TRUST AND COVID VACCINES: DOES
TRUST IN GOVERNMENT AFFECT
COVID VACCINATION RATES AROUND
THE WORLD
Università degli Studi di Milano-Bicocca
Course of Econometrics
Prof. Luca Stanca
Capodivento Gaia, Riehn Lorenzo, Genoviva Valerio, Rupenovic Sara
April 2022
Contents
1. Introduction 2
2. Literature Review 3
3. Data 4
4. Econometric Analysis 9
4.1 Base Model 9
4.1 First Specifications 10
4.2 Ramsey´s RESET Test: Form Misspecifications 12
4.3 Turning points 15
4.4 Beta Coefficients 16
4.5 Homoscedasticity 17
4.6 Residual Analysis 18
5. Conclusion 19
References 21
Do Files: 21
1
1. Introduction
Covid 19 vaccines have proven not to be exclusively a “health care” matter. Instead, they are
somehow connected to politics, society, and personal beliefs in general. Besides its health
benefits, the vaccine was also intended to help jumpstart the economy after a worldwide
pandemic crisis that caused immense discontent among the world’s population. For this reason,
it is plausible to think that the relationship between people and their government can influence
people’s decision to take the vaccine.
Through a cross-country analysis, in this paper we try to analyze the effect of people’s trust in
public institutions on Covid-19 vaccine intake with a sample of 106 countries. We also try to
model the role of ideology in developing skeptical positions on vaccines, since studies showed
that ideological extremisms are more likely signs of some type of opposition toward vaccine
effectiveness. Our econometric model is created to investigate which factors might affect
vaccination rates, focusing mostly on the relationship between governmental trust and
vaccination rates all around the world and controlling for other important determinants. The
theory underlying our model is that countries with higher levels of government trust are
expected to be the ones associated with higher rates of fully vaccinated people.
However, we find that named effect is moderate compared to the impact of other variables.
Basic conditions in each country, such as income inequality, the age structure of the population
and the availability of vaccines also have a significant influence on vaccination rates. Trust in
science and the severity of the policy response to Covid-19 are significant as well.
In the sections below we will provide: a literature review of the past and current studies
concerning our topic; a detailed explanation of every single variable we believe to have a role
in explaining countries’ different vaccination rates; an econometric analysis in order to assess
the effect and the significance of each one of these variables. At the end, we will discuss the
results obtained, display the conclusions and give our last comments about the project.
2
2. Literature Review
The literature available on Covid-19 vaccination rates is being constantly updated since it is a
very recent topic of worldwide interest. Although there are many papers trying to explain the
possible determinants of differences in Covid-19 vaccine intakes between countries, very few
address the role of trust in public institutions and of political ideology in affecting these
differences.
As for developed countries mostly, a recent paper published by OECD (OECD, 2022) focused
on assessing discrepancies in vaccine coverage against COVID-19 between its members by
analyzing the key factors that influence vaccine intakes. The study found that variables such as
trust in government and in other public institution, demographic structure, and policy stances
towards vaccinating children have a role in explaining the differences in overall population
rates, even though the link is not always straight-forward. The implementation of mandatory
vaccination in some OECD countries could be a determinant, but the authors believe it is too
early to assess its effects.
This analysis concentrated in just the 38 countries members. For the sake of the project, we
decide to include more countries into our analysis, adding the smaller and less developed ones.
Moreover, we analyze other key factors that influence vaccine intakes such as, for instance, the
stringency index, the human development index and GDP per capita. As a result, we notice that
some variables or indicators are suitable and relevant to the purpose of the paper.
Another study, published by the academic journal “Policy Sciences” (Debus & Tosun, 2021)
analyzed the role of political views in explaining vaccination willingness. The study found that
ideological extremism on both ends of the spectrum explains skepticism of vaccination. The
authors concluded that policymakers should address this by creating broad alliances among
parties and societal groups in order to increase trust in and public support for vaccines against
COVID-19.
Further literature on the role of trust in the overall pandemic crisis was published by “The
Lancet” (COVID-19 National Preparedness Collaborators, 2022). The analysis found that trust
in government and interpersonal trust, as well as less government corruption, is significantly
3
associated with lower infection rates and with higher COVID-19 vaccine coverage among
middle-income and high-income countries, where vaccine availability was more widespread.
