Influenza risk perception and travel-related health protection behavior in the US: Insights for the aftermath of the COVID-19 outbreak.

Influenza is a contagious virus affecting both one's health and economic productivity. This study evaluates uses a survey of 2168 individuals across the U.S. Ordered logit regressions are used to model risk perception and generalized ordered logit regressions are used to model risk mitigation travel-related decisions. Models are estimated for three influenza outbreak scenarios, specifically an individual's travel-related: 1) risk perceptions, 2) risk mitigation decisions when infected and the individual wants to prevent spreading it, and may want treatment, and 3) risk mitigation decisions when not infected and the individual wants to reduce exposure. Risk perception results show that a recent personal experience with influenza-like symptoms and being female significantly increased risk perception at mandatory and medical trip locations. Risk mitigation model results show that males are less likely to alter their travel patterns in response to the possible spreading of the virus or increasing exposure. Knowing the difference between influenza and the stomach flu is more influential in reducing travel than a recent influenza experience in one's household. Individuals proactive with their health (i.e., receive the vaccine, have health insurance) are also proactive in seeking medical attention and reducing influenza spread. Lastly, aligned with the Protection Motivation Theory, individuals reduce travel to locations in which they perceived medium or high risk. However, increased risk perceived at one's work location did not significantly reduce travel. The findings provide insight into the risk perception and mitigation behavior of the American public during the COVID-19 pandemic and after restrictions are lifted.

Introduction

Within the United States, influenza affects an estimated 25–50 million people per year, with 150,000 hospitalizations and 30,000–40,000 deaths (Rauš et al., 2015), resulting in an annual economic impact in the billions of dollars (Schoenbaum, 1987). Three types of influenza affect human health: seasonal, Zoonotic or variant, and pandemic influenza (WHO, 2014). Seasonal influenza can affect individuals multiple times in their lifetimes and is predominant during winter months, in regions with a temperate climate. Zoonotic or variant forms of influenza are spread by the interaction between humans and animals and have been recorded several times in the past century (WHO, 2014). Whereas, a virus is pandemic when it has not previously circulated amongst humans and creates significant epidemics, for example, the H1N1 influenza (2009) pandemic (WHO, 2014) and the Coronavirus disease (COVID-19) pandemic (Chappell, 2020). Pandemics can lead to much higher death tolls worldwide if not contained quickly through social distancing measures and vaccines.

Influenza spreads from person to person through sneezing, coughing (Stelzer-Braid et al., 2009), or touching contaminated surfaces (Jones, 2011). COVID-19 can live on surfaces for several days (Gray, 2020) and be airborne for up to three hours (Harvard Medical School, 2020). Individuals can be exposed to the virus at a variety of locations based on their daily activities, such as work, school, restaurants, libraries, hospitals, and doctor's offices. Seasonal influenza vaccinations are developed based on predictions of which virus strains will be prevalent in a given year, but the effectiveness varies, meaning those who are vaccinated can become ill after exposure to a strain that is not well matched by the vaccine or if the individuals are unable to develop a protective immune response.

Chances of contracting or spreading the infection can be reduced if a person alters their behavior effectively when they perceive high influenza risk during their daily activities. For the purposes of this paper, risk is a combination of consequences and likelihood. The consequences are contracting influenza and the severity of the symptoms, which are a function of individual health and mitigation measures. The likelihood can be considered a function of exposure and susceptibility. The influenza virus is a threat to an individual's health if he/she is not immune.

Previous studies have linked the preventative health measures an individual takes with the risk perceived (Marathe et al., 2011; Chen et al., 2010; Chai et al., 2010). For example, the Health Belief Model, a theoretical social cognition model in psychology, shows that an individual takes action to reduce risk after considering the related benefits and costs of that action (Rosenstock et al., 1988; Conner and Norman, 2005). Later, Rogers, 1975, Rogers, 1983 hypothesized the Protection Motivation Theory (PMT), which expands on the Health Belief Model by emphasizing two cognitive processes: threat appraisal and coping appraisal.

Threat appraisal is related to the perception of the severity of the event (related to how wide spread the virus is, potential exposure, and consequence in our risk terminology) and maladaptive response reward. Maladaptive response reward is the perceived benefit gained by not engaging in health protection behavior to avoid getting sick (Rogers, 1983; Ruthig, 2016). In other words, threat appraisal is the estimation of the chance of contracting the disease and estimation of the seriousness of a disease (i.e., severity) (Rogers, 1975, 1983; Boer and Seydel, 1996). Coping appraisal looks at the “individual's expectancy that carrying out recommendations can remove the threat (response efficacy) and the belief in one's ability to execute recommended courses of action successfully (self-efficacy)” (Boer and Seydel, 1996).

