Modeling COVID-19 Preventive Behavior: Impact of Neighborhood Characteristics,Cognitive Appraisals,and Information Use

Abstract

With bases in the Socio-Ecological Model (SEM), Protection Motivation Theory, and the Extended Parallel Process Model, this study builds a multilevel model including neighborhood risk factors, cognitive appraisals, and media information use with research hypotheses and questions developed at the individual, community, and cross levels. The research expectations are tested with 2020 individual-level survey data (N = 995) and 2019 and 2020 neighborhood-level data (N = 41) in New York City. In the context of COVID-19, multilevel modeling documented that the bulk of variance in the outcome variables was at the individual level. At the individual level, perceived efficacy, as well as its interaction with perceived threat, significantly predicted preventive behaviors, information scanning was significantly associated with perceived efficacy and threat, and information seeking was significantly associated with efficacy. At the community level, poverty rate was significantly associated with perceived efficacy and threat and preventive behaviors. At the cross level, community-level risk factors (e.g., poverty rate, crime rate) significantly moderated the effects of individual-level factors (i.e., perceived efficacy and information scanning) on preventive behaviors, which are indicative of the interdependence of factors at different levels in the SEM.

Keywords

socio-ecological model sociological factors media information use quantitative survey COVID-19 pandemic

Despite decades-old calls for a greater focus on multilevel inquiry in public health and communication (Diez Roux, 2001; Pan & McLeod, 1991), important questions remain unanswered. Socioecological factors influence health behaviors and outcomes (Viswanath & Emmons, 2006), but less is known about how they function relative to and interactively with individual-level cognitive and communication factors. This constitutes a gap in scientific knowledge, hindering the development of public health messaging, including how it can be tailored specially for individuals in different types of communities. The inclusion of media information use permits an innovative look at how communication effects vary relative to individuals’ cognitive processes and according to their neighborhoods of residence. This is important given that communication inquiry tends to be limited to the individual level and has not previously taken a multilevel approach to testing fear appeal theory in this manner.

Research expectations are articulated on the effects of individual-level factors (i.e., cognitive appraisals, media information use) and community-level risk factors (e.g., poverty rate), as well as the effects of cross-level interactions between community-level risk factors and individual-level factors. This inquiry is in the context of the COVID-19 pandemic, where research has documented the prevalence of fear (Vann et al., 2022), the influence of community structural factors on outcomes (Schulz et al., 2020), and the influence of behavioral recommendations on infection rates and other outcomes (Hanson et al., 2021; Lu et al., 2021). In March 2020, the World Health Organization described the mass of available COVID-19 information as an “infodemic” (United Nations, 2020), with COVID-19 a common topic in news and news use (Mitchell & Oliphant, 2020; Ogbodo et al., 2020).

Conceptual Model

Given its inclusion of individual-level factors, as well as factors of communities and societies where individuals live, the Socio-Ecological Model (SEM) is adopted as an overall theoretical framework, with fear appeals theory used to guide individual-level processes in a fear-rife pandemic scenario. Bronfenbrenner (1977) conceived four levels of the SEM: microsystem, mesosystem, ecosystem, and macrosystem. The conceptual model (see Figure 1) is based in the SEM of Ivankovich and colleagues (2013), which holds that an individual’s behavior is a function of factors at four levels: (1) individual, (2) relationships, (3) community, and (4) society. The effects of factors at different levels can be direct and independent of one another, but also interactive and dependent on one another (Elder et al., 2007). In terms of interactive influence, for example, community factors can moderate the effects of individual factors.

Figure 1.

Conceptual Model of Multilevel Processes (With Research Expectations Embedded).

Individual-Level Influence

This article does not test mediation processes of cognitive appraisals and uses Protection Motivation Theory (PMT) and the Extended Parallel Process Model (EPPM) to theorize upon other fear appeal functions. PMT holds that a persuasive message’s impact on a recommended behavior results from perceived vulnerability, severity, self-efficacy, and response efficacy (Rogers, 1975). In refinement, the EPPM frames how individuals develop perceptions of efficacy and threat and subsequent behaviors. Encompassing severity and susceptibility, threat is “a danger or harm that exists in the environment whether we know it or not” (Witte et al., 1996, p. 332). Encompassing response efficacy and self-efficacy, efficacy entails “the effectiveness, feasibility, and ease with which a recommended response impedes or averts a threat” (Witte et al., 1996, p. 332). How fear appeals stimulate these cognitive appraisals results in different behavioral processes, including danger control, fear control, and no response (Witte, 1992).

The greater the threat that individuals perceive, the more fear they experience and the more they are motivated to respond to the threat and adopt preventive behavior (Witte, 2001). COVID-19 research found that perceived severity and susceptibility are positively associated with preventive behavior (Nazione et al., 2021), neither perceived severity nor susceptibility has a significant effect on personal hygiene behaviors (i.e., facemask wearing, handwashing, and covering mouth when coughing), and perceived susceptibility has a positive effect on social distancing (Yoon et al., 2022). The inconsistency of these findings may result from how the influence of cognitive appraisals on preventive behaviors varies according to cultural, situational, and demographic factors. In addition, individuals appraise whether they can address a threat and whether such addressment will be successful or not. The greater the efficacy individuals perceive, the more able they will be to adopt a preventive behavior (Witte, 2001). COVID-19 research demonstrated that efficacy is positively associated with preventive behavior (Nazione et al., 2021), threat and efficacy are positively associated with stay-at-home intention (Tsoy et al., 2022), and response efficacy and self-efficacy are positively associated with personal hygiene behaviors and social distancing (Yoon et al., 2022):

Hypothesis 1a (H1a): COVID-19 threat and efficacy are positively associated with COVID-19 preventive behaviors.