3. Data
Our analysis is based on cross-sectional data taken from different international sources (see
References). Every dataset we take into consideration contains information and indicators that
are useful to explain the concept of our project. The section below describes all the variables
and indices we use to understand the relationship between vaccination rates among countries
and the people’s trust in their government.
Our dependent variable from which our research takes shape, is the share of the fully
vaccinated population. We examine how the population for each country takes commitment
in completing all vaccination doses. The data we collected is from “Our World in Data” for the
date 06.04.2022 for all countries (Our World In Data, 2022). As a result, we observe the
different percentages comparing states among each other, which will be explained with later
variables.
The “Rule of Law” is one of the six worldwide governance indicators. As defined by the World
Bank, it is the extent to which people have confidence in and abide by the rules of the society,
and in particular quality of contract enforcements, property rights, the police, the courts, as well
as the likelihood of crime and violence (World Bank, 2021). We use this variable, among the
total two, to measure people’s trust in government.
The share of people who trust their national government in 2020 analyzes respondents who
answered, “a lot” or “some” to the question: “How much do you trust your national
government?”. This index, based on a large sample of surveys, helps us to understand the link
between people’s trust in national government and its influence on inducing every citizen to get
vaccinated (Our World In Data, 2022). The same survey has been done also to measure, for
each country, the share of people who trust in science, share of people who trust their
neighborhood, and the share of people who trust journalists.
4
“Total deaths per million” index displays cumulative confirmed Covid-19 deaths. The deaths
caused by the virus have been marked growing constantly in “Our World in Data” despite
improvements in medical technology and increase in vaccinations. Our intuition is that the
deadlier the virus hits the population the higher is the willingness to take a vaccine.
“Mandatory vaccines” expresses which countries have law enforced obligatory vaccination.
The metric used is (0) for no obligation in regard, while as (1) are rated the states that do.
European states are in majority for enforcing these laws: Italy for those over the age of 50 and
Greece over the age of 60, while Austria's new law mandates all adults to be vaccinated against
Covid. Such major obligations are also present in Asia and Latin America. Where in some
countries the vaccination is mandatory even for eligible minors, as for everyone above the age
of five in Ecuador (Statista, 2022).
The UNICEF Supply Division Collection collects information of total COVAX deliveries,
doses allocated, and doses ordered. We used its source to get data for our “Population share
covered by supply” index. This index shows that not all countries with lower percentage of
fully vaccinated population are such because of no trust in government, but that a possible cause
could be insufficient supply of vaccines (Unicef Supply Division, 2022).
The Covid Stringency Index is taken from “Our World in Data” website. To implement this
measure in our model, we considered the average stringency index between 2020 and 2021 for
each country. This index is based on different measures such as: school closures, workplace
5
closure and travel bans adopted to contrast further spread of the virus. We believe that political
decisions to limit the spread of the virus can affect people behavior in different ways, especially
if we talk about vaccination choices.
The Human Development Index (HDI) (Our World In Data, 2022) is a composite index
measuring average achievement in three basic dimensions of human development: a long and
healthy life; knowledge and education; a decent standard of living. We chose this index because
vaccination rates often move along with changes in overall wealth, life expectancy and
education.
“Age dependency ratio” is a ratio of the population of ages 65 and over to the population of
ages 15 to 64, expressed as the number of dependents per 100 people of working age (ages 15-
64) (United Nations Development Programme, n.d.). We believe that an on average old
population could be more willing to vaccinate against Covid, since age comes with greater
health risks.
Even GDP per Capita index could be associated to Covid-19 vaccine doses administrated to
each country’s population. “Our World in Data´s studies explain very well how rich countries
are more likely the ones with more vaccinated people; the graph below shows the positive
relationship between countries with higher GDP per capita and Covid-19 vaccination rates.
The Wellcome Global Monitor is the world's largest study into how people around the world
think and feel about science and major health challenges. It surveys over 140,000 people from
more than 140 countries. The index measuring the share of people that disagrees that vaccines
are effective analyzes individuals who responded, “strongly disagree” or “somewhat disagree”
to the statement: “Vaccine are effective”. We used this index to create our variable on the share
of people who disagree vaccines are effective.