After perceiving a high health risk, the protective actions an individual chooses are dependent on the mitigation strategies available to the individual. Based on response efficacy and self-efficacy, an individual can either adopt or reject a protective action. For example, social distancing such as avoiding public places can reduce exposure to the virus, reducing the likelihood of contracting influenza, while vaccination can reduce susceptibility to virus strains included in the vaccine. Visiting the doctor while ill can lead to treatments to reduce the severity of the symptoms. All risk mitigation actions investigated in this paper either modify the exposure component of risk for the individual or others or the severity component of the consequences.

Several past studies focused on connecting risk perception and risk mitigation. In a study on the Australian population, Barr et al. (2008) found that respondents with higher levels of health-risk perception reported more willingness to comply with public health interventions during influenza outbreaks. Similar results were found in Hong Kong (Leung et al., 2003), Italy (Di Giuseppe et al., 2008), and Australia (Jacobs et al., 2010), where respondents with increased perception of risk were more likely to engage in risk-reducing behaviors.

The purpose of this study is to evaluate three scenarios using data from a survey of 2168 individuals across the U.S. Ordered logit regressions are used to model risk perception and generalized ordered logit regressions are used for risk mitigation travel-related decisions. Three scenarios are tested with respect to an influenza outbreak, specifically an individual's travel-related: 1) risk perceptions, 2) risk mitigation decisions when infected, the individual wants to prevent spreading it, and may want treatment, and 3) risk mitigation decisions when not infected and the individual wants to reduce exposure. This study adds to the existing literature through these scenarios and the specific locations considered (i.e., rather than social-distancing in general). Scenario 1 includes the locations for work, school, restaurants, libraries, doctor's offices, and hospitals. Understanding the risk perceptions of contracting influenza at these locations may provide insights for how the public will view the risk of contracting other viruses (e.g., COVID-19) when travel restrictions are lifted but before vaccines are available. Scenario 2's risk mitigation models pertain to visiting doctors' offices, avoiding public places, avoiding public transit, and staying home. The factors identified in these models may be helpful for researchers analyzing disease spread. Scenario 3's risk mitigation models pertain to the last three locations in Scenario 2 and the factors associated with these models may indicate which individuals are more likely to voluntarily social-distance even when not requested by governing entities.

The remainder of this paper is divided into five sections. Section 2 discusses the data and methodology of this study. Section 3 summarizes the results of the risk perception scenario. Section 4 contains the results from the two risk mitigation scenarios. Section 5 discusses the policy implications of the study's findings and Section 6 provides conclusions and recommendations for future research.

Data and methodology

The data for this study came from a survey that was developed by researchers at Virginia Tech. It was distributed by the GfK Group (formerly Knowledge Networks) using a sample from GfK's KnowledgePanel®, a probability-based web panel designed to be representative of the United States. GfK recruits panel members by using address-based sampling methods. Once household members are recruited for the panel and assigned to a study sample, they are notified by email for survey taking and panelists can visit their online member page. The sample for this study consisted of participants who spoke English, were at least 18 years old, and resided in the United States. From the 3604 fielded participants in March 2016, a total of 2168 participants (close to 60%) completed the survey. The majority of the respondents were based in California (11.2%), followed by Texas (7.3%), Florida (6.7%) and New York (5.9%).

Survey participants were provided the following text prompt for the symptoms of influenza: “The purpose of this survey is to learn about your experience with influenza. Influenza is an illness that has symptoms such as: fever, cough, sore throat, muscle aches, runny or stuffy nose, headaches, and fatigue. Influenza what people refer to as the stomach flu. When completing this survey, influenza and not the stomach flu as you answer the questions.”

Table 1 outlines the information collected from respondents that were used to build the models for the three scenarios, including both the independent and dependent variables. Scenario 1 contains models that estimate factors that influence a person's risk perception at work, school, restaurant, library, hospital, and doctor office locations. Scenario 2 contains travel-related behaviors for four locations when a person is sick with influenza and possibly trying to prevent spreading the virus or to seek medical attention. Scenario 3 contains three models for when someone is not sick, but possibly wants to reduce exposure.

Table 1

Variables in the ordered logit models.

