Abstract

This study focuses on various policies that the US Federal government could enact to increase COVID-19 vaccination rates and stabilize the spread of COVID-19 with herd immunity. It compares a vaccine mandate enforced with a fine to various incentives such as tax rebates and sporting tickets. It also takes into account the accessibility of vaccination centers, measured by the proximity of vaccination centers from residencies, as well as combinations of these different policy attributes. The target population for this study are all Americans 18 years and older who have not received a dose of a COVID-19 vaccine. Below are lists of all possible values of each attribute included in the choice questions:

  • Type of incentives for compliance: cash, grocery store gift cards, internet rebates, sporting tickets
  • Value of incentives for compliance ranging from $0-$1000 in increasing increments
  • Value of fines for non-compliance ranging from $0-$1000 in increasing increments
  • Distance from vaccination centers ranging from 0 miles from residences - 10 miles from residences in increasing increments

By collecting choice information on these potential attributes, we hope to gain insight on what incentives and/or penalties are the most effective in getting non-vaccinated U.S. residents above the age of 18 vaccinated.

We found that respondents are more likely to take the vaccine with incentives, as we saw a larger probability of vaccination with incentives, and that respondents seem resistant to penalties, as penalties drastically decreased the probability of vaccination. We particularly saw an increase in vaccination probability with grocery store gift cards as incentives, and a slight increase in vaccination probability for vaccination centers within 1-3 miles of residences. However, our confidence in whether the rest of the population will have the same preferences as the respondents is limited by the high confidence interval and major variation within subgroups.

Introduction

Despite the rollout of COVID-19 vaccines, vaccine hesitancy prevented the United States from achieving herd immunity, the level of immunity that a population needs to control the spread of a virus. Below is a chart depicting the number of daily vaccinations over time as a percentage of the population, followed by a chart depicting the percentage of the American population that is vaccinated:

On September 9th, 2021, U.S. President Joe Biden announced vaccine mandates for 100 million Americans, including federal employees and private-sector employees part of firms with over 100 people. This has been met with praise from some experts who claim that forcing Americans to get vaccinated is crucial, criticism from other experts who claim that resistance to vaccine mandates will cause labor shortages and is not effective enough to vaccinate enough of the population [for herd immunity], and hostile resistance, particularly from conservatives, who attack the Constitutional provision of the mandates. View this article to read some of the reactions to Biden’s order.

Vaccine mandates have been touted for months since the COVID-19 vaccine has become available, but before they were enacted many state and local governments started giving out positive incentives such as money or commodities. All levels of government, from the federal government to local, enacted initiatives to increase access to the vaccine by fighting misinformation and setting up mobile vaccination sites.

Our study analyzes the population and the effectiveness of various policies to increase vaccination rates. Each policy in our study consists of different attributes, which vary across each policy. The goal is to optimize the attributes to get as many people vaccinated as possible. The attributes to be considered are:

Accessibility

Accessibility refers to the distance which a vaccination clinic would be placed from your home. It varies with 0 (door-door vaccinations) - 10 miles from your residence.

Incentives

Incentives refers to the type of commodity that would be given out at tax season for being fully vaccinated against COVID-19 by the tax return deadline. It varies by the following cash tax rebates and cash equivalents (gift cards):

  • Cash
  • Grocery stores
  • Internet providers
  • Professional sporting events

Incentive Value

Incentive value refers to the value of commodities that would be given out at tax season for being fully vaccinated against COVID-19 by the tax return deadline. It varies by increasing increments from $0 to $1000.

Penalty

Penalty refers to the penalty due at tax season for failing to get a being fully vaccinated against COVID-19 by the tax return deadline under a vaccine mandate. It varies by increasing increments from $0 to $1000.

Survey Design

In order to see what sort of policy would be most appealing to those who are currently unvaccinated, the survey is designed to only collect responses from those who would be able to choose to take the vaccine after seeing the policy options. This consists of adults 18 years or older who have not received any doses of a COVID-19 vaccine. Those under the age of 18 and who have already received a vaccine dose were excluded because they are unable to choose to take the vaccine. This survey is also restricted to United States residents only, as this policy would only be directed towards United States residents.

The survey also collects demographic information relating to age, gender identity, race, income, political affiliation, and geographic location to better inform which policies are most effective for which demographics. Educational material presented to the respondents include information on accessibility, incentives, incentive value, and mandate value, which make up the survey’s attributes.

  • Accessibility ranged from 0-10 miles for the distance a vaccination clinic would be placed from a respondent’s home. The accessibility range starts at 0 miles because it is in the most convenient spot possible for a respondent to receive a vaccine, and the range ends at 10 miles because most respondents should have a vaccination location outside of 10 miles from their home.
  • Incentive value ranged from $0-$1000 to see if there was a value below $1000 that would sway more people to receive the vaccine. This range stopped at $1000, as anything above that number would most likely be infeasible to implement into a policy.
  • Incentives were provided as cash or cash equivalents (gift cards) redeemable at grocery stores, internet providers, and professional sporting events. This range of incentives were chosen as they are items that are necessary and/or popular among the general public.
  • Penalties ranged from $0-$1000 to see if there was a value below $1000 that would sway more people to receive the vaccine. This range stopped at $1000, as anything above that number would most likely be infeasible to implement into a policy.