Threat and efficacy may interact to influence preventive behaviors (Rogers, 1975; Witte, 1992). If individuals perceive efficacy to be low, they may perceive a threat to be beyond their behavioral control, resulting in fear avoidance. If individuals perceive efficacy to be high, they may perceive the threat to be within their behavioral control, resulting in behavior change. COVID-19 threat and efficacy interact to predict intention to stay home (Tsoy et al., 2022):

Hypothesis 1b (H1b): COVID-19 threat and efficacy interact to predict COVID-19 preventive behaviors.

Media information use is included in the conceptual model given its capacity to influence individuals’ perceived threat and coping strategies for threat. Information seeking entails “purposive acquisition of information from selected information carriers” (Johnson & Meischke, 1993, pp. 343–344). Information scanning is different, going “beyond incidental exposure, to include information a person encounters in the normal flow of information but decides to attend to” (Shim et al., 2006, p. 158) and tying in with news attention, which entails how individuals exert cognitive effort to process media messages (Chaffee & Schleuder, 1986). Determining its effectiveness (Witte & Allen, 2000), media messaging contains content specific to cognitive appraisals (Beaudoin, 2002; Klemm et al., 2016). COVID-19 research found that information seeking is positively associated with cognitive appraisals (Beaudoin & Hong, 2021):

Hypothesis 2a (H2a): COVID-19 information seeking and information scanning are positively associated with COVID-19 threat and efficacy.

Media information use can foster improvements in health behaviors. Media attention is positively associated with healthy lifestyle intention (Banerjee & Ho, 2020), and information seeking is positively associated with COVID-19 preventive behaviors (Beaudoin & Hong, 2021; T. K. Lee & Kim, 2024; Liu, 2020):

Hypothesis 2b (H2b): COVID-19 information seeking and information scanning are positively associated with COVID-19 preventive behaviors.

Community-Level Influence

Community factors can impact COVID-19 preventive behavioral processes. Where people live shapes their behavioral processes, with protective factors helping individuals’ health outcomes and risk factors hindering them (Centers for Disease Control and Prevention, 2021). Health behaviors and outcomes vary at the community level (Duncan et al., 1993), but such effects are much smaller than individual-level effects, sometimes explaining less than 1% of variance. Considered to be compositional (Macintyre et al., 2002), community-level factors include poverty rate, COVID-19 prevalence, health care access, and crime rate. Increased poverty makes communities less stable, limiting individuals’ access to health care and resources and impeding their development of health behaviors (Turner-Musa et al., 2020). Poverty can undercut health behaviors as a result of resource constraints and distrust of public health messaging. Neighborhood poverty rate is positively associated with COVID-19 infection rates (Karmakar et al., 2021).

Hypothesis 3 (H3): Community-level poverty rate is inversely associated with COVID-19 threat and efficacy and preventive behaviors.

Infectious disease rates vary according to community-level factors such as neighborhood disease rate (Kumar et al., 2012) and urbanity versus rurality (Bekalu et al., 2017). COVID-19 prevalence may predict preventive behavioral processes in two ways. First, increased COVID-19 prevalence in a community may increase residents’ perceptions of threat and efficacy and preventive behaviors. Second, increased COVID-19 prevalence in a community may occur if residents have lower perceived threat and efficacy and preventive behaviors:

Research Question 1 (RQ1): Is community-level COVID-19 prevalence positively or inversely associated with COVID-19 threat, efficacy, and preventive behaviors?

Health care access permits individuals to secure health and medical treatment. COVID-19 brought about decreases in health care access and use (Pujolar et al., 2022; von Humboldt et al., 2022). Health care access helps prevent negative health outcomes, including illness and death (Núñez et al., 2021). Specific to COVID-19, perceived health care access reduced perceptions of health vulnerability and fear, which reduced preventive behaviors (Vann et al., 2022):

Hypothesis 4 (H4): Health care access is inversely associated with COVID-19 threat, efficacy, and preventive behaviors.

Crime rate is a proxy for community instability, which upsets the social context of a neighborhood and its embedded social relationships (Durkheim, 1897/1951), leaving communities socially disorganized (Sampson & Groves, 1989) and prone to unhealthy behaviors and outcomes (Marmot & Wilkinson, 1999; Scheft & Fontenette, 2005). Crime rate can influence preventive behavioral processes by undercutting emotional and mental health, utilitarian physical activity, and social support networks. Neighborhood crime is positively associated with sexually transmitted disease rates (Cohen et al., 2000):

Hypothesis 5 (H5): Community-level crime rate is inversely associated with COVID-19 threat, efficacy, and preventive behaviors.

Cross-Level Influence

Cross-level processes entail how community-level factors can moderate the effects of individual-level factors. Neighborhood environmental factors moderate the effects of individual-level self-efficacy on physical activity (Carlson et al., 2012; Van Holle et al., 2015) and self-rated health (Y.-H. Lee & Fan, 2023). There could be synergies or conflicts in how such interactions, respectively, promote or deter preventive behaviors (Elder et al., 2007). Specific to COVID-19, the influence of community risk factors could synergistically amplify the effects of perceived threat as a result of heightened salience or, in a conflictual manner, mitigate the effects as a result of pandemic fatigue:

Research Question 2 (RQ2): Do community-level risk factors moderate the effects of COVID-19 threat and efficacy on COVID-19 preventive behaviors?