Variable name	Parameter definition
Scenario 1: risk perception of contracting influenza
Doctor's office risk rating	Three-factor response scale: low, medium, high
Hospital's risk rating
Library risk rating
Restaurants risk rating
School risk rating
Work risk rating
Scenario 2: actions when sick with influenza
Visit doctor's office	Three-factor response scale: never, sometimes, always
Avoid public places
Avoid public transit
Stay at home
Scenario 3: actions to reduce exposure
Avoid public places	Three-factor response scale: never, sometimes, always
Avoid public transit
Stay at home
Independent variables
Age	Respondent's age in years
Apartment/mobile	Respondent lives in either an apartment or mobile home (0 = No, 1 = Yes).Other responses for detached family home, attached family home, boat/RV/van.
BS degree	Respondent has at least a bachelors' degree (0 = No, 1 = Yes)
Experience	Respondent or HH member experienced influenza-like symptoms in the last 6 months (0 = No, 1 = Yes)
Exposure_Work	Respondent's job involves “a lot” of contact with the public (1 = yes, 0 = no)
Health insurance	Respondent has health insurance (0 = No, 1 = Yes)
HHsize	Household size
Income	HH income in ten-thousand dollars. Pseudo continuous income based on range mid-points
Income/HHSize	HH income in ten-thousand dollars divided by the number of members in the household
InfoSourceWordofMouth	Respondent received influenza outbreak information by word of mouth (1 = yes, 0 = no)
Knowledge	Respondent knows the difference between ‘stomach flu’ and influenza (0 = No, 1 = Yes)
Kids_0_12	At least one child in the HH between the ages of 0 and 12 years old (0 = No, 1 = Yes)
Male	Gender (0 = Female, 1 = Male)
Metro	Respondent lives in a metropolitan statistical area (MSA)
Race White (non-Hispanic)	Race and ethnicity of the participant (0 = Not Caucasian 1 = Caucasian).Not Caucasian includes responses for Black (non-Hispanic), Other races (non-Hispanic or 2+ races Hispanic/non-Hispanic).
Single	Respondent has a marital status of divorced, never married, or widowed (0 = No, 1 = Yes)
Single parent	Single parent (1 = yes, 0 = no – both dual parent households and no children in the household)
Stores risk rating	Medium or High level of risk perception for store visit (0 = No, 1 = Yes)
Transit risk rating	Medium or High level of risk perception for transit systems (0 = No, 1 = Yes)
Vaccine	Participant gets the influenza vaccine every year or some years (0 = No, 1 = Yes)
Work risk rating	Medium or High level of risk perception for the workplace (0 = No, 1 = Yes)
Working	Respondent is employed (0 = No, 1 = Yes)

Additional questions regarding each respondent's demographics and home provided information that could be used to predict their risk perceptions and their risk mitigation actions in the three scenarios. Respondents selected their household income range and, for this study, the income ranges were converted to a continuous format by averaging the minimum and maximum bounds of each range. The housing type (apartment/mobile) and race/ethnicity (race white, non-Hispanic) variables, respectively, were grouped to increase variable significance in the models. For example, with the race/ethnicity variable, all combinations were tested, such as Black/non-Hispanic and White/non-Hispanic vs Any/Hispanic and Other/non-Hispanic, Any/Hispanic and Black/non-Hispanic vs White/non-Hispanic and Other/non-Hispanic, Black/non-Hispanic vs all other races, White/non-Hispanic vs all other races, etc. Other race and housing type groupings reduced variable significance and, at times, the reduction resulted in variable insignificance. When significant, the risk perception ratings were included as independent variables in the Scenario 2 and 3 models.

The dependent variables for all models were indicated on three-factor scales (low, medium, high or never, sometimes, always). These responses have an intuitive order but no mathematical equivalent representation. An ordered logit model is most commonly used to account for this property in the dependent variable, and the ordered logit was used for Scenario 1. However, the difference in the transition from never to sometimes and the difference from sometimes to always is not always equal. Many variables in the Scenario 2 and 3 models had relationships that violated the parallel regression assumption of the ordered logit. This was indicated through the results of the Brant Test, which significantly rejected the null hypothesis that the parallel regression assumption holds (Williams, 2006). This means there was not a single uniform trend across the three responses. For example, in this study, a large number of respondents with past influenza experience answered ‘sometimes’ to avoiding public places when sick while very few answered ‘never’ or ‘always.’ This relationship violates the parallel regression assumption. The ordered logit cannot be used because a single coefficient cannot pick up this changing effect across the three ordered responses. Therefore, the generalized ordered logit used in Scenarios 2 and 3 gives more than one coefficient to parameters violating the assumption to overcome the ordered logit's shortcomings (Williams, 2006).