Each respondent receives 10 choice questions that have two options “I will take the vaccine under these conditions” (where the conditions are listed in the option) and “I will not take the vaccine under these conditions”. There is one alternative per choice question (the vaccine conditions) and a no-choice option, where respondents are informed that it means they would not take the vaccine under the conditions listed in the first alternative. An example figure of a random conjoint question can be seen below.

Data Analysis

Sample Description

The final pilot survey sample after cleaning consists of 107 respondents. The choice data sheet contains of 20 rows per respondent, denoting 2 alternatives per question (will you take the vaccine under this condition or not?) and 10 questions each. The following tables show a breakdown of some of our important survey respondents’ demographics.

Gender Number of Respondents
Male 52
Female 54
Prefer not to say 1

We had a pretty even split of gender with slightly more females.

Year of Birth Number of Respondents
1950-1959 7
1960-1969 8
1970-1979 5
1980-1989 27
1990-1999 36
2000-2002 11

Most of our respondents were born in the 1990s, and our survey respondents seemed to skew more towards a younger population of 20 and 30 year olds.

Ethnicity Number of Respondents
Asian 6
Black 9
Hispanic 6
Native American 3
White 80
Self-Identify 3

An overwhelming majority of our survey population was white. The 3 people who self-identified had written in their ethnicity as “TX” “American” and “North carolina”.

Income (in thousands of $) Number of Respondents
< 10 3
10 - 20 7
20 - 30 13
30 - 40 9
40 - 50 14
50 - 75 19
75 - 100 15
100 - 150 14
150 - 200 6
> 200 3
Prefer not to say 5

Our respondents came from all income backgrounds.

Political Orientation Number of Respondents
Strongly Conservative 8
Conservative 21
Leaning Conservative 7
Moderate 19
Leaning Liberal 5
Liberal 16
Strongly Liberal 8
Independent 19
Prefer not to say 4

There was a diverse distribution of political orientations for our respondents.

Education Received Number of Respondents
No High school 3
High School 18
Vocational 5
Some College 27
Associate’s Degree 11
Bachelor’s Degree 34
Graduate Degree 8
Prefer not to say 1

The majority of our survey population had some college education or more.

Resistance Number of Respondents
No Resistance 12
Mixed 47
Complete Resistance 48

From our data we identified different groups with different levels of resistance to the COVID-19 vaccine. The no resistance group consisted of 12 survey respondents who chose to take the vaccine for all 10 choice questions. The mixed resistance group consisted of 47 survey respondents who sometimes chose to take the vaccine and sometimes did not choose to take the vaccine. The complete resistance group consisted of 48 respondents who chose not to take the vaccine under any conditions.

Data Cleaning

When we first opened the responses from the three separate components, we noticed a lot of null values to numerous attributes, including session ids that seemed to be from unsuccessful attempts to view our survey, or from bots on the internet. To remove these values gracefully, we had to join these three components together by the session id, a unique identifier for each participant associated with the survey. To do so, we took out duplicate columns and removed null values for session. We also computed the time took to complete each survey component in unique columns. Then, we filtered out respondents that did not answer the choice questions and did not meet eligibility criteria. We only filtered out respondents based on if they didn’t answer any choice questions because no one only partially answered th choice questions, so we could filter out only those who didn’t answer any. There were 697 total respondents, but we trimmed it down to 107 because many did not meet the eligibility criteria and were not able to get to the choice questions. We also filtered out respondents who didn’t pass the attention check, but no one was filtered out in this process as every respondent passed the attention check.

Modeling

Our utility model for each policy attribute is as follows:

The first two beta values represent continuous attributes, and for every change in value and penalty, the utility for policy j will increase by the beta value for that attribute * the level of the attribute. The last 8 beta values are for categorical attributes that can either be 0 or 1, with 0 denoting the attribute is not present and 1 denoting the presence of an attribute. If present, the utility for policy j will increase by the beta value for that attribute. The first four of those 8 beta values represent categorical variables for distance to a vaccination center, in miles from respondents’ residences. The first accessibility coefficient represents door-door vaccination campaigns. The last four terms represents the type of incentive given out - cash, grocery store gift cards, internet rebates, or sporting event tickets.

Using the model and the choice data, we then ran the logitr model using preference space to estimate the beta coefficients. Running a summary of the results returned to us the estimated coefficients and the standard error.

coefficients st_errors
value 0.0005111 0.0001911
penalty -0.0004686 0.0002416
accessibility_1 0.0804890 0.1857847
accessibility_3 0.0735859 0.1837196
accessibility_10 0.0285740 0.1868493
incentive_grocery_store 0.5278555 0.1803210
incentive_internet 0.1210913 0.1912552
incentive_sport_tickets 0.0098003 0.1928311
outsideGood 0.9830817 0.1963739

We ran the mixed logit model on the data. We tried multiple combinations of attributes to vary with a normal distribution, and found that the optimal parameters were value, penalty, and every incentive type varying across a normal distribution. We also found that since the mixed logit model is not convex, using 50 multistarts optimizes the model. We tried multistarts ranging from 2 to 100, and found that the optimal model is found after 50. However, we found that even the optimal mixed logit model is not a good fit, as the expected value for each of the coefficients does not mirror the expected values of the coefficients when using a multinomial logit model.