Community factors can moderate the effects of media information use. On one hand, when community-level risk factors are high, the beneficial effects of media information use may be weakened. Individuals who live in rural areas, which are prone to inequalities and health care barriers, benefit less from media information use in developing HIV testing intention (Bekalu et al., 2017) and social capital (Beaudoin & Thorson, 2004). In addition, community-level protective factors enhance how individuals use media information to achieve beneficial outcomes, including community participation (Paek et al., 2005) and civic participation (Kang & Kwak, 2003). On the other hand, when community-level risk factors are high, the beneficial effects of media information use may be strengthened. Media Dependency Theory holds that, when individuals are in a social system rife with conflict and crisis, they tend to depend more on and be more influenced by media messaging (Ball-Rokeach & DeFleur, 1976). Residents in rural areas benefit more than urban residents from network television news use in developing social capital (Beaudoin & Thorson, 2004):

Research Question 3 (RQ3): Do community-level risk factors moderate the effects of COVID-19 information seeking and information scanning on COVID-19 threat, efficacy, and preventive behaviors?

Method

The individual-level data came from a cross-sectional survey of adults who lived in Bronx, Brooklyn, Manhattan, and Queens in New York City (aged 18 and older) from April 28 to May 6, 2020 (N = 995).¹ Qualtrics derived the sample and hosted the online survey.² The community-level data are from 41 neighborhoods in the four boroughs. Poverty and crime rate data, from 2019 and 2020, respectively, were obtained from the Furman Center for Real Estate and Urban Policy at New York University (NYU Furman Center, 2021). The New York City Department of Health and Mental Hygiene (NYC Health) provided the COVID-19 prevalence data from April 28 to May 6, 2020.³ The health care access data also came from the prior source (New York City Department of Health and Mental Hygiene, 2025). Institutional review board (IRB) approval was acquired at Boston University. A prior multilevel study used similar data sets, but diverged methodologically, theoretically, and conceptually with a focus on the Theory of Planned Behavior (Beaudoin, 2024).

Individual-Level Measurement

Descriptive statistics are reported in Table 1. The survey questionnaire was specific to the coronavirus COVID-19 pandemic, using the term “coronavirus” in questions (see survey items in Supplementary Files). COVID-19 information scanning had two items specific to attention (Chaffee & Schleuder, 1986; Shim et al., 2006)—online news information (M = 7.15, SD = 2.02) and television news information (M = 7.13, SD = 2.22)—which were added and divided by two to represent information scanning (r = .57, p < .001). There was one item for COVID-19 information seeking (Beaudoin, 2023; National Cancer Institute, 2019). Responses were recoded in terms of days per month (0–30).

Table 1.

Descriptive Statistics.

Variable	M	SD	Min	Max
Individual level
Age	36.60	13.38	18	72
Male	50.15%		0	1
Education	5.48	1.49	1	7
White	58.49%		0	1
Household income	5.93	2.40	1	9
Employment	57.69%		0	1
Health insurance	9.05%		0	1
Political party ID	2.99	1.83	1	6
Tested positive for COVID-19	3.82%		0	1
Information seeking	17.32	12.50	0	30
Information scanning	7.14	1.88	1	9
Efficacy	4.23	0.81	1	5
Threat	3.44	0.87	1	5
Preventive behaviors	4.37	0.89	1	5
Community level
Crime rate	.00	1.81	−1.98	7.26
COVID-19 prevalence	.00	1.95	−2.70	7.55
Poverty rate	.25	0.18	0.04	0.79
Health care access	.00	1.89	−2.28	3.98

Note. Individual-level and community-level sample sizes are 995 and 41, respectively. Percentages are reported for dichotomous variables.

Likert-type scales were used for COVID-19 cognitive appraisals (Witte, 2001). Severity and susceptibility had two items each, which were added and divided by four to represent threat (α = .94). Self-efficacy and response efficacy had six items each, which were added and divided by 12 to comprise efficacy (α = .75). Five-point scales were used to measure sheltering in place and wearing a face mask. The two items were added and divided by two to constitute COVID-19 preventive behaviors (r = .46, p < .001).

Sociodemographics included male, White, age, education, household income, full-time employment, health insurance, tested positive for COVID-19, and political party identification from strong Democrat (1) to strong Republican (6) (Institute for Social Research, 2021).

Community-Level Measurement

Neighborhoods were the community-level unit (Diez Roux, 2001). Close-ended survey questions specified the neighborhood of residence of each respondent. Two neighborhood 2020 indicators measured crime rate: property crime (M = 11.73, SD = 11.09) and violent crime (M = 6.76, SD = 5.30). After standardization, the items were grouped to represent crime rate (r = .64, p < .001). One neighborhood 2019 indicator measured poverty rate. Zip code−level data from April 28 to May 6, 2020, measured cumulative COVID-19 cases and cumulative COVID-19 deaths in the neighborhoods. The indicators were averaged to create daily cumulative neighborhood totals: COVID-19 cases (M = 277.84, SD = 194.92) and COVID-19 deaths (M = 31.76, SD = 19.89). After standardization, the indicators were grouped to constitute COVID-19 prevalence (r = .90, p < .001). Two neighborhood 2019−2020 indicators measured health care access: percentage of adults without health insurance (M = 19.00, SD = 13.28) and percentage of adults without needed medical care (M = 18.72, SD = 10.95). After standardization, the items were grouped to represent health care access (r = .79, p < .001).

