Page 149 - GHES-3-2

Global Health Economics and
Sustainability
Vaccine hesitancy in the US, India, and China

3.4.1.2. Ridge regression group. The income categories were redefined as follows:
Ridge assigns the L2 penalty, which is the squared magnitude “low-income group” for total household income <US$
of the overemphasized coefficients, with λ determining the 34,999, “middle-income group” for incomes between US$
weight assigned to the penalty. The larger the value of λ, the 34,999 and 74,999, and “high-income group” for those
more likely the coefficients approach zero, but the L2 penalty earning more than US$ 75,000. The education categories
does not help in finding the optimum estimates. Unlike were reclassified into: (i) “High school” (including less
LASSO, the Ridge model will not shrink these coefficients to than high school and some high school), (ii) “High school
precisely zero. The likelihood is represented by Equation VII: graduate” (high school graduate or equivalent), (iii) “some
college, no degree received,” (iv) “Associate/bachelor’s
l n i1 [ y x log1 e x i )] p j1 2 j (VII) degree,” and (v) “Graduate degree.” The responses related
R
(

to depression factors, such as frequency of anxiety, worry,
i

i
interest, and feeling down for over 2 weeks, were regrouped
Here, β is the L2 penalty. It uses the L2 penalty. The into three categories: “Not at all,” “Several days,” and
2
L2 penalty does not induce sparsity. With sparsity, one “Always,” which includes responses for “More than half the
can perform variable selection and also provide a level of days” and “Nearly every day.” Other forms of impairment,
interpretability of parameters. such as hearing, seeing, remembering, and mobility, were
also included and recategorized as “Impaired” (including
3.4.1.3. Elastic net
some difficulty, a lot of difficulty, and cannot do at all) and
Elastic net is a convex combination of LASSO and Ridge, “Not impaired” (no difficulty). In addition to the selected
with the effectual reduction in the effect of coefficients with factors for investigation in the MLR model to understand
L2 norm and exactly setting some coefficients to zero with their impact toward vaccine hesitancy, we also included
L1 norm. The likelihood is represented by Equation VIII: the following variables: race (White, Black, and others);
gender (male, female, and transgender); health insurance
2
p j1 1( ) p j1 | j | (VIII) status (yes, no); whether respondents received MH services
j
(yes, no); whether MH medicines (psychopharmacological
Here, λ is the penalty as a mixture of the previous drugs) were prescribed (yes, no); and region (Midwest,
two approaches. With the L1 norm, some regression Northeast, South, and West). Analysis indicated that
coefficients can be set to zero, thereby decreasing the individuals in the Midwest and West regions seemed
number of parameters, and hence, it is not surprising that it to be more vaccine-hesitant than those in the Northeast
can outperform LASSO on data with positively correlated or South. Furthermore, these regions were divided by
variables. topography into eight regions: West, Northwest, Midwest,
Southwest, Southeast, Mid-Atlantic, Northeast, and Great
Theoretically beneficial and computationally Lakes (Table 1). Additional subsets were identified among
advantageous penalized methods will be explored in future
publications but are described here.
Table 1. Regions of the United States
4. Results Region States
4.1. Pre-processing Northeast Connecticut, Rhode Island, Massachusetts, Vermont,
New Hampshire, Maine
Pre-processing relates to the HPD dataset. The responses
toward the intention of getting a vaccine were categorized Mid-Atlantic Virginia, West Virginia, Pennsylvania, New York,
New Jersey, Maryland, Delaware
into the following groups: (i) “Not hesitant” for those
who definitely plan to get vaccinated. (ii) “Unsure” for Southeast Georgia, Florida, South Carolina, North Carolina,
Mississippi, Alabama, Tennessee, Kentucky
those who are likely to get vaccinated but have some Great Lakes Minnesota, Wisconsin, Illinois, Indiana, Ohio, Michigan
uncertainty. (iii) “Hesitant” for those who probably will
not get vaccinated. (iv) “Strongly hesitant” for those who Midwest Iowa, Nebraska, Kansas, Missouri
definitely will not get vaccinated. This study aims to gain Southwest Arizona, New Mexico, Texas, Oklahoma, Louisiana,
an in-depth understanding of the factors contributing to Arkansas
vaccine hesitancy. The sample size is adequate to draw Northwest Alaska, Idaho, Montana, North Dakota, South
inferences related to a broader population. Dakota, Wyoming, Utah, Colorado
West Washington, Oregon, California, Nevada, Hawaii
For the application of MLR analysis, we combined the
“Hesitant” and “Strongly hesitant” categories into a single Mid-South Louisiana, Mississippi, Alabama, Arkansas,
Tennessee, Kentucky

Volume 3 Issue 2 (2025) 141 https://doi.org/10.36922/ghes.2958