coefficients st_errors
value_mu 0.0022262 0.0007514
penalty_mu -0.0008791 0.0021181
accessibility_1 0.5889784 0.6222866
accessibility_3 -0.5088493 0.7641872
accessibility_10 0.2904307 0.6313879
incentive_grocery_store_mu 15.3158601 13.0733657
incentive_internet_mu 0.0407058 0.5707278
incentive_sport_tickets_mu 3.2118784 3.2823391
outsideGood 1.1480483 0.5721541
value_sigma 0.0000108 0.0014957
penalty_sigma 0.0136262 0.0118530
incentive_grocery_store_sigma 43.4853586 36.9758657
incentive_internet_sigma 3.3756724 2.8526296
incentive_sport_tickets_sigma 12.8052311 12.9609789

We ran the logitr model with resistant respondents, those who chose not to take the vaccine under all conditions (denoted by outsideGood). We called this subgroup the “resistant” group. We obtained this group by marking each respondent with a 0 or 1 denoting if they answered “no” to all choice questions, and running the logitr model with this group using separate attributes.

coefficients st_errors
value 0.0007100 0.0002828
penalty -0.0004837 0.0003302
accessibility_1 0.1958957 0.2670157
accessibility_3 0.0587616 0.2653845
accessibility_10 0.1885783 0.2714119
incentive_grocery_store 0.5945227 0.2620557
incentive_internet 0.1087567 0.2655115
incentive_sport_tickets 0.2330551 0.2847370
outsideGood 0.4305177 0.2766491
outsideGood_resistant 0.9852725 0.4126940
value_resistant -0.0003627 0.0004121
penalty_resistant -0.0002322 0.0005347
accessibility_1_resistant -0.3975556 0.4015675
accessibility_3_resistant -0.0384331 0.3870407
accessibility_10_resistant -0.3220526 0.3963647
incentive_grocery_store_resistant -0.1738697 0.3832617
incentive_internet_resistant -0.1735595 0.4140298
incentive_sport_tickets_resistant -0.2535831 0.4106109

We ran the logitr model with conservative, liberal, and independent respondents, who self-identified with each respective political party. We obtained these groups by making dummy columns for each unique political orientation marking each respondent with a 0 or 1 denoting if they have a specific political orientation. We decided to narrow down the number of political groups to reflect our sample size by merging “strongly liberal” and “leaning liberal” into “liberal” and merging “strongly conservative” and “leaning conservative” into “conservative”. We then ran the logitr model on these groups and independents using unique attributes.

coefficients st_errors
value 0.0007494 0.0004551
penalty -0.0007930 0.0005237
accessibility_1 -0.3628354 0.3043856
accessibility_3 -0.5992260 0.3218997
accessibility_10 -0.3049389 0.3573930
incentive_grocery_store 0.8077133 0.3839628
incentive_internet -0.2547931 0.4246870
incentive_sport_tickets 0.3149509 0.4401965
outsideGood 0.4124942 0.3951194
value_liberal 0.0000072 0.0005852
penalty_liberal 0.0005274 0.0006997
accessibility_1_liberal 0.5646333 0.2949376
accessibility_3_liberal 0.8355463 0.2894474
accessibility_10_liberal 0.4735110 0.4454765
incentive_grocery_store_liberal -0.3286184 0.5213877
incentive_internet_liberal -0.0561608 0.5525032
incentive_sport_tickets_liberal -0.3111666 0.5638608
outsideGood_liberal -0.2461253 0.4944326
value_conservative 0.0002351 0.0006199
penalty_conservative 0.0004864 0.0007465
accessibility_1_conservative 0.9932580 0.5398139
accessibility_3_conservative 1.1545165 0.5475154
accessibility_10_conservative 0.5914997 0.6041454
incentive_grocery_store_conservative -1.1227352 0.5906551
incentive_internet_conservative 0.6716591 0.5821913
incentive_sport_tickets_conservative -0.8618877 0.6382369
outsideGood_conservative 2.0021718 0.6260375
value_independent -0.0006199 0.0006589
penalty_independent 0.0006692 0.0007991
accessibility_1_independent 0.3203842 0.5878214
accessibility_3_independent 0.9647243 0.5580106
accessibility_10_independent -0.1226066 0.6310279
incentive_grocery_store_independent -0.2292768 0.6019469
incentive_internet_independent 0.8813157 0.6527790
incentive_sport_tickets_independent 0.0599380 0.6483903
outsideGood_independent 1.0120944 0.6364934

We ran the logitr model with low-income and high-income respondents, who self-identified with each respective income group. We obtained these groups by making dummy columns for each income bracket and marking each respondent with a 0 or 1 denoting if they fall in a particular income group. We decided to narrow down the number of political groups to reflect our sample size by merging respondents who earned less than $50,000 into the low-income group and merging respondents who earned $100,000 or more into the high-income group. We then ran the logitr model on these groups using unique attributes.