Income Inequality is calculated through the Gini coefficient (Our World In Data, 2022), that
is a measure of the deviation of income distribution among individuals or households within a
country from a perfectly equal distribution. Since skepticism for covid vaccine might be high
in countries with high GDP as well as high income inequality, this nourishes the hate towards
6
the big pharmaceutical and rich companies, since the vaccines have been mostly developed by
privates.
The median age of each country’s population is another index we could use to represent the
age structure of each country (Our World In Data, 2022).
Lastly, we tried to capture the leading political ideology by looking at the political orientation
of the first party in that country (Scartascini, Cruz, & Keefer, 2020). In order to capture political
orientation in our model, we created two dummy variables, for each left- or right-wing
orientation. The right-wing [left-wing] dummy takes values:
- (1) if the country’s first party that is right-wing [left-wing] oriented,
- (0) if the country’s first party that is left-wing [right-wing] or doesn’t fit in neither of
these two categories.
Table 1. Summary Statistics
Variables Description Mean Std. dev. Min Max Obs
fvps Fully vaccinated population share 54,05538 25,09904 4,41 92,6 106
trustng Trust in the national government in this 54,22358 18,62367 15,8 94,5 106
country
rol Rule of law 0,114246 0,9853244 -2,35 2,078883 106
popcov Population share covered by supply 176,4156 161,927 9,9505 660,5293 101
deathpm Total deaths per million 1452554 1434070 1777 6366898 106
medianage Median age 33,2717 9,118007 16,4 48,2 106
gdppc GDP per capita 20284,1 15976,2 1697,707 67335,29 106
hdi Human development index 77,11132 13,57826 43,4 95,7 106
stringency Covid Stringency index 54,08836 11,10874 10,15419 74,57913 104
oldage Age dependency ratio 18,16541 10,82617 3,81877 48,00553 104
manvacc Mandatory vaccines 0,0660377 0,2495279 0 1 106
gini Income inequality 36,81887 7,49499 24,2 63 106
disagreff Share of people who disagree vaccines 5,837739 3,488062 0,0448544 18,2799 105
are effective
rightp Right wing party 0,2735849 0,4479162 0 1 106
leftp Left wing party 0,254717 0,4377719 0 1 106
7
trustsc Share of people who trust in science 77,23019 13,67667 37,5 97 106
trustneigh Share of people who trust 71,09245 14,43234 22,6 93,1 106
neighbourhood
trustjourn Share of people trust journalists 56,0217 14,1651 11,3 88,9 106
8
4. Econometric Analysis
4.1 Base Model
Previously we have shown that we aim at explaining the variance in Covid vaccination rates
worldwide, whereas we are mostly interested in the trust of the population in the national
government as a possible explanatory factor. As our first step we use the following OLS
multiple regression analysis that includes all our variables of interest and begins with a linear
functional form.
𝑓𝑣𝑝𝑠 = 𝛽0 + 𝛽1 𝑟𝑜𝑙 + 𝛽2 𝑝𝑜𝑝𝑐𝑜𝑣 + 𝛽3 𝑑𝑒𝑎𝑡ℎ𝑝𝑚 + 𝛽4 𝑔𝑑𝑝𝑝𝑐 + 𝛽5 𝑔𝑖𝑛𝑖 + 𝛽6 𝑑𝑖𝑠𝑎𝑔𝑟𝑒𝑓𝑓
+ 𝛽7 ℎ𝑑𝑖 + 𝛽8 𝑠𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑐𝑦 + 𝛽9 𝑜𝑙𝑑𝑎𝑔𝑒 + 𝛽10 𝑡𝑟𝑢𝑠𝑡𝑠𝑐 + 𝛽11 𝑡𝑟𝑢𝑠𝑡𝑛𝑒𝑖𝑔ℎ
+ 𝛽12 𝑡𝑟𝑢𝑠𝑡𝑗𝑜𝑢𝑟𝑛 + 𝛿1 𝑟𝑖𝑔ℎ𝑡𝑝 + 𝛿2 𝑙𝑒𝑓𝑡𝑝 + 𝛿3 𝑚𝑎𝑛𝑣𝑎𝑐𝑐 + 𝑢
Where:
- 𝛽0 is the intercept
- 𝛽1 , … , 𝛽12 are coefficients for discrete variables
- 𝛿1 , … , 𝛿3 are coefficients for the dummy variables
- 𝑢 is the error term
9
These are our first results:
. reg fvps rol popcov deathpm gdppc hdi stringency oldage manvacc gini disagreff rightp leftp trustsc trustneigh trustjourn
Source SS df MS Number of obs = 99
F(15, 83) = 12.56