Statistical Analysis

Stata 17 was used for multilevel modeling (MLM), which specifies and fits multilevel and hierarchical models with nested levels of independent variables. Maximum likelihood estimation was implemented, with robust standard errors to mitigate concern for how assumed nonnormality and model misspecification can derive inaccurate standard errors and jeopardize hypothesis testing (Garson, 2020; Rabe-Hesketh & Skrondal, 2005). Independent variables were grand-mean centered (Hofmann & Gavin, 1998) and entered in three staged models: (1) a full unconditional (or null) model; (2) a random-intercept full conditional model with independent variables at the individual and community levels; and (3) another random-intercept full conditional model with cross-level interactions. The individual-level sample of 995 respondents and community-level sample of 41 neighborhoods (M = 24.3 respondents per neighborhood; range = 10–31) are generally adequate for MLM (Maas & Hox, 2005). The statistical power is sufficient for the number of independent variables and the multilevel research design. Interclass correlations (ICC) are provided, and significant interactions are plotted at the mean (M) minus one standard deviation (SD), at the M, and at the M plus one SD (Aiken & West, 1991). There were significant correlations between the community-level risk factors, but the calculation of variance inflation factors (VIFs) raised no significant concerns about multicollinearity.⁴

Results

The intercepts were nonsignificant, specifying the absence of overall community effects, and the residuals were significant, specifying the presence of within-group variance in dependent variables (see Model 1 in Tables 2 –4). The ICCs were low in the models. Even in such situations, MLM provides advantages to modeling multilevel processes (Garson, 2020; Rabe-Hesketh & Skrondal, 2005).

Table 2.

Multilevel Model of COVID-19 Preventive Behaviors.

Variable	Model 1	Model 2	Model 3
Fixed effects
Individual level
Intercept	4.38*** (.03)	4.40*** (.02)	4.40*** (.02)
Age		.004* (.00)	.003 (.00)
Male		−.12*** (.04)	−.12*** (.03)
Education		.06*** (.02)	.06*** (.02)
White		−.04 (.06)	−.03* (.06)
Household Income		.02 (.01)	.02 (.01)
Employment		−.13 (.05)	−.13 (.05)
Health insurance		.05 (.09)	.04 (.09)
Political party ID		−.01 (.01)	−.01 (.01)
Tested positive for COVID-19		−.22* (.09)	−.19* (.08)
Information scanning (SC)		.03 (.02)	.002 (.02)
Information seeking (SE)		.002 (.00)	.02 (.00)
Efficacy (EF)		.62*** (05)	.62*** (.04)
Threat (TH)		.03 (.03)	.04 (.03)
EF × TH		−.10** (.04)	−.12** (.04)
Community level
COVID-19 prevalence		.004 (.01)	.001 (.01)
Poverty rate		.23* (.10)	.17 (.11)
Health care access		−.01 (.01)	−.02 (.01)
Crime rate		−.01 (.01)	.002 (.01)
Cross level
SC × COVID-19 Prevalence			−.01 (.04)
SC × Poverty Rate			.34** (.01)
SC × Health Care Access			−.01 (.11)
SC × Crime Rate			−.03* (.01)
SE × COVID-19 Prevalence			.001 (.01)
SE × Poverty Rate			.003 (.00)
SE × Health Care Access			.000 (.00)
SE × Crime Rate			−.002 (.00)
TH × COVID-19 Prevalence			−.01 (.01)
TH × Poverty Rate			−.20 (.15)
TH × Health Care Access			.03 (.03)
TH × Crime Rate			−.01 (.01)
EF × COVID-19 Prevalence			.03 (.03)
EF × Poverty Rate			−.72* (.29)
EF × Health Care Access			−.01 (.03)
EF × Crime Rate			.06** (.02)
Random effects
Level 1 (Residual)	.78***	.38***	.37***
Level 2 (Intercept)	.000	.000	.000
Interclass correlation	.000	.000	.000

Note. Unstandardized coefficients and standard errors (in parentheses) are reported. Individual-level and community-level sample sizes are 995 and 41, respectively.

p < .05. **p < .01. ***p < .001.

Table 3

Multilevel Model of Threat.

Variable	Model 1	Model 2	Model 3
Fixed effects
Individual level
Intercept	3.44*** (.03)	3.44*** (.03)	3.43*** (.03)
Individual-level variables
Age		.004** (.00)	.004** (.00)
Male		.01 (.05)	.01 (.05)
Education		−.003 (.02)	−.01 (.02)
White		.07 (.07)	.06 (.06)
Household income		.03 (.01)	.03 (.01)
Employment		−.05 (.07)	−.04 (.07)
Health insurance		.07 (.10)	.06 (.10)
Political party ID		−.05** (.02)	−.05** (.02)
Tested positive for COVID-19		.27 (.16)	.27 (.16)
Information scanning (SC)		.08*** (.01)	.09*** (.01)
Information seeking (SE)		.002 (.00)	.002 (.00)
Community level
COVID-19 prevalence		.03 (.01)	.02 (.02)
Poverty rate		−.55** (.20)	−.45* (.21)
Health care access		.03 (.01)	.03 (.02)
Crime rate		.01 (.03)	−.01 (.01)
Cross level
SC × COVID-19 Prevalence			−.01* ().01
SC × Poverty Rate			−.13 (.08)
SC × Health Care Access			−.004 (.01)
SC × Crime Rate			.02* (.01)
SE × COVID-19 Prevalence			.002 (.00)
SE × Poverty Rate			.01 (.01)
SE × Health Care Access			.001 (.00)
SE × Crime Rate			−.002 (.00)
Random effects
Level 1 (Residual)	.76***	.70***	.70***
Level 2 (Intercept)	.000	.000	.000
Interclass correlation	.01	.000	.000

Note. Unstandardized coefficients and standard errors (in parentheses) are reported. Individual-level and community-level sample sizes are 995 and 41, respectively.

p < .05. **p < .01. ***p < .001.