x
value 0.0003445
penalty -0.0011474
accessibility_1 0.1270044
accessibility_3 0.1178104
accessibility_10 -0.3246856
incentive_grocery_store 0.4872539
incentive_internet 0.1613869
incentive_sport_tickets 0.0789400
outsideGood 0.8599015
value_low 0.0004641
penalty_low 0.0008499
accessibility_1_low -0.0020351
accessibility_3_low -0.0912626
accessibility_10_low 0.1989877
incentive_grocery_store_low -0.0453873
incentive_internet_low 0.1695055
incentive_sport_tickets_low -0.1603067
outsideGood_low 0.3508514
value_high 0.0000381
penalty_high 0.0016762
accessibility_1_high -0.1538972
accessibility_3_high 0.0323651
accessibility_10_high 1.1737559
incentive_grocery_store_high 0.2786345
incentive_internet_high -0.4680978
incentive_sport_tickets_high 0.0701226
outsideGood_high 0.0912208
x
value 0.0003284
penalty 0.0004240
accessibility_1 0.3110987
accessibility_3 0.3158577
accessibility_10 0.3188004
incentive_grocery_store 0.3073506
incentive_internet 0.3248201
incentive_sport_tickets 0.3210829
outsideGood 0.3192067
value_low 0.0004380
penalty_low 0.0005801
accessibility_1_low 0.4253876
accessibility_3_low 0.4260864
accessibility_10_low 0.4409104
incentive_grocery_store_low 0.4235103
incentive_internet_low 0.4422044
incentive_sport_tickets_low 0.4435092
outsideGood_low 0.4493175
value_high 0.0005521
penalty_high 0.0006664
accessibility_1_high 0.5100550
accessibility_3_high 0.5017537
accessibility_10_high 0.5074341
incentive_grocery_store_high 0.4880623
incentive_internet_high 0.5299878
incentive_sport_tickets_high 0.5312624
outsideGood_high 0.5413572

We ran the logitr model with Asian, Black, Hispanic, and Native American respondents, who self-identified with each respective race. We obtained these groups by making dummy columns for each race and marking each respondent with a 0 or 1 denoting if they fall in a particular ethnic group. Due to the fact that racial groups have unique biases, we could not merge racial groups and had to model Asians, Blacks, Hispanics, and Native Americans uniquely. We then ran the logitr model on these groups using unique attributes.

x
value 0.0005767
penalty -0.0003769
accessibility_1 0.1493537
accessibility_3 0.0995948
accessibility_10 0.0911452
incentive_grocery_store 0.2767176
incentive_internet 0.1446156
incentive_sport_tickets -0.0048925
outsideGood 1.1936386
value_asian -0.0004947
penalty_asian -0.0015811
accessibility_1_asian 0.1329034
accessibility_3_asian -1.1830281
accessibility_10_asian -1.0854955
incentive_grocery_store_asian 0.1798753
incentive_internet_asian -0.4854084
incentive_sport_tickets_asian -1.4625230
outsideGood_asian -3.3031581
value_black -0.0009031
penalty_black 0.0003545
accessibility_1_black -0.0174029
accessibility_3_black 0.1345240
accessibility_10_black 0.6291846
incentive_grocery_store_black 0.3805330
incentive_internet_black -0.6688335
incentive_sport_tickets_black 0.3029948
outsideGood_black -0.9511424
value_hispanic -0.0003316
penalty_hispanic -0.0003742
accessibility_1_hispanic -1.1800256
accessibility_3_hispanic 0.4838921
accessibility_10_hispanic -0.9817333
incentive_grocery_store_hispanic 1.7012276
incentive_internet_hispanic 2.1524001
incentive_sport_tickets_hispanic 2.2733010
outsideGood_hispanic 1.4130574
value_native 0.0010442
penalty_native -0.0015685
accessibility_1_native 0.2819474
accessibility_3_native 0.9975117
accessibility_10_native -0.0405203
incentive_grocery_store_native 2.0041884
incentive_internet_native 0.0333943
incentive_sport_tickets_native -0.8685169
outsideGood_native 0.0937451
x
value 0.0002245
penalty 0.0002816
accessibility_1 0.2193722
accessibility_3 0.2187689
accessibility_10 0.2200292
incentive_grocery_store 0.2149491
incentive_internet 0.2193150
incentive_sport_tickets 0.2229741
outsideGood 0.2317993
value_asian 0.0010520
penalty_asian 0.0013554
accessibility_1_asian 1.0890372
accessibility_3_asian 0.9494852
accessibility_10_asian 1.0216612
incentive_grocery_store_asian 1.1495492
incentive_internet_asian 1.2041302
incentive_sport_tickets_asian 1.1761442
outsideGood_asian 1.3543555
value_black 0.0006619
penalty_black 0.0009126
accessibility_1_black 0.6797840
accessibility_3_black 0.6524295
accessibility_10_black 0.6686678
incentive_grocery_store_black 0.6223760
incentive_internet_black 0.6939441
incentive_sport_tickets_black 0.6753842
outsideGood_black 0.6886401
value_hispanic 0.0011081
penalty_hispanic 0.0013126
accessibility_1_hispanic 1.0119626
accessibility_3_hispanic 0.8717907
accessibility_10_hispanic 1.0262839
incentive_grocery_store_hispanic 1.1957915
incentive_internet_hispanic 1.2885827
incentive_sport_tickets_hispanic 1.2897061
outsideGood_hispanic 1.2391928
value_native 0.0015067
penalty_native 0.0017405
accessibility_1_native 1.6091822
accessibility_3_native 1.3592319
accessibility_10_native 1.4966115
incentive_grocery_store_native 1.4367922
incentive_internet_native 1.2694133
incentive_sport_tickets_native 1.6821636
outsideGood_native 1.4531344

We ran the logitr model with poorly-educated and highly-educated respondents, who self-identified with each respective education level. We obtained these groups by making dummy columns for each education level and marking each respondent with a 0 or 1 denoting if they have earned a particular educational degree. We decided to narrow down the number of groups to reflect our sample size by merging respondents who did not hold a Bachelor’s degree into the low-education group and merging respondents who earned a graduate degree into the high-income group. We then ran the logitr model on these groups using unique attributes.