The generalized ordered logit model has a similar probability equation to the ordered logit. However, since the generalized ordered logit model gives one coefficient to parameters meeting the parallel regression assumption and more than one coefficient to those violating the assumption, the equation is adjusted to meet this specification. Eq. (1) gives the generalized logit probability equation (for individual i selecting an option with a coding greater than j) for a model where both X and X meet the parallel regression assumption and therefore β and β can be used across all j's. The X variable did not meet the assumption, therefore β varies across the j alternatives. Eqs. (2), (3), (4) are used in conjunction with Eq. (1) to get the separate probabilities of individual i choosing each of the M ordinal responses (Williams, 2006).

Risk perception results

Models for people's risk perceptions were developed for select locations, including mandatory trips (i.e., work and school), discretionary trips (i.e., restaurants and libraries), and medical trips (i.e., hospitals and doctor's offices). The ordered logit methodology was used as all models passed the parallel regression assumption. Table 2 presents the six models, where the work model was estimated only on responses from employed participants and the school model on responses from households that had children 12 years old and younger. The McFadden pseudo-R2 was used to indicate the model goodness-of-fit, which typically has low values and should not be interpreted as a linear regression R-squared value. Previous researchers (Norušis, 2012; Grilli and Rampichini, 2014) using the ordered logistic regression have reported values from 0.012 to 0.075.

Table 2

Scenario 1 travel-related risk perceptions for contracting influenza.

	Mandatory		Discretionary		Medical
	Work	School	Restaurant	Library	Hospitals	Doctor
Male	−0.342⁎⁎⁎	−0.582⁎⁎⁎				−0.224⁎⁎⁎
Experience	0.971⁎⁎⁎	0.741⁎⁎⁎	0.551⁎⁎⁎	0.486⁎⁎⁎	0.505⁎⁎⁎	0.465⁎⁎⁎
Exposure_Work	1.796⁎⁎⁎
SingleParent	0.594⁎⁎	−0.830⁎⁎
Rent	−0.303⁎⁎
Income			−0.0240⁎⁎⁎	−0.0172⁎⁎
Metro				0.325⁎⁎⁎
Kids_0_12					0.190⁎	0.195⁎
Knowledge					0.227⁎⁎
InfoSourceWordOfMouth	0.429⁎⁎⁎	0.528⁎⁎			0.176⁎⁎	0.263⁎⁎⁎
InfoSourceTV			0.270⁎⁎⁎	0.219⁎⁎		0.302⁎⁎⁎
Constant cut 1	−0.147	−1.688⁎⁎⁎	−1.177⁎⁎⁎	−0.198	−1.081⁎⁎⁎	−1.394⁎⁎⁎
Constant cut 2	2.395⁎⁎⁎	0.196	1.329⁎⁎⁎	1.854⁎⁎⁎	0.488⁎⁎⁎	0.415⁎⁎⁎
Observations	1144	358	2024	1964	2007	2023
AIC	2112.683	696.1333	4018.824	4077.401	4095.517	4009.812
BIC	2153.022	719.416	4046.888	4110.897	4129.143	4049.099
McFadden adj. R-squared	0.138	0.050	0.012	0.010	0.011	0.016

Note: Significance. Codes: ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1.

As indicated in Table 2, males significantly perceived lower risk at work, school and the doctor's office. This finding is consistent with literature where, in a review of studies related to gender and risk perception, Gustafsod found that while men and women are concerned about the same things, women frequently worry more (Gustafsod, 1998). Davidson and Freudenburg suggested that the traditional gender role of women being nurturers and care providers within a family could lead to higher health risk perceptions (Davidson and Freudenburg, 1996).

If the participant or someone in their household had been infected with influenza in the past 6 months, that person had a heightened risk perception at all of the locations with work having the greatest increase. This is similar to the findings in Leung et al. (2003), where in the context of severe acute respiratory syndrome (SARS), those with relevant symptoms were more likely to take precautionary measures.

Employees of jobs that involve close contact with the public viewed their workplaces as having higher risk. This type of employment allows greater opportunities for exposure to viruses through the exhalation of infected individuals (Fabian et al., 2008) as well as coughing and sneezing. Renters perceived their workplaces to have lower risk on average. Single parent households on average saw the workplace as having a higher risk of contracting influenza compared to non-single parent households. However, on average, they viewed the school environment to have lower risk than their counterparts.

Household income was indirectly related to perceived risk for discretionary trips. This indirect relationship is consistent with other literature that found that higher income individuals/households were more capable of making purchases to reduce their risk (Dosman et al., 2001). Those living in a metropolitan area perceived higher risk at the library compared to those not living in a metropolitan area. Those with children 12 years old and younger did not have significantly different risk perceptions for work and discretionary trip locations, but did view medical locations as having higher risk. This may be due to children having immature immune systems compared to adults, where immunity memory is established during the first 18 years of a person's life (Dietert et al., 2000). Also, having knowledge of the difference between influenza and the stomach flu increased risk perception at hospitals.