Table 4.

Multilevel Model of Efficacy.

Variable	Model 1	Model 2	Model 3
Fixed effects
Individual level
Intercept	4.23*** (.03)	4.23*** (.02)	4.23*** (.02)
Individual-level variables
Age		.003* (.00)	.003* (.00)
Male		−.08 (.05)	−.08 (.05)
Education		−.01 (.02)	−.01 (.02)
White		−.03 (.06)	−.03 (.06)
Household income		.03* (.01)	.03* (.01)
Employment		−.05 (.06)	−.06 (.06)
Health insurance		.23* (.11)	.24* (.11)
Political party ID		−.04*** (.01)	−.04*** (.01)
Tested positive for COVID-19		−.28 (.18)	−.26 (.18)
Information seeking (SE)		.01*** (.00)	.01*** (.00)
Information scanning (SC)		.12*** (.15)	.11*** (.01)
Community level
COVID-19 prevalence		.01 (.01)	.02 (.01)
Poverty rate		−.63** (.23)	−.61** (.22)
Health care access		.04 (.02)	.03 (.02)
Crime rate		.000 (.01)	.01 (.01)
Cross level
SC × COVID-19 Prevalence			.01 (.01)
SC × Poverty Rate			.29** (.10)
SC × Health Care Access			−.03*** (.01)
SC × Crime Rate			−.01 (.01)
SE × COVID-19 Prevalence			−.001 (.00)
SE × Poverty Rate			−.01 (.02)
SE × Health Care Access			−.001 (.00)
SE × Crime Rate			.002 (.00)
Random effects
Level 1 (Residual)	.000	.000	.000
Level 2 (Intercept)	.65***	.54***	.53***
Interclass correlation	.000	.000	.000

Note. Unstandardized coefficients and standard errors (in parentheses) are reported. Individual-level and community-level sample sizes are 995 and 41, respectively.

p < .05. **p < .01. ***p < .001.

Support for H1a is limited to the significant association between efficacy and preventive behaviors (b = .62; see Model 2 in Table 2). Supportive of H1b, the interaction of threat and efficacy significantly predicted preventive behaviors (b = −.10; see Model 2 in Table 2). In Supplemental Figure 2, for low levels of efficacy, there was a positive relationship between threat and preventive behaviors, but for high levels of efficacy, the relationship was less positive.

Partially supporting H2a, information scanning was positively associated with threat (b = .08) and efficacy (b = .12), and information seeking was positively associated with efficacy (b = .01), but not threat (b = .002; see Model 2 in Tables 3 and 4). Unsupportive of H2b, information scanning (b = .03) and information seeking (b = .002) were not significantly associated with preventive behaviors (see Model 2 in Table 2).

Support for H3 includes that poverty rate was inversely associated with threat (b = −.55) and efficacy (b = −.63; see Model 2 in Tables 3 and 4). The association between community-level poverty rate and preventive behaviors was significant, but had a positive coefficient (b = .23; see Model 2 in Table 2).

In terms of RQ1, community-level COVID-19 prevalence was not significantly associated with threat (b = .03), efficacy (b = .01), or preventive behaviors (b = .004; see Model 2 in Tables 2 –4). Unsupportive of H4, health care access was not significantly associated with threat (b = .03), efficacy (b = .04), or preventive behaviors (b = −.01; see Model 2 in Tables 2 –4). Unsupportive of H5, community-level crime rate was not significantly associated with threat (b = .01), efficacy (b = .000), or preventive behaviors (b = −.01; see Model 2 in Tables 2 –4).

In terms of RQ2, there were two significant interactions in predicting preventive behaviors—Efficacy × Poverty Rate (b = −.72), and Efficacy × Crime Rate (b = .06; see Model 3 in Table 2 and Supplemental Figures 3 and 4). In terms of RQ3 (see Model 3 in Tables 3 and 4), there were two significant interactions in predicting threat—Information Scanning × COVID-19 Prevalence (b = −.01), and Information Scanning × Crime Rate (b = .02; see Supplemental Figures 5 and 6)—and two significant interactions in predicting efficacy—Information Scanning × Poverty Rate (b = .29), and Information Scanning × Health Care Access (b = −.03; see Supplemental Figures 7 and 8). There were two significant interactions in predicting preventive behaviors—Information Scanning × Poverty Rate (b = .34), and Information Scanning × Crime Rate (b = −.03; see Supplemental Figures 9 and 10).

The eight significant cross-level interactions can be split into two sets. In Supplemental Figure 3, for low levels of poverty rate, there was a positive relationship between efficacy and preventive behaviors, but for high levels of poverty rate, the relationship became negative. A generally similar pattern can be seen in Supplemental Figures 5, 7, 8, and 10. In Supplemental Figure 4, for low levels of crime rate, there was a negative relationship between efficacy and preventive behaviors, but for high levels of crime rate, the relationship became positive. A generally similar pattern can be seen in Supplemental Figures 6 and 9.