Model 44859.6694 15 2990.64463 Prob > F = 0.0000
Residual 19767.1079 83 238.157927 R-squared = 0.6941
Adj R-squared = 0.6389
Total 64626.7774 98 659.456912 Root MSE = 15.432
fvps Coefficient Std. err. t P>|t| [95% conf. interval]
rol 3.516991 3.452439 1.02 0.311 -3.349771 10.38375
popcov .0225203 .0237229 0.95 0.345 -.0246635 .0697041
deathpm -2.37e-06 1.58e-06 -1.50 0.137 -5.51e-06 7.66e-07
gdppc -.0002639 .0002833 -0.93 0.354 -.0008275 .0002996
hdi 1.076148 .3658478 2.94 0.004 .348491 1.803804
stringency .4373343 .1696207 2.58 0.012 .0999655 .7747031
oldage .1372462 .3846359 0.36 0.722 -.6277791 .9022715
manvacc 8.162243 6.546462 1.25 0.216 -4.858407 21.18289
gini .5629992 .315877 1.78 0.078 -.0652675 1.191266
disagreff -.9668897 .5410657 -1.79 0.078 -2.043048 .1092683
rightp -.8458892 4.29383 -0.20 0.844 -9.386144 7.694366
leftp 1.627146 4.247382 0.38 0.703 -6.820726 10.07502
trustsc .6075361 .2534787 2.40 0.019 .1033771 1.111695
trustneigh -.3695144 .2435507 -1.52 0.133 -.853927 .1148981
trustjourn .0971167 .1810354 0.54 0.593 -.2629555 .4571888
_cons -92.95317 28.34145 -3.28 0.002 -149.3232 -36.58316
Here, the first thing we notice is that the direction of the correlation between rule of law and
Covid vaccination shares is positive as expected but evaluating the p-value this factor appears
to be insignificant at the 5% level in this first model. On the other side, many variables as the
human development index, stringency of restrictions, income inequality, doubts about vaccines
and trust in science seem to be highly significant at first glance.
4.1 First Specifications
The required assumptions for unbiased estimators consist of linearity in parameters, strict
exogeneity of regressors and no perfect multicollinearity. However, in this model we assume a
relatively high level of multicollinearity because we observe high correlations between several
explanatory variables. That can cause disruption in efficiency of the estimators and, in turn,
problems in the inference of the model.
Consequently, we implement a Variance Inflation Factor test, which indicates how strongly the
basic size of the standard error of a coefficient of a regressor is being inflated by correlation
with other regressors. In fact, we observe that the standard errors of hdi, oldage, trustneigh,
10
trustsc, gdppc, popcov and rol are inflated by more than twice their basic size. On that basis we
decide to eliminate hdi, trustneigh and gdppc from the model and to replace oldage and rol with
medianage and trustng, respectively.
. vif
Variable VIF 1/VIF
hdi 10.40 0.096131
gdppc 8.40 0.119002
oldage 7.33 0.136364
popcov 6.10 0.163826
trustneigh 5.40 0.185290
trustsc 5.20 0.192146
rol 4.83 0.207167
trustjourn 2.63 0.379597
gini 2.31 0.432209
deathpm 2.01 0.496762
rightp 1.55 0.643271
disagreff 1.47 0.680348
stringency 1.46 0.683727
leftp 1.42 0.706457
manvacc 1.17 0.854282
Mean VIF 4.11
. vif
Variable VIF 1/VIF
medianage 5.07 0.197056
trustsc 3.73 0.268246
popcov 3.28 0.304989
trustng 2.39 0.418064
trustjourn 2.31 0.432978
deathpm 2.12 0.471263
gini 1.72 0.581668
rightp 1.52 0.655855
leftp 1.36 0.735506
disagreff 1.34 0.744076
manvacc 1.16 0.863163
stringency 1.09 0.918775
Mean VIF 2.26
Now we evaluate the joint significance of the political orientation of the most voted party in
each country with a F-test. It does not result in a p-value low enough to reject the hypothesis of
joint insignificance of a leftwing or rightwing orientation of the most voted party at a 5% level.