The source and type of information on influenza spread affects how people perceive health risks (Suciu, 2019). Survey responses could indicate more than one information source, so they were not mutually exclusive. Finding out about an influenza outbreak by word of mouth increased risk perception for mandatory and medical trips. However, influenza outbreak information disseminated through television had a greater impact on risk perception for discretionary trips than for the other two trip categories. Social media as a dissemination channel of an outbreak was also tested in each of the models, but was insignificant. Other studies show that while social media is increasingly used as an information source (Weber Shandwick, 2018), findings show people are less likely to believe the health information they are reading from social media compared to other channels (Miraglia and Kinman, 2017). This could lead to social media's insignificant ability to increase risk perception.

Risk mitigation results

Results for Scenario 2 (i.e., travel-related actions when the respondent is infected) are discussed in Section 4.1 and presented in Table 3a, Table 3b . Results for Scenario 3 (i.e., travel-related actions during an influenza outbreak and the respondent is not infected) are discussed in Section 4.2 and presented in Table 4 . None of the models in this section passed the parallel regression assumption and, therefore, were required to use the generalized ordered logit methodology.

Table 3a

Scenario 2 travel-related actions to obtain treatment or prevent the spread of disease.

	Visit doctor's office					Avoid public places
	Parameters		Marginal effects			Parameters		Marginal effects
	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Sometimes	Always	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Sometimes	Always
Male						−0.385⁎⁎⁎	−0.385⁎⁎⁎	2.89%	6.19%	−9.08%
Kids_0_12
Experience						0.0343	−0.474⁎⁎⁎	−0.26%	11.25%	−10.99%
Vaccine	0.573⁎⁎⁎	0.573⁎⁎⁎	−11.12%	3.86%	7.26%
Health insurance	0.735⁎⁎⁎	0.735⁎⁎⁎	−15.61%	7.70%	7.90%	0.763⁎⁎⁎	0.763⁎⁎⁎	−7.54%	−9.14%	16.67%
Knowledge						0.240⁎⁎	0.240⁎⁎	−1.92%	−3.66%	5.58%
Working						−0.478⁎⁎⁎	−0.478⁎⁎⁎	3.39%	8.03%	−11.42%
Apartment/mobile	0.204⁎	0.204⁎	−3.70%	0.86%	2.84%
Race White, non-Hisp	−0.200⁎	−0.548⁎⁎⁎	3.65%	4.27%	−7.92%
Constant cuts 1 & 2	0.144	−2.376⁎⁎⁎				2.030⁎⁎⁎	−0.589⁎⁎⁎
Observations	2127					1955
AIC	4022.749					3521.207
BIC	4062.387					3565.832
McFadden adj. R-squared	0.021					0.026

Note: Significance. Codes: ‘⁎⁎⁎’ 0.001 ‘⁎⁎’ 0.01 ‘⁎’ 0.05 ‘.’ 0.1 ‘ ’ 1.

Table 3b

Scenario 2 additional travel-related actions to prevent the spread of disease.

	Avoid public transit					Stay home
	Parameters		Marginal effects			Parameters		Marginal effects
	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Sometimes	Always	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Sometimes	Always
Male	−1.691⁎⁎⁎	−0.257	19.78%	−14.66%	−5.13%	−0.361⁎⁎⁎	−0.361⁎⁎⁎	2.35%	5.93%	−8.29%
Kids_0_12
Experience						0.167	−0.498⁎⁎⁎	−1.06%	12.27%	−11.21%
Health insurance						0.857⁎⁎⁎	0.857⁎⁎⁎	−7.64%	−10.25%	17.89%
Knowledge						0.315⁎⁎⁎	0.315⁎⁎⁎	−2.23%	−4.86%	7.09%
Working						−0.705⁎⁎⁎	−0.705⁎⁎⁎	4.16%	12.49%	−16.65%
Income/HHSize						−0.0467⁎⁎⁎	−0.0467⁎⁎⁎	0.31%	0.76%	−1.06%
Age	0.0235⁎⁎⁎	0.0235⁎⁎⁎	−0.29%	−0.18%	0.47%
Constant cuts 1 & 2	1.725⁎⁎⁎	−1.842⁎⁎⁎				2.346⁎⁎⁎	−0.491⁎⁎
Observations	191					1955
AIC	363.545					3387.816
BIC	379.807					3438.02
McFadden adj. R-squared	0.035					0.035

Note: Significance. Codes: ‘⁎⁎⁎’ 0.001 ‘⁎⁎’ 0.01 ‘⁎’ 0.05 ‘.’ 0.1 ‘ ’ 1.