Discussion

Individual-Level Processes

Consistent with multilevel research (Duncan et al., 1993), this study demonstrated that individual-level factors explain the bulk of variance in the outcomes. That behaviors were predicted by efficacy, but not threat, is counter to theory, but consistent with COVID-19 research (Yoon et al., 2022), suggesting the greater salience of individuals’ perceptions of their capacity to perform a behavior over their perceptions of the potential harm posed by the pandemic. This effect pattern may result from how assured respondents were in the preventive benefits of face masks and sheltering in place and, conversely, how the development of fear appeals may have been relatively immature at the time as a result of few early infections experienced by respondents or perceptions that reported deaths were to at-risk individuals. That said, the significant interaction of efficacy and threat supports prior research (Tsoy et al., 2022), indicating the importance of both cognitive appraisals.

Information scanning was positively associated with threat and efficacy, whereas information seeking was positively associated with only efficacy. Neither information measure was significantly associated with preventive behaviors, which is counter to some research (Shim et al., 2006), including that on COVID-19 (Beaudoin & Hong, 2021). Media information use may be effective in eliciting changes in cognitive appraisals, but can come up short in terms of behaviors, which are tenacious and resistant to change (Beaudoin et al., 2007). These nonsignificant effects may also have resulted from the sociopolitical landscape, with political polarization concerning preventive behaviors and related media content. As compared with information seeking, information scanning played a more substantive role at the individual level, as well as at the cross level, which speaks to the importance of individuals’ cognitive exertion in media use (Chaffee & Schleuder, 1986).

Community-Level Processes

Significant community-level effects were limited to poverty rate, which was inversely associated with efficacy and threat, but positively associated with preventive behaviors. Consistent with prior research (Cohen et al., 2000; Karmakar et al., 2021), the first two negative effects indicate how poverty makes communities less stable, limiting access to health care and resources (Turner-Musa et al., 2020). The third positive effect may result from how some U.S. cities actively addressed COVID-19 prevention in higher-risk communities (Mullachery et al., 2022).

Cross-Level Processes

The first set of five interactions (see Supplemental Figures 3, 5, 7, 8, and 10) suggests that the positive effects of information scanning, as well as efficacy in one case, were weakened for individuals who lived in communities with worse risk factors. These interactions involved poverty in two cases, as well as COVID-19 prevalence, health care access, and crime rate in one case each. These results are consistent with research on community-level risk factors (Beaudoin & Thorson, 2004; Bekalu et al., 2017) and community-level protective factors (Kang & Kwak, 2003; Paek et al., 2005). The four media-related interactions pertain to developing a community-level segmentation approach for health campaign message dissemination. Indicating how cross-level processes can be conflictual (Elder et al., 2007), individuals who lived in neighborhoods with higher collective risk factors were less likely to benefit from health information.

The second set of three interactions (see Supplemental Figures 4, 6, and 9) suggests that the positive effects of information scanning, as well as efficacy in one case, were strengthened for individuals who lived in communities with worse risk factors. These interactions involved crime rate in each of the three cases. These results pertain to how individuals who live in communities with greater social instability may augment media dependence and media effects (Ball-Rokeach & DeFleur, 1976). The two media-related interactions pertain to developing a community-level segmentation approach for health campaign message dissemination. Indicating how cross-level processes can be synergistic (Elder et al., 2007), individuals who lived in neighborhoods with higher collective risk factors were more likely to benefit from health information.

Conclusion

Behavioral outcomes were a greater function of individual-level predictors such as cognitive appraisals and media information use than community-level risk factors. Thus, while “where you live” matters, “who you are” matters more. The results cast light on the independent and interactive influence of cognitive appraisals and media information use on preventive behaviors. Two primary limitations deserve acknowledgment. First, scholars should be cautious in extrapolating the results. There may be a variation in results in different cities and across different phases of the pandemic, including in regard to pandemic fatigue and vaccination accessibility. Second, given its use of cross-sectional survey data, this study does not permit the derivation of causal inferences. It is recommended that future research employ panel survey data to help address this limitation.

Supplemental Material

sj-docx-1-heb-10.1177_10901981251337678 – Supplemental material for Modeling COVID-19 Preventive Behavior: Impact of Neighborhood Characteristics, Cognitive Appraisals, and Information Use

Supplemental material, sj-docx-1-heb-10.1177_10901981251337678 for Modeling COVID-19 Preventive Behavior: Impact of Neighborhood Characteristics, Cognitive Appraisals, and Information Use by Christopher E. Beaudoin in Health Education & Behavior

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Christopher E. Beaudoin

Supplemental Material

Supplemental material for this article is available online at

Notes

References

Aiken

L. S.

West

S. G.

(1991). Multiple regression: Testing and interpreting interactions. Sage.

Allison

P. D.

(2012, September 10). When can you safely ignore multicollinearity? https://statisticalhorizons.com/multicollinearity

Ball-Rokeach

DeFleur

(1976). A dependency model of mass-media effects. Communication Research, 3, 3–21. https://doi.org/10.1177/009365027600300101

Banerjee

S. S.

(2020). Applying the theory of planned behavior: Examining how communication, attitudes, social norms, and perceived behavioral control relate to healthy lifestyle intention in Singapore. International Journal of Healthcare Management, 13(Suppl. 1), 496–503. https://doi.org/10.1080/20479700.2019.1605687

Beaudoin

C. E.

(2002). Exploring antismoking ads: Appeals, themes, and consequences. Journal of Health Communication, 7, 123–137. https://doi.org/10.1080/10810730290088003

Beaudoin

C. E.

(2023). Do social media matter? The effects of information seeking on COVID-19 psychological and behavioral processes. Telematics and Informatics, 83, 102027. https://doi.org/10.1016/j.tele.2023.102027

Beaudoin

C. E.

Fernandez

Wall

J. L.

Farley

T. A.

(2007). Promoting healthy eating and physical activity: Short-term effects of a mass media campaign. American Journal of Preventive Medicine, 32(3), 217–223.