x
value 0.0003598
penalty -0.0006134
accessibility_1 0.0661708
accessibility_3 -0.2540634
accessibility_10 0.0879980
incentive_grocery_store 0.5380771
incentive_internet -0.2216536
incentive_sport_tickets 0.2010983
outsideGood 0.6532283
value_low 0.0002229
penalty_low 0.0001824
accessibility_1_low -0.0243030
accessibility_3_low 0.5894149
accessibility_10_low -0.1382827
incentive_grocery_store_low -0.0633475
incentive_internet_low 0.5188284
incentive_sport_tickets_low -0.2295321
outsideGood_low 0.6198316
value_high 0.0009449
penalty_high 0.0008621
accessibility_1_high 0.6248925
accessibility_3_high 0.5401381
accessibility_10_high 0.3625476
incentive_grocery_store_high 0.3688962
incentive_internet_high 0.9575404
incentive_sport_tickets_high -0.4836343
outsideGood_high 0.8185979
x
value 0.0003061
penalty 0.0004300
accessibility_1 0.2910949
accessibility_3 0.3101552
accessibility_10 0.2833100
incentive_grocery_store 0.2824072
incentive_internet 0.3118029
incentive_sport_tickets 0.3028663
outsideGood 0.3033357
value_low 0.0004043
penalty_low 0.0005356
accessibility_1_low 0.3956180
accessibility_3_low 0.3998634
accessibility_10_low 0.3944497
incentive_grocery_store_low 0.3812181
incentive_internet_low 0.4099136
incentive_sport_tickets_low 0.4066200
outsideGood_low 0.4137962
value_high 0.0008768
penalty_high 0.0009951
accessibility_1_high 0.7343250
accessibility_3_high 0.7455972
accessibility_10_high 0.7819365
incentive_grocery_store_high 0.7716949
incentive_internet_high 0.8025723
incentive_sport_tickets_high 0.8633954
outsideGood_high 0.8737230

Results

Baseline model

For the baseline model, we found that the general sample had a favorable view towards incentives (prefer to take the vaccine with incentives) as utility is positively correlated with incentive value, but an unfavorable view for fines (prefer not to take the vaccine with fines) as utility is negatively correlated with penalty value. Moreover, respondents prefer grocery store gift cards the most. We also saw that respondents in general preferred vaccination centers within 3 miles of residences over door-door vaccinations.

Resistant model

We examined the model attributes for resistant respondents, with non-resistant respondents as a reference. We found resistant respondents had an unfavorable view towards incentives as utility is negatively correlated with incentive value, and an unfavorable view for fines as utility is also negatively correlated with penalty value. However, resistant respondents seem to be slightly less sensitive to fines, but there is a high confidence interval for this attribute that spans 0, indicating that there is a high degree of uncertainty about the actual utility term for this attribute. Moreover, we found that resistant respondents. We also saw that respondents in general preferred door-door vaccinations and preferred cash as an incentive, however there is a high confidence interval for these attributes that spans 0, limiting our accuracy as to the actual coefficients.

Political model

We examined the model attributes for conservative, liberal, and independent respondents, using moderate respondents as a reference. We found respondents in political groups have a favorable view towards incentives as utility is positively correlated with incentive value, and an unfavorable view for fines as utility is also negatively correlated with penalty value. Liberal and conservative respondents did not seem to be persuaded by incentives, given that incentive value is not correlated with utility, and the coefficients for those groups is close to 0, but they had a more favorable view on penalties given the positive correlation between penalty value and utility. Independents seemed to have a more favorable view on penalties and a more unfavorable view on incentives, given the correlation between value and utility. Respondents of a political party (liberal, conservative, independent) tended to prefer vaccination centers, with a slight preference towards vaccination centers 3 miles or less from residences, given higher coefficients for those subgroups. Also given higher coefficients, conservative and independent respondents also prefer internet the most, while liberal respondents prefer cash more than other incentives. However, there is a high confidence interval that spans 0 across most subgroups and continuous attributes, indicating that there is massive uncertainty about the value of the coefficients.

Income model

We examined the model attributes for low-income and high-income respondents. We found that utility for low-income respondents with regards to penalties and incentives is positively correlated with both penalty value and incentive value, indicating that they are more favorable towards those policies. The expected coefficient for incentive value is close to 0 with high-income respondents, indicating that they are indifferent about incentives. However, the coefficient for penalties with high-income respondents is much higher, in the positive range, indicating more favorability towards penalties. Moreover, based on the coefficients, we found that low-income respondents in the sample slightly prefer internet rebates and vaccination centers within 10 miles, and that high-income respondents in the sample slightly prefer grocery store gift cards and vaccination centers within 10 miles. However, the high confidence interval in all attributes prevents us from making accurate guesses.

Ethnicity model

We examined the model attributes for Asian, Black, Hispanic, and Native respondents. Given the positive expected coefficients for Native respondents with regards to incentive value, and the negative coefficients for Black, Asian, and Hispanic respondents with regards to incentive value, we found that Native respondents have a favorable view to incentives, while Black, Asian, and Hispanic respondents have a more unfavorable view on incentives. We also found that Black respondents have a more favorable view on penalties, while Asian, Hispanic, and Native respondents have a more unfavorable view on penalties for non-compliance with vaccine mandates. We also found based on the categorical coefficients that Asian respondents prefer to travel within 1 mile to get a vaccine dose, Hispanic and Native respondents are willing to travel 3 miles, and Black respondents are willing to travel 10 miles. Finally, we found based on model coefficients that Asian respondents slightly preffered grocery store gift cards over cash, and majorly preferred grocery store gift cards over internet rebates and sporting event tickets, that Native respondents mostly prefer grocery store gift cards as incentives, that Hispanic respondents have a favorable view on non-cash incentives, and that Black respondents have a preference towards grocery store gift cards and sporting event tickets.