Table 4

Scenario 3 travel-related actions to reduce exposure to influenza.

	Avoid public places					Avoid public transit					Stay home
	Parameters		Marginals			Parameters		Marginals			Parameters		Marginals
	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Some-times	Always	Never vs Sometimes and Always	Never and Sometimes vs Always	Never	Some-times	Always	Never vs Some-times and Always	Never and Sometimes vs Always	Never	Some-times	Always
Male	−0.464⁎⁎⁎	−0.115	4.66%	−2.29%	−2.37%	−0.639	0.671⁎	9.47%	−20.02%	10.55%	−0.206⁎⁎	−0.206⁎⁎	3.11%	0.18%	−3.29%
Kids_0_12	−0.398⁎⁎⁎	−0.398⁎⁎⁎	4.37%	3.46%	−7.82%
Experience
Knowledge
Stores risk	0.718⁎⁎⁎	0.249⁎⁎	−8.36%	3.37%	4.99%						0.602⁎⁎⁎	20.70%	−9.99%	6.79%	3.20%
Transit risk						1.061⁎	1.061⁎	−18.85%	5.47%	13.37%
Work risk											−0.257⁎⁎	−0.257⁎⁎	3.78%	0.42%	−4.20%
Age	0.0114⁎⁎⁎	0.0114⁎⁎⁎	−0.11%	−0.12%	0.24%	0.0267⁎⁎⁎	0.0267⁎⁎⁎	−0.40%	−0.03%	0.43%
Income/HHsize	−0.116⁎⁎⁎	−0.116⁎⁎⁎	1.16%	1.23%	−2.39%	−0.187⁎⁎⁎	−0.187⁎⁎⁎	2.81%	0.18%	−2.99%	−0.103⁎⁎⁎	−0.103⁎⁎⁎	1.56%	0.09%	−1.65%
Race White	−0.202⁎⁎	−0.202⁎⁎	1.96%	2.28%	−4.24%						−0.363⁎⁎⁎	−0.363⁎⁎⁎	5.23%	0.83%	−6.06%
Single						0.839⁎⁎	0.839⁎⁎	−12.90%	−0.02%	12.92%
BS degree											−0.320⁎⁎⁎	−0.320⁎⁎⁎	4.94%	0.06%	−5.00%
Metro											−0.229⁎	−0.229*	3.30%	0.50%	−3.81%
Working											−0.741⁎⁎⁎	−0.741⁎⁎⁎	10.33%	2.43%	−12.76%
Constant cuts 1 & 2	1.808⁎⁎⁎	−0.953⁎⁎⁎				−0.228	−3.808⁎⁎⁎				2.726***	0.167
Observations	2023					178					1921
AIC	3692.154					337.625					3563.719
BIC	3748.277					363.079					3624.886
McFadden adj. R-squared	0.2					0.041					0.043

Note: Significance. Codes: ‘⁎⁎⁎’ 0.001 ‘⁎⁎’ 0.01 ‘⁎’ 0.05 ‘.’ 0.1 ‘ ’ 1.

Scenario 2

Table 3a, Table 3b present the results for visiting the doctor, avoiding public places, avoiding public transit, and staying home in Scenario 2. In the ‘Visit Doctor's Office’ model, the Race White (non-Hispanic) variable did not meet the parallel regression assumption and, therefore has different coefficients and trends change across the marginal effects percentages. That is, self-identified Caucasian, non-Hispanics were 3.65% more likely to respond ‘Never’ to visiting the doctor when sick compared to other races/ethnicities, 4.27% more likely to respond ‘Sometimes,’ and 7.92% less likely to respond ‘Always.’ This trend increases then decreases across the ordered responses. In the same model, the Vaccine variable did pass the parallel regression assumption, causing both of the variable's coefficients to be equal at 0.573.

The results show that males were less likely to prevent the spread of the disease, as they were statistically less likely than females to avoid public places, avoid public transit, and stay at home. This is noticeable in the marginal effects, where male respondents were, for example, 2.89% more likely to never avoid public places when experiencing influenza-like symptoms compared to female respondents. Males, however, were not statistically significantly less likely to visit the doctor and, therefore, were not any less active in obtaining treatment than females. Due to this insignificance, the gender variable is not included in the Visit Doctor's Office model.