Beaudoin

C. E.

Hong

(2021). Predictors of COVID-19 preventive perceptions and behaviors among millennials: Two cross-sectional survey studies. Journal of Medical Internet Research, 23(8), Article e30612. https://doi.org/10.2196/30612

Beaudoin

C. E.

Thorson

(2004). Social capital in rural and urban communities: Testing differences in media effects and models. Journalism & Mass Communication Quarterly, 81(2), 378–399. https://doi.org/10.1177/107769900408100210

10.

Beaudoin

C. E.

(2024). A multilevel study of preventive behavioral outcomes: The relative and interactive influences of media information use and neighborhood risk factors. Journalism & Mass Communication Quarterly. Advance online Publication. https://doi.org/10.1177/10776990241262348

11.

Bekalu

M. A.

Eggermont

Viswanath

K. V.

(2017). HIV/AIDS communication inequalities and associated cognitive and affective outcomes: A call for a socioecological approach to AIDS communication in Sub-Saharan Africa. Health Communication, 32(6), 685–694. https://doi.org/10.1080/10410236.2016.1167999

12.

Bronfenbrenner

(1977). Toward an experimental ecology of human development. American Psychologist, 32(7), 513–531. https://doi.org/10.1037/0003-066X.32.7.513

13.

Carlson

J. A.

Sallis

J. F.

Conway

T. L.

Saelens

B. E.

Frank

L. D.

Kerr

Cain

K. L.

King

A. C.

(2012). Interactions between psychosocial and built environment factors in explaining older adults’ physical activity. Preventive Medicine, 54(1), 68–73. https://doi.org/10.1016/j.ypmed.2011.10.004

14.

Centers for Disease Control and Prevention. (2021). The social-ecological model: A framework for prevention. https://www.cdc.gov/violenceprevention/about/social-ecologicalmodel.html

15.

Chaffee

S. H.

Schleuder

(1986). Measurement and effects of attention to media news. Human Communication Research, 13(1), 76–107. https://doi.org/10.1111/j.1468-2958.1986.tb00096.x

16.

Cohen

Spear

Scribner

Kissinger

Mason

Wildgen

(2000). “Broken windows” and the risk of gonorrhea. American Journal of Public Health, 90(2), 230–236. https://doi.org/10.2105/ajph.90.2.230

17.

Diez Roux

A. V

. (2001). Investigating neighborhood and area effects on health. American Journal of Public Health, 91(11), 1783–1789. https://doi.org/10.2105/ajph.91.11.1783

18.

Duncan

Jones

Moon

(1993). Do places matter? A multi-level analysis of regional variations in health-related behaviour in Britain. Social Science & Medicine, 37(6), 725–733. https://doi.org/https://doi.org/10.1016/0277-9536(93)90366-C

19.

Durkheim

(1951). Suicide. Free Press. (Original work published 1897)

20.

Elder

J. P.

Lytle

Sallis

J. F.

Young

D. R.

Steckler

Simons-Morton

Stone

Jobe

J. B.

Stevens

Lohman

Webber

Pate

Saksvig

B. I.

Ribisl

(2007). A description of the social-ecological framework used in the trial of activity for adolescent girls (TAAG). Health Education Research, 22(2), 155–165. https://doi.org/10.1093/her/cyl059

21.

Garson

G. D.

(2020). Multilevel modeling. Sage.

22.

Hanson

C. L.

Crandall

Barnes

M. D.

Novilla

M. L.

(2021). Protection motivation during COVID-19: A cross-sectional study of family health, media, and economic influences. Health Education & Behavior, 48(4), 434–445. https://doi.org/10.1177/10901981211000318

23.

Hofmann

D. A.

Gavin

M. B.

(1998). Centering decisions in hierarchical linear models: Implications for research in organizations. Journal of Management, 24(5), 623–641. https://doi.org/10.1016/S0149-2063(99)80077-4

24.

Institute for Social Research. (2021). Party identification. https://www.icpsr.umich.edu/web/pages/instructors/setups/notes/party-identification.html

25.

Ivankovich

M. B.

Fenton

K. A.

Douglas

J. M.

Jr. (2013). Considerations for national public health leadership in advancing sexual health. Public Health Reports, 128, 102–110. https://doi.org/10.1177/00333549131282s112

26.

Johnson

J. D.

Meischke

(1993). A comprehensive model of cancer-related information seeking applied to magazines. Human Communication Research, 19(3), 343–367. https://doi.org/10.1111/j.1468-2958.1993.tb00305.x

27.

Kang

Kwak

(2003). A multilevel approach to civic participation: Individual length of residence, neighborhood residential stability, and their interactive effects with media use. Communication Research, 30(1), 80–106. https://doi.org/10.1177/0093650202239028

28.

Karmakar

Lantz

P. M.

Tipirneni

(2021). Association of social and demographic factors with COVID-19 incidence and death rates in the US. JAMA Network Open, 4(1), Article e2036462. https://doi.org/10.1001/jamanetworkopen.2020.36462

29.