Thus, it appears that ideological extremism in a country is a factor that affects vaccination rates
negatively (Debus & Tosun, 2021), while the political sphere alone has no significant impact.
. test leftp rightp
( 1) leftp = 0
( 2) rightp = 0
F( 2, 86) = 0.77
Prob > F = 0.4683
11
Finally, the variables trustjourn, deathpm and disagreff result with an p-value above the 5%
significance level. We decide to drop them.
4.2 Ramsey´s RESET Test: Form Misspecifications
Our resulting model is the following:
. reg fvps trustng medianage stringency manvacc gini trustsc popcov
Source SS df MS Number of obs = 99
F(7, 91) = 24.92
Model 42468.7882 7 6066.96975 Prob > F = 0.0000
Residual 22157.9891 91 243.494386 R-squared = 0.6571
Adj R-squared = 0.6308
Total 64626.7774 98 659.456912 Root MSE = 15.604
fvps Coefficient Std. err. t P>|t| [95% conf. interval]
trustng .2703415 .1027145 2.63 0.010 .0663117 .4743713
medianage 1.470627 .3448432 4.26 0.000 .7856385 2.155616
stringency .5378633 .146545 3.67 0.000 .2467697 .8289569
manvacc 8.52337 6.421327 1.33 0.188 -4.231808 21.27855
gini .8821418 .2559421 3.45 0.001 .3737443 1.390539
trustsc .334881 .1940913 1.73 0.088 -.0506576 .7204196
popcov .0241693 .0166789 1.45 0.151 -.0089612 .0572998
_cons -101.976 20.26271 -5.03 0.000 -142.2254 -61.72666
Next, we control for misspecifications in our model. An useful tool for that purpose is the
Ramsey Regression Specification Error Test (RESET), which investigates if a model requires
non-linear variables in order to explain optimally the dependent variable. The following low p-
value leads to a rejection of the null hypothesis that there are no misspecifications in our
regression. Therefore, we search for significant quadratic variables and interaction terms.
. estat ovtest
Ramsey RESET test for omitted variables
Omitted: Powers of fitted values of fvps
H0: Model has no omitted variables
F(3, 88) = 3.86
Prob > F = 0.0121
• Quadratic forms: Observing the graphical illustration of stringency, popcov, gini,
trustsc and we suspect increasing or diminishing effects for these variables. In fact, we
find that popcov, gini and trustsc are jointly significant with their respective quadratic
forms and that the latter add substantial explanatory power. We opt to integrate the
quadratic variables for popcov and gini into the model.
12
. test trustsc trustsc2
( 1) trustsc = 0
( 2) trustsc2 = 0
F( 2, 90) = 3.55
Prob > F = 0.0329
. test gini gini2
( 1) gini = 0
( 2) gini2 = 0
F( 2, 90) = 9.17
Prob > F = 0.0002
. test popcov popcov2
( 1) popcov = 0
( 2) popcov2 = 0
F( 2, 90) = 5.23
Prob > F = 0.0071
• Interaction terms: It seems reasonable to us that the positive effect of trust in science
decreases through augmenting age. Thus, we test additionally for a possible interaction
between trustsc and medianage and find a term that is significant on its own and jointly
with its basic variables as well. This interaction is implemented instead of a quadratic
form of trustsc.
. test medianage_trustsc medianage trustsc
( 1) medianage_trustsc = 0
( 2) medianage = 0
( 3) trustsc = 0
F( 3, 90) = 13.79
Prob > F = 0.0000
In order to investigate if countries that implemented laws to enforce vaccinations
experience different effects of the other regressors than countries without mandatory
vaccines we create interaction terms with the corresponding dummy and several
variables. However, none of these interactions result to be jointly significant. We decide
to eliminate the dummy for mandatory vaccines, since it is not significant at a level of
5%.