For all Scenario 2 models, families with at least one child 12 years old or younger did not show statistically significantly different behavior than families without a child in that age range. Previous experience with influenza was a statistically significant predictor of behavior for avoiding public places and staying at home, where respondents with experience had a high likelihood of sometimes taking each travel-related action. This means that having experience with influenza decreased the chances of never or always taking these actions. Also, respondents who had received the vaccine and had health insurance were more likely to always visit the doctor's office if infected with influenza. These findings support that people who receive the annual vaccine are likely proactive about taking health measures, and having access to the health insurance allowed them to obtain affordable medical treatment.

Respondents who are employed were 11.42% less likely to always avoid public places and 16.65% less likely to always stay at home compared to those not currently employed. Similarly, respondents with a higher family income per capita showed a reduced rate in taking travel-related social distancing actions by being less likely to stay at home. The reduced rate of mitigation behavior of the employed could be due to the inability to take time off from work, either for financial or work policy reasons, even when infected with influenza. The frequency of presenteeism (i.e., working while sick) has been analyzed in literature and it has been “estimated that employee presenteeism is between 1.8 to 10 times more frequent than absenteeism” (Suciu, 2019). It is important to note that government regulations and stay-at-home orders for non-essential personnel during the COVID-19 pandemic likely prevented presenteeism that was seen in previous influenza outbreaks, such as seasonal.

Also, Scenario 2 models show that respondents who had knowledge of the difference between influenza and a stomach flu were more likely to avoid public places and to stay home when sick with influenza compared to their counterparts. Those identifying as Caucasian, non-Hispanic had decreased chances of visiting a doctor's office when having influenza. Older individuals were more likely to avoid transit to reduce the spread of influenza than their younger counterparts. Living in an apartment or mobile home increased the chances of seeking medical attention when infected.

Scenario 3

Table 4 presents the generalized ordered logit model results for Scenario 3, which includes the same travel-related actions as Scenario 2 with the exception of visiting the doctor as the respondent is not infected in this scenario and does not need to seek medical attention. Similar to the findings in Scenario 2, males were less likely to avoid public places and to stay at home to reduce exposure to influenza. However, males were much less likely to take action when already infected compared to women (Scenario 2) than when not infected (Scenario 3). Across all models in both scenarios, having a child younger than 12 years old in the household was only statistically significant in avoiding public places in Scenario 3. Having children in the household reduces one's ability to avoid public places when they are not sick. This could be due to obligations that the caregiver or other children in the household cannot get out of easily as well as the necessary errands a caretaker needs to make. Neither knowledge nor experience statistically significantly impacted a person's travel-related risk mitigation. Risk perception was significant in all three models. Respondents were more likely to avoid public places and stay at home if there was increased perceived risk in stores and more likely to avoid transit when there is increased risk perceived of contracting influenza in the transit system. However, respondents perceiving medium or high risk of contracting influenza at work were less likely to stay at home. This may be reasonable since employees may only be able to miss work when they are actually sick themselves.

In Scenario 2, household income per capita, self-identifying as Caucasian, non-Hispanic, and employment are indirectly related to taking action to prevent the spread when infected and the respondent's age was directly related to taking action. These indirect and direct relationships are also seen in Scenario 3's results with trying to reduce exposure to influenza. This suggests that people who take travel-related actions when infected with influenza to prevent the spread are more likely to take travel-related actions to prevent getting infected during an influenza outbreak.

Also, the significant variables in Scenario 3's models show that respondents with a higher education (i.e., at least a bachelor's degree) were less likely to stay at home to reduce exposure during an influenza outbreak. Single individuals (i.e., divorced, never married, or widowed) were more likely to avoid transit compared to their married counterparts, while those living in a metropolitan area were less likely to stay home. In general, findings from both Scenarios 2 and 3 support that wealthier and educated respondents (i.e., high household income per capita, are employed, and have at least a bachelor's degree) were less likely to reduce travel in response to influenza. This is evident from the statistical significance of the Income/HHSize, BS Degree, and Working variables. Specifically, in Scenario 2 high household income per capita was statistically significantly linked to less likely staying at home and those employed were less likely to avoid public places and stay at home. In Scenario 3, a high household income per capita was linked to respondents being less likely to avoid public places, avoid public transit, and stay at home. Those with a bachelor's degree or higher were less likely to stay at home. Also, those who were employed were less likely to stay at home.

Policy implications

This study shows the links between an individual's health, public health, risk perception and travel behavior. The findings first suggest that certain channels are more effective in disseminating risks associated with an influenza outbreak. Given that those receiving influenza outbreak information through word of mouth perceived higher risk at mandatory and medical trips while television lead to higher risk perception for discretionary trips, it is possible that the effectiveness of any one media channel is not consistent across people and trip purposes. This provides supportive evidence that government public outreach should utilize several dissemination channels during a virus outbreaks. To the authors' knowledge, there are no studies in literature on the changing effect of information source on risk perception by trip type to validate this finding.