Education model

We examined the model attributes for less-educated and more-educated respondents. We found that utility for less-educated respondents with regards to penalties is positively correlated with penalty value, and that utility for less-educated respondents with regards to incentives is positively correlated with incentive value, although the coefficient for incentive value is close to 0. The expected coefficients for incentive value and penalty is postive with more-educated respondents, indicating that they are more likely to take the COVID-19 vaccine under incentives or penalties. Moreover, based on the coefficients, we found that less-educated respondents in the sample prefer internet rebates and vaccination centers within 3 miles of their residences, although some less-educated respondents are open to travelling 10 miles to get a vaccine, and that more-educated respondents in the sample slightly prefer internet rebates and vaccination centers within 3 miles. However, the high confidence interval that spans 0 in all attributes prevents us from making accurate guesses.

Market scenarios

We devised a total of 22 scenarios where we varied either incentive value or penalty with different accessibility levels (distances to vaccination centers) or incentive types, with door-to-door vaccinations and cash as an incentive being the reference values for accessibility and incentive type, respectively. Only value or penalty could be non-zero, exclusive to each scenario as we wanted to see the effects of each attribute individually. For each non-zero level of value or penalty, there were 7 possibilities, varying across each of the categorical attribute levels, including reference levels. Each scenario has two options, one with a combinations of choice attributes and outsideGood set to 0 denoting that a respondent would take a vaccine, and the other option with all chocie attributes set to 0 and outsideGood set to 1 denoting that a respondent would not take a vaccine in that scenario.

Continuous Parameters

For continuous attributes (value and penalty), we varied values between 0 (when the other attribute was in effect) and 1000. We decided to include 100 as a variable to take into account variation in coefficients.

value penalty
0 0
100 100
1000 1000

Categorical Parameters

For categorical attributes (incentive type and accessibility), we varied values between 0 and 1 using dummy columns, with 1 denoting that a certain policy involves an incentive type or accessibility level. Since each scenario is exclusive to value or penalty, for every scenario involving a penalty, we only plotted scenarios that varied by accessibility and kept all incentive type values to 0.

accessibility_1 accessibility_3 accessibility_10 incentive_grocery_store incentive_internet incentive_sport_tickets outsideGood
0 0 0 0 0 0 0
1 1 1 1 1 1 1

Plotting Scenarios

We used the predict function from the stats package to calculate the logit probabilities of each alternative in the scenario, simulating the probability that an arbitrary respondent would take the vaccine in a particular scenario.

We made a table of the predicted probabilities that a respondent would take a vaccine. We also made two tornado plots - one plot of the change in probability of vaccination with types of incentives with reference to cash, and another with probabilities of the accessibility levels with reference to door-door vaccinations. We also made sensitivity plots of each of the incentive types, including cash, with plotting scenarios with different values of each type of unique incentive individually against a baseline scenario (no incentives, vaccination centers, or penalties).

obsID predicted_prob predicted_prob_lower predicted_prob_upper altID value penalty accessibility_1 accessibility_3 accessibility_10 incentive_grocery_store incentive_internet incentive_sport_tickets outsideGood
1 1 0.4003244 0.3228230 0.4855994 1 100 0 0 0 0 1 0 0 0
3 2 0.3077035 0.2343901 0.3889756 1 100 0 0 0 0 0 1 0 0
5 3 0.2845162 0.2182217 0.3665695 1 100 0 0 0 0 0 0 1 0
7 4 0.2825254 0.2136240 0.3601909 1 100 0 0 0 0 0 0 0 0
9 5 0.2991224 0.2270157 0.3858334 1 100 0 1 0 0 0 0 0 0
11 6 0.2976772 0.2283098 0.3734484 1 100 0 0 1 0 0 0 0 0
13 7 0.2883533 0.2210155 0.3638758 1 100 0 0 0 1 0 0 0 0
15 8 0.5139713 0.4098747 0.6113384 1 1000 0 0 0 0 1 0 0 0
17 9 0.4131728 0.3170597 0.5211468 1 1000 0 0 0 0 0 1 0 0
19 10 0.3864746 0.2959924 0.4931808 1 1000 0 0 0 0 0 0 1 0
21 11 0.3841534 0.2877469 0.4850797 1 1000 0 0 0 0 0 0 0 0
23 12 0.4033642 0.2982846 0.5131289 1 1000 0 1 0 0 0 0 0 0
25 13 0.4017040 0.3079128 0.5175374 1 1000 0 0 1 0 0 0 0 0
27 14 0.3909354 0.2869989 0.4932312 1 1000 0 0 0 1 0 0 0 0
29 15 0.2630954 0.1959421 0.3421789 1 0 100 0 0 0 0 0 0 0
31 16 0.2789949 0.2095503 0.3661932 1 0 100 1 0 0 0 0 0 0
33 17 0.2776084 0.2113207 0.3493797 1 0 100 0 1 0 0 0 0 0
35 18 0.2686726 0.2022515 0.3413678 1 0 100 0 0 1 0 0 0 0
37 22 0.1897426 0.1223689 0.2932131 1 0 1000 0 0 0 0 0 0 0
39 23 0.2024269 0.1310650 0.3053517 1 0 1000 1 0 0 0 0 0 0
41 24 0.2013147 0.1336166 0.2982784 1 0 1000 0 1 0 0 0 0 0
43 25 0.1941745 0.1269072 0.2846540 1 0 1000 0 0 1 0 0 0 0

We found that increasing the value of incentives noticeably increased the value of the probability of vaccination, while increasing the penalties noticeably decreased the probability of vaccination. The probability of vaccination especially increased (10% higher) when the incentive was a grocery store gift card - implying that respondents were more inclined by grocery store incentives, and that respondents were hostile to penalties given the extremely low probabilities with high penalties. Moreover, we found that vaccination clinics within 1-3 miles of residences were most effective, as the probability of vaccination increased from door-door vaccinations, but decreased slightly with vaccination centers within 10 miles of residences.