13
At this point we obtain our final model:
. reg fvps trustng medianage stringency gini trustsc popcov popcov2 gini2 medianage_trustsc
Source SS df MS Number of obs = 99
F(9, 89) = 24.88
Model 46244.6999 9 5138.29999 Prob > F = 0.0000
Residual 18382.0775 89 206.540196 R-squared = 0.7156
Adj R-squared = 0.6868
Total 64626.7774 98 659.456912 Root MSE = 14.372
fvps Coefficient Std. err. t P>|t| [95% conf. interval]
trustng .2651596 .0948913 2.79 0.006 .0766125 .4537066
medianage 3.802018 1.187526 3.20 0.002 1.44243 6.161606
stringency .46514 .1362266 3.41 0.001 .1944606 .7358193
gini 4.051901 1.431358 2.83 0.006 1.207824 6.895979
trustsc 1.050019 .4453433 2.36 0.021 .1651316 1.934907
popcov .1607867 .0436791 3.68 0.000 .0739974 .2475761
popcov2 -.0002235 .0000758 -2.95 0.004 -.0003742 -.0000728
gini2 -.0413465 .0174387 -2.37 0.020 -.0759968 -.0066963
medianage_trustsc -.033332 .0156701 -2.13 0.036 -.0644683 -.0021958
_cons -211.1392 40.19807 -5.25 0.000 -291.012 -131.2665
This model confirms the positive impact of our variable of interest, namely trust in the national
government, on vaccination rates worldwide. The model shows that for one percentage point
increase in the trust in government index, the fully vaccinated population share increases by
0.265 percentage points.
We can then address the impacts of a generic one unit change in the non-linear variables of our
model (gini, popcov, medianage, trustsc) by using average values.
Considering the model in its functional form:
14
We find the following partial derivatives:
We find that basic conditions of the respective nation as income inequality, the age structure of
the population and the basic availability of vaccines have positive effects (evaluated at their
average value) on vaccine rates. While the effect of trust in science evaluated at the average
median age is negative and close to 0.
Once more we use the RESET test in order to search for misspecifications in the model. This
time the p-value results under the 5% significance level, which implicates that our model is free
of misspecifications as omitted variables or functional errors. Therefore, we can assume that
the model is not affected by omitted variable bias.
. estat ovtest
Ramsey RESET test for omitted variables
Omitted: Powers of fitted values of fvps
H0: Model has no omitted variables
F(3, 86) = 2.67
Prob > F = 0.0524
4.3 Turning points
The next step is to compute turning points of our quadratic variables and to evaluate if these
points are still in a realistic value span. We use the following formula:
𝛽𝑖
𝑇𝑃 = −
2𝛽𝑗
15
where i represents the original variable and j its quadratic form.
As already proved, increasing popcov has a positive but diminishing impact on fvps. After a
coverage of 359% of the population with vaccine doses this effect becomes negative. That
seems reasonable, since exceeding the triple of doses required to vaccinate the whole population
represents a negative sign for vaccination willingness or logistical incompetence.
The variable gini results in a similar output. It’s positive but diminishing effect on fvps turns
negative after a gini-coefficient of 0,491. A possible interpretation is that growth of income
inequality from a low level relates to economic prosperity and its intuitive that economic well-
being is related to large vaccine rates. However, if inequality turns to extreme levels, most of
wealth is owned by a few and the rest of the population remains poor and unsatisfied which
affects vaccine rates negatively.
4.4 Beta Coefficients
Now that we have our final model, we can proceed computing beta coefficients. These allow
us to set the units of each regressor on the same level and to compare their impacts. In this
process the dependent and independent variables are standardized. The resulting Beta
coefficients tell us by how many standard deviations the dependent variable changes after a
change of one standard deviation of an independent variable. That is useful since medianage is
represented on another scale than the rest of the variables. To be more precise, the non-
standardized coefficients express the change of vaccination rates in percentage points after an
increase of one year in medianage and after a change of one percentage point in the rest of the
variables.
1
For a Gini coefficient that takes values from 0 to 1. It would be 49 if measured as in our dataset (where it takes
values from 0 to 100)
16
fvps Coefficient Std. err. t P>|t| Beta
trustng .2651596 .0948913 2.79 0.006 .1932764
medianage 3.802018 1.187526 3.20 0.002 1.363784
stringency .46514 .1362266 3.41 0.001 .2013213
gini 4.051901 1.431358 2.83 0.006 1.18446
trustsc 1.050019 .4453433 2.36 0.021 .5736751
popcov .1607867 .0436791 3.68 0.000 1.016524
popcov2 -.0002235 .0000758 -2.95 0.004 -.6739788
gini2 -.0413465 .0174387 -2.37 0.020 -.9799789
medianage_trustsc -.033332 .0156701 -2.13 0.036 -1.362822
_cons -211.1392 40.19807 -5.25 0.000 .