Second, findings suggest certain policies could alter an individual's travel behavior during an influenza outbreak and could promote the health of the general public. For example, government notices and policies encouraging individuals to be more proactive about their health, as well as recommendations for self-quarantine (as in the COVID-19 situation), could reduce an individual's travel when sick and, therefore, could reduce the spread of the disease within the general public. This study found individuals who are proactive in their health are also proactive in protecting others from getting the disease. This was shown through the statistical significance of receiving the annual vaccine and maintaining health insurance coverage with reduced travel and greater likelihood of visiting the doctor when sick.

Third, this study highlights the importance of educating the public about influenza, how it is spread, and how it differs from a stomach flu. Individuals who knew the difference between influenza and the stomach flu before the survey were more likely to reduce travel when sick to reduce the spread to the general public. This knowledge had more influence than even an individual having an influenza experience in their household in the last 6 months (either themselves or a household member).

Fourth, the findings indicate that individuals who are wealthier, educated, and employed may be less likely to exhibit social distancing by avoiding public places and staying at home. In Scenario 2, when the individual is already infected with influenza, those with a higher family income per capita and those who were employed both were less likely to stay home to prevent the spread of the disease to others. Similarly, in Scenario 3, when the individual is not infected during an influenza outbreak, individuals who had at least a bachelor's degree, higher family income per capita, and/or were employed were less likely to reduce travel that could increase exposure to influenza. From a public health perspective, it could be beneficial to promote more flexible work hours and telecommuting (Liao et al., 2012).

Lastly, findings from Scenario 3 suggest that those who perceived higher risk for contracting influenza in the workplace were less likely to stay at home. Other trip types found a direct relationship between threat appraisal and self-efficacy in Scenario 3, such as for trips to the store and using transit. Therefore, model results indicate low levels of self-efficacy for work trips compare to other trip types. This counterintuitive relationship between risk perception and behavior was evident during the COVID-19 pandemic where employees with high interaction with the public (e.g., medical care staff, grocery shelves stocker, cashier, and food deliverer) were considered “essential” and therefore were required to continue their high-risk jobs (Hoff, 2020).

Conclusions and future research direction

This study contained several generalized ordered and ordered logit models surrounding an individual's risk perception and travel behavior during an influenza outbreak. Males were statistically significantly less likely than females to perceive risk as well as alter their travel patterns by voluntarily social distancing in response to the possibility of spreading the disease or increased exposure to the disease. Having children ages 12 and younger in the household did not impact an individual's likelihood of visiting a doctor's office, avoiding public places, avoiding transit, or staying home when already sick and the individual was less likely to avoid public places when they were not infected. This suggests that households with younger children may face challenges with voluntary social distancing when they and their household members are well. According to the marginal effects, having knowledge about influenza and how it is different from a stomach flu (before the survey) was more effective in reducing travel (avoiding public places and staying home) when the respondent is already sick than having experience with influenza in one's household in the last six months. Individuals who were proactive about taking care of their health (i.e., received annual vaccines and had health insurance coverage) were also proactive in reducing their exposure and spread of influenza. Lastly, in alignment with the Protection Motivation Theory, in general, individuals tended to reduce travel to locations in which they perceived medium or high risk of contracting influenza, such as stores. However, perception of increased risk at work locations did not show an increased likelihood to stay home.

Future research in this area could combine the findings from this study with agent-based models that track individual's travel participation and responses to virus outbreaks. This could lead to more refined estimates of transmission and levels of exposure. The effectiveness of workplace policies, other location specific policies, government orders, and general interventions may be tested with such models.

Additional studies could look more into the link between information channel and risk perception, including a validation and more in-depth analysis of why the effectiveness of information channels changes by trip type. Also, this link should be analyzed under several situations. Specifically, in this study, social media was found to be insignificant while word of mouth and television was significant. However, given that COVID-19 required social distancing and quarantine measures, it would be interesting to see if a person's reduced ability to get word of mouth information about a trip's risk meant other channels became more significant, such as social media. Lastly, future studies on the COVID-19 pandemic could evaluate the effects of telecommuting and e-learning options on travel-related risk mitigation.

Funding statement

This work was supported in part by the Grant number 1R01GM109718.

CRediT authorship contribution statement

Susan Hotle: Methodology, Formal analysis, Writing - review & editing. Pamela Murray-Tuite: Conceptualization, Methodology, Writing - review & editing, Funding acquisition. Kunal Singh: Writing - review & editing.

Declaration of competing interest

None.