Final Recommendations and Conclusions

In general, findings from our analysis were ultimately disappointing- our survey data found that although some people may respond to incentives and receive the vaccine, the vast majority of the unvaccinated population seems to be set on not receiving the vaccine. When asked for feedback on the survey, one respondent commented: “Under no circumstances including death will I ever take the covid 19 vaccine”. We classified 49 out of our 107 respondents as resistant, meaning that they answered that they would not take the COVID-19 vaccine under any of the incentives given to them in all 10 choice questions. To put into context how resistant this group is, some of our incentives and penalties valued up to $1000 and they still chose to not take the vaccine.

Despite all of this, we were able to gather some interesting findings. Out of all the incentives offered, our population had the largest preference for our grocery store incentive, even over cash. This finding was surprising, considering people would have had more ways to use the incentive if given cash instead of a grocery store incentive. When looking at the value of the incentive, the dollar amount seemed to be insignificant. This is because as the incentive value increased by hundreds of dollars, up to $1000, the value found diminishing returns. If an incentive were to be implemented, a standard value of $100 would allow policymakers to receive the most bang for their buck.

Our survey results also showed that having accessibility to the vaccine could act as a nice supplement to a grocery store incentive for a vaccine policy. A preferable accessibility distance to the vaccine for our survey respondents was within 3 miles of their home. Bringing in the accessibility within 1 mile only swayed a few more respondents, while not many were swayed by accessibility within 10 miles. Specific values can be seen in our accessibility tornado plot.

Based on the tornado plot of policy attributes, we recommend against penalties for non-compliance. Respondents became more and more averse to receiving a vaccine as the penalty value increased. Not only did penalties fail to convince respondents to take the vaccine, but it made them even more resistant. This resistance to compliance showed that a positive incentive would be much better received than a penalty.

Due to having a small sample size, our subgroups were fairly small and had a large confidence interval. This hindered our ability to guess the preferences of people in these subgroups, which hinders our confidence in making accurate predictions of optimal policies to increase vaccinations.

Limitations

There were many limitations to this survey, the largest being the sample size. With only 107 respondents, a much larger sample size would have allowed a much better confidence interval and standard deviation in all of our data. We would also be able to better see the preferences between the different demographics such as age, ethnicity, education, income, location, and political orientation. This could be useful in determining where to implement certain policies. For example, the sporting ticket incentive was least preferred when compared to cash, a grocery store incentive, and an internet incentive. However, in a state like Texas, where football is a religion and there are countless beloved professional and college sports teams, a subgroup analysis of Texans could find a huge preference for a sporting ticket incentive. Different groups of people will have different preferences and our analysis was only able to scratch the surface in discovering these. Gathering more survey data would be essential in having a more robust and confident analysis, which can inform policymakers to create the most effective policy.

Furthermore, with more time, the interactions between different variables could also be studied. As mentioned, the grocery store incentive was found to be more effective as an incentive for people to take the vaccine than cash. The relationship between the cash incentive and the grocery store incentive could be studied, along with other attributes, to be sure of any interaction effects. If this interaction were studied, it could be found that cash is just as effective as a grocery store incentive.

Appendix

Survey Link: https://covid-19-vaccine-policy-final.formr.org

Introduction

Welcome to our survey!

Welcome and thank you for your interest in our survey! The purpose of this survey is to develop a quantitative model of consumer choice to inform the design decisions of a COVID vaccine policy. This survey is being conducted for a research initiative at The George Washington University. You must be a resident of the United States above the age of 18 and who has not received a dose of a COVID-19 vaccine to partake in this survey. You must also not have a health condition or disability that prevents you from being eligible to take the COVID-19 vaccine.

This survey involves questions that ask whether a certain condition would result in you receiving the COVID-19 vaccine. These conditions can be seen as positive, negative, or neutral. These conditions will not actually occur and you will not need to take a COVID-19 vaccine based on your answers in this survey. At the end of the survey we will also be collecting demographic information.

If you have any questions about this survey, then please contact .

Education

Great work! Now that you’ve shared a bit about yourself, we’d like you to consider policies designed to encourage COVID-19 vaccinations. Let’s learn about policy attributes.

Policy Attributes

Below is a description of the various policy attributes that will vary across every policy - each policy will be a combination of a universal vaccine mandate, enforced by a fine at tax season for not complying, different types of monetary incentives given out at tax season for being fully vaccinated against COVID-19, and the administration of vaccination clinics. Each attribute contains an associated image helping to explain the attribute.

Accessibility

Accessibility refers to the distance in which a vaccination clinic would be placed from homes. It varies with increasing increments from 0-10 miles from a constituent’s residence. A distance of 0 miles indicates door-door in-residence vaccination campaigns, a distance greater than 0 miles indicates that the policy involves a mobile or indoor public vaccination clinic.