4.5 Homoscedasticity
Furthermore, we want to make sure that our estimates are homoscedastic, which means that the
variance of the residuals is constant and is not dependent from omitted variables.
Heteroscedasticity does not affect the unbiasedness and consistency of our estimators but would
cause the standard errors to be biased. That, in turn, hampers inference on the significance of
our model. Therefore, we utilize the Breusch-Pagan Test and control for a linear relationship
between squared residuals and our regressors. The resulting p-value does not allow the rejection
of the null-hypothesis that the model is homoscedastic.
. estat hettest
Breusch–Pagan/Cook–Weisberg test for heteroskedasticity
Assumption: Normal error terms
Variable: Fitted values of fvps
H0: Constant variance
chi2(1) = 0.12
Prob > chi2 = 0.7243
The White Test is a further test for homoskedasticity, which adds squares and cross-products
of all independent variables to the regression on the residuals. This test searches consequently
not only for heteroskedasticity, but for specification bias as well. As in case of our model the
White Test can be inefficient if there are too many used regressors. We apply the test anyway
and as expected we cannot reject the hypothesis of homoscedasticity at a 5% level.
17
. imtest, white
White's test
H0: Homoskedasticity
Ha: Unrestricted heteroskedasticity
chi2(51) = 66.95
Prob > chi2 = 0.0663
4.6 Residual Analysis
Finally, we are interested in the structure of the residuals. After having evidence through tests,
the following graph shows us as well that our model is homoscedastic.
As expected, the residuals result with a mean of near zero and a standard deviation of 13.7.
With other words, the predictions of our model for vaccine rates diverge on average by 13,7
percentage points from the expected error of zero. Assuming that our OLS estimators are
consistent, this is a problem that can be solved with a larger sample.
18
The two graphs below indicate that the distribution of the residuals is approximately normal
with a few exceptions on both extremes. Apparently there some more positive extreme errors
than negative ones. However, the approximate normality of the residuals confirms the validity
of inference through our model.
5. Conclusion
Our econometric model is created to investigate which factors might affect vaccination rates,
focusing mostly on the relationship between governmental trust and vaccination rates all
around the world and controlling for other important determinants.
This model confirms our assumption on the initially raised question: The positive impact of
our variable of interest, namely trust in the national government, on vaccination rates
worldwide. The model shows that for one percentage point increase in the trust in government
index, the fully vaccinated population share increases by 0.265 percentage points.
Moreover, we find that basic conditions of the respective nation as income inequality, the age
structure of the population and the basic availability of vaccines have positive effects
(evaluated at their average value) on vaccine rates.
The model shows additionally a positive relation between trust in science and the dependent
variable. However, this effect diminishes the older the population is.
We also observe a positive relation with the stringency index; in other words, countries with
higher restriction measures proved it encourages people to get vaccinated.
19
To make a concrete example, during the first wave and first measures implied to stop covid-
19 in Italy, one of the neighboring countries did not take such a severe initiative. Italy had one
of the highest stringency indexes, while the Croatian index was relatively low which in turn
resulted in low vaccination rates.
Gini index and covid vaccine supply proved positive to our theory given that both variables
have a positive impact on our dependent variable. However, these effects decrease and
become negative after a certain value of income inequality and coverage of the population
with vaccine doses.
On the other hand, we find that variables as mandatory vaccination laws, confirmed Covid
deaths, political orientation of the first party, the human development index, GDP per capita,
trust in journalism and trust in neighborhood are not significant in the context of our model.
Even though our research gives a clear overview on the subject, we need to take in
consideration that this is up to current times. It is still early to know the real implications of
trust in government on share of fully vaccinated people rates since the vaccine campaign is
still ongoing and not all the countries have yet had access to enough covid vaccine supply. For
this reason, it will be necessary to re-examine and carry out the analysis with updated data
once the pandemic crisis will be definitely over.
Overall, our results show how important it is to consider trust in society and public
institutions as an important factor to work on in order to build an healthy environment and a
well-functioning country, in each field.
