 4th edition. Addison Wesley
Longman. 2001. April 18, 2005.
Anil Kumar. “Who Doesn’t Have Health Insurance and Why?” Federal Reserve Bank of Dallas.
2004. April 18, 2005.
“Blood Pressure.” American Heart Association. 2004. April 18, 2005.
“Causes of Heart Disease.” Texas Heart Institute. December 8, 2004. April 18, 2005.
“Coronary.” Mama’s Health. April 18, 2005. April 18, 2005.
“Diabetes: Type 2 Diabetes.” WebMD. June, 2004. April 18, 2005.
“Health and Wealth – A Fundamental Look.” Region of Peel. 2001. April 18, 2005.
“Heart Disease: Alcohol and Your Heart.” WebMD. June, 2004. April 18, 2005.
“Heart Disease: Exercise for a Healthy Heart.” WebMD. June, 2004. April 18, 2005.
“Heart Disease: Risk Factors For Heart Disease.” WebMD. June, 2004. April 18, 2005.
“Heart Disease: Smoking and Heart Disease.” WebMD. June, 2004. April 18, 2005.
“Obesity and Heart Disease.” American Heart Association. 1997. April 18, 2005.
“Risk factors for coronary heart disease and level of education.” The Tromso Heart Study. 1988.
April 18, 2005.
“What is Coronary Artery Disease?” National Heart, Lung, and Blood Institute. August, 2003.
April 18, 2005.