Incentives

Incentives refers to the type of commodity that would be given out at tax season as an incentive for being fully vaccinated against COVID-19 by the tax return deadline. It varies by cash (tax rebates) and cash equivalents (gift cards) redeemable at:

  • Grocery stores
  • Internet providers
  • Professional sporting events

Incentive Value

Incentive value refers to the monetary value of commodities that would be given out at tax season for being fully vaccinated against COVID-19 by the tax return deadline. It varies by increasing increments from $0 to $1000.

Penalty Value

This value refers to the penalty due at tax season for failing to get a being fully vaccinated against COVID-19 by the tax return deadline under a vaccine mandate. It varies by increasing increments from $0 to $1000.

Practice Question

We would like to begin with one practice question below. Your response to this question will not be recorded.

If these were the only options available, would you choose to take the vaccine under the policy below?

Mandate: Vaccine mandated, $100 fine for non-compliance Accessibility: Vaccination clinic will be located within 3 miles from your house Incentive: $50 in cash"

  • Yes
  • No

Conjoint Questions

Great work!

We will now begin with the questions. On the next few pages we will show you different combinations of policy options regarding COVID-19 vaccinations. There will be a total of 10 choice questions for you to answer. (Image shown below is an example of a choice question)

Respondent Information

Nice job!

We’re almost done! We’d just like to ask just a few more questions about you which we will use solely to analyze our survey data.

  1. In what year were you born?
  • All years from 1920-2003 are listed
  1. What is your current gender identity?
  • Male
  • Female
  • Trans male/trans man
  • Trans female/trans woman
  • Genderqueer/gender non-conforming
  • Prefer not to say

Different identity (please state):

  1. I identify my race as (select all that apply):
  • Native American or Alaskan Native (e.g., Navajo Nation, Blackfeet Tribe, Mayan, Aztec, Native Village of Barrow Inupiat Traditional Government, Nome Eskimo Community, etc.)
  • Asian (e.g., Chinese, Filipino, Indian, Vietnamese, Korean, Japanese, Malaysian, etc.) African (e.g., African American, Jamaican, Haitian, Nigerian, Ethiopian, Somalian, etc.)
  • Hispanic or Latino (e.g., Mexican or Mexican American, Puerto Rican, Cuban, Salvadoran, Dominican, Colombian, etc.)
  • Middle Eastern or North African (e.g., Lebanese, Iranian, Egyptian, Syrian, Moroccan, Algerian, etc.)
  • Native Hawaiian or Pacific Islander (e.g., Native Hawaiian, Samoan, Chamorro, Tongan, Fijian, Marshallese, etc.)
  • Caucasian (e.g., German, Irish, English, Italian, Polish, French, etc.) Prefer not to say

Different identity (please state):

  1. What is your annual household income (from all sources) before taxes and other deductions from pay?
  • Less than $10,000
  • $10,000 - $14,999
  • $15,000 - $24,999
  • $25,000 - $34,999
  • $35,000 - $49,999
  • $50,000 - $74,999
  • $75,000 - $99,999
  • $100,000 - $149,999
  • $150,000 - $199,999
  • $200,000 or more
  • Prefer not to say
  1. What is your political orientation?
  • Strongly conservative
  • Conservative
  • Leaning conservative
  • Moderate
  • Leaning liberal
  • Liberal
  • Strongly liberal
  • Independent
  • Prefer not to say
  1. What is your primary mode of transportation?
  • Personal vehicle
  • Public transportation (Bus, Rail, Ferry, etc.)
  • Ride-hailing service (e.g. Taxi, Lyft, Uber, Via, UberPool)
  • Para-transit services
  • Prefer to self-describe
  1. What state do you live in?
  • Alabama
  • Alaska
  • Arizona
  • Arkansas
  • California
  • Colorado
  • Connecticut
  • Delaware
  • District of Columbia
  • Florida
  • Georgia
  • Hawaii
  • Idaho
  • Illinois
  • Indiana
  • Iowa
  • Kansas
  • Kentucky
  • Louisiana
  • Maine
  • Maryland
  • Massachusetts
  • Michigan
  • Minnesota
  • Mississippi
  • Missouri
  • Montana
  • Nebraska
  • Nevada
  • New Hampshire
  • New Jersey
  • New Mexico
  • New York
  • North Carolina
  • North Dakota
  • Ohio
  • Oklahoma
  • Oregon
  • Pennsylvania
  • Rhode Island
  • South Carolina
  • South Dakota
  • Tennessee
  • Texas
  • Utah
  • Vermont
  • Virginia
  • Washington
  • West Virginia
  • Wisconsin
  • Wyoming
  1. What is the highest degree or level of school you have completed? If currently enrolled, please use the highest degree received.
  • Less than a high school diploma
  • High school degree or equivalent (e.g. GED)
  • Some college or university, no college degree
  • Trade/technical/vocational training, no degree awarded
  • Associate’s degree (e.g. AA, AS)
  • Bachelor’s degree (e.g. BA, BS)
  • Graduate or Professional Degree (e.g. PhD, MD, JD, MS)
  • Prefer not to say
  1. Do you have a religious exemption that prevents you from taking the COVID-19 vaccine?
  • Yes
  • No

We have reached the end. Please let us know if you have any other thoughts or feedback on this survey.

Your feedback will help us make future improvements :) Thank you so much for taking this survey.

(Open text response)