20
References
COVID-19 National Preparedness Collaborators. (1. February 2022). Pandemic Pandemic
preparedness and COVID-19: an exploratory analysis of infection and fatality rates, and
contextual factors associated with preparedness in 177 countries,from Jan 1, 2020, to Sept
30, 2021. The Lancet.
Debus, M., & Tosun, J. (16. June 2021). Political ideology and vaccination willingness: implications for
policy design. Policy Sciences.
OECD. (2022). Understanding differences in vaccination uptake among OECD countries. Economics
Department Working Papers No. 1704.
Our World In Data. (2022). Our World In Data. Von https://ourworldindata.org/ abgerufen
Scartascini, C., Cruz, C., & Keefer, P. (2020). The Database of Political Institutions 2020 (DPI2020). Von
Inter-American Development bank: https://publications.iadb.org/en/database-political-
institutions-2020-dpi2020 abgerufen
Statista. (2022). Statista. Von https://www.statista.com/chart/25326/obligatory-vaccination-against-
covid-19/ abgerufen
Unicef Supply Division. (2022). Von https://www.unicef.org/supply/covid-19-vaccine-market-
dashboard abgerufen
United Nations Development Programme. (n.d.). Human Develpment Reports. Retrieved from
https://hdr.undp.org/en/data
World Bank. (2021). Rule of Law. Von https://databank.worldbank.org/databases/rule-of-law
abgerufen
Do Files:
// Base Model, Correlations and VIF Test
cor rol popcov deathpm gdppc hdi stringency oldage manvacc gini disagreff rightp leftp trustsc
trustneigh trustjourn trustng medianage
reg fvps rol popcov deathpm gdppc hdi stringency oldage manvacc gini disagreff rightp leftp trustsc
trustneigh trustjourn
vif
reg fvps trustng deathpm medianage stringency manvacc gini disagreff rightp leftp trustsc trustjourn
popcov
vif
21
// F-Test
reg fvps trustng deathpm medianage stringency manvacc gini disagreff rightp leftp trustsc trustjourn
popcov
test rightp leftp
// RESET Test and Specifications
reg fvps trustng deathpm medianage stringency manvacc gini disagreff trustsc trustjourn popcov
reg fvps trustng medianage stringency manvacc gini trustsc popcov
estat ovtest
reg fvps trustng medianage stringency manvacc gini trustsc popcov popcov2
test popcov popcov2
estat ovtest
reg fvps trustng medianage stringency manvacc gini trustsc popcov trustsc2
test trustsc trustsc2
estat ovtest
reg fvps trustng medianage stringency manvacc gini trustsc popcov gini2
test gini gini2
estat ovtest
reg fvps trustng medianage stringency manvacc gini trustsc popcov stringency2
test stringency stringency2
estat ovtest
reg fvps trustng medianage stringency manvacc gini trustsc popcov medianage_trustsc
test medianage_trustsc medianage trustsc
estat ovtest
22
// Test for interaction of manvacc
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2 manvacc_disagreff
test manvacc manvacc_disagreff
estat ovtest
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2 manvacc_gini
test manvacc manvacc_gini
estat ovtest
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2 manvacc_popcov
test manvacc manvacc_popcov
estat ovtest
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2 manvacc_rol
test manvacc manvacc_rol
estat ovtest
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2
manvacc_stringency
test manvacc manvacc_stringency
estat ovtest
reg fvps rol popsupply deathpm stringency manvacc gini disagreff trustsc popcov2 manvacc_trustsc
test manvacc manvacc_trustsc
estat ovtest
23
//second RESET test
reg fvps trustng medianage stringency gini trustsc popcov popcov2 gini2 medianage_trustsc
estat ovtest
// Beta Coefficients and Homoscedasticity
reg fvps trustng medianage stringency gini trustsc popcov popcov2 gini2 medianage_trustsc, beta
reg fvps trustng medianage stringency gini trustsc popcov popcov2 gini2 medianage_trustsc
estat hettest
imtest, white
// Residual Analysis
reg fvps trustng medianage stringency gini trustsc popcov popcov2 gini2 medianage_trustsc
predict yhat
predict uhat, res
rvfplot, yline(0)
qnorm uhat
histogram uhat, normal density
sum uhat
24