Klemm

Das

Hartmann

(2016). Swine flu and hype: A systematic review of media dramatization of the H1N1 influenza pandemic. Journal of Risk Research, 19(1), 1–20. https://doi.org/10.1080/13669877.2014.923029

30.

Kumar

Quinn

S. C.

Kim

K. H.

Musa

Hilyard

K. M.

Freimuth

V. S.

(2012). The social ecological model as a framework for determinants of 2009 H1N1 influenza vaccine uptake in the United States. Health Education & Behavior, 39(2), 229–243. https://doi.org/10.1177/1090198111415105

31.

Kutner

Greenberg

Jin

Paulsen

(2006). The health literacy of America’s adults: Results from the 2003 National Assessment of Adult Literacy (NCES 2006-483). National Center for Education Statistics.

32.

Lee

T. K.

Kim

H. K.

(2024). Extending the knowledge gap hypothesis to narrative persuasion: Parents’ information-seeking intention and learning about adolescent children’s Covid-19 vaccination. Health Education & Behavior, 51(2), 270–279. https://doi.org/10.1177/10901981231216742

33.

Lee

Y.-H.

Fan

S.-Y.

(2023). Psychosocial and environmental factors related to physical activity in middle-aged and older adults. Scientific Reports, 13(1), 7788. https://doi.org/10.1038/s41598-023-35044-4

34.

Liu

P. L.

(2020). COVID-19 information seeking on digital media and preventive behaviors: The mediation role of worry. Cyberpsychology, Behavior, and Social Networking, 23(10), 677–682. https://doi.org/10.1089/cyber.2020.0250

35.

Liu

Yuan

Y. C.

Burns

K. S.

(2021). Source trust and COVID-19 information sharing: The mediating roles of emotions and beliefs about sharing. Health Education & Behavior, 48(2), 132–139. https://doi.org/10.1177/1090198120984760

36.

Maas

Hox

(2005). Sufficient sample sizes for multilevel modeling. Methodology: European Journal of Research Methods for the Behavioral and Social Sciences, 1(3), 86–92. https://doi.org/10.1027/1614-2241.1.3.86

37.

Macintyre

Ellaway

Cummins

(2002). Place effects on health: How can we conceptualise, operationalise and measure them? Social Science & Medicine, 55(1), 125–139. https://doi.org/https://doi.org/10.1016/S0277-9536(01)00214-3

38.

Marmot

Wilkinson

R. G.

(1999). Social determinants of health. Oxford University Press.

39.

Mitchell

Oliphant

J. B.

(2020). Americans immersed in COVID-19 news; most think media are doing fairly well covering it. Pew Research Center. https://www.pewresearch.org/journalism/2020/03/18/americans-immersed-in-covid-19-news-most-think-media-are-doing-fairly-well-covering-it/

40.

Mullachery

P. H.

Melly

Kolker

Barber

Diez Roux

A. V.

Bilal

(2022). Inequities in spatial accessibility to COVID-19 testing in 30 large US cities. Social Science & Medicine, 310, 115307. https://doi.org/10.1016/j.socscimed.2022.115307

41.

National Cancer Institute. (2019). Health Information National Trends Survey (HINTS) 5, Cycle 3. National Cancer Institute, Division of Cancer Control and Population Sciences. https://hints.cancer.gov/dataset/HINTS5_Cycle3_Stata_20210305.zip

42.

Nazione

Perrault

Pace

(2021). Impact of information exposure on perceived risk, efficacy, and preventative behaviors at the beginning of the COVID-19 pandemic in the United States. Health Communication, 36(1), 23–31. https://doi.org/10.1080/10410236.2020.1847446

43.

New York City Department of Health and Mental Hygiene. (2025). Community health profiles. https://a816-health.nyc.gov/hdi/profiles/

44.

Núñez

Sreeganga

S. D.

Ramaprasad

(2021). Access to healthcare during COVID-19. International Journal of Environmental Research and Public Health, 18(6), 2980. https://doi.org/10.3390/ijerph18062980

45.

NYU Furman Center. (2021). New York neighborhood data profiles. Furman Center for Real Estate and Urban Policy. https://furmancenter.org/neighborhoods

46.

Ogbodo

J. N.

Onwe

E. C.

Chukwu

Nwasum

C. J.

Nwakpu

E. S.

Nwankwo

S. U.

Nwamini

Elem

Iroabuchi Ogbaeja

(2020). Communicating health crisis: A content analysis of global media framing of COVID-19. Health Promotion Perspectives, 10(3), 257–269. https://doi.org/10.34172/hpp.2020.40

47.

Paek

H.-J.

Yoon

S.-H.

Shah

D. V.

(2005). Local news, social integration, and community participation: Hierarchical linear modeling of contextual and cross-level effects. Journalism & Mass Communication Quarterly, 82(3), 587–606. https://doi.org/10.1177/107769900508200307

48.

Pan

McLeod

J. M.

(1991). Multilevel analysis in mass communication research. Communication Research, 18(2), 140–173. https://doi.org/10.1177/009365091018002002

49.

Pujolar

Oliver-Anglès

Vargas

Vázquez

M. L.

(2022). Changes in access to health services during the COVID-19 pandemic: A scoping review. International Journal of Environmental Research and Public Health, 19(3). https://doi.org/10.3390/ijerph19031749

50.

Rabe-Hesketh

Skrondal

(2005). Multilevel and longitudinal modeling using Stata. Stata Press.

51.

Rogers

R. W.

(1975). A Protection Motivation Theory of fear appeals and attitude change. Journal of Psychology, 91, 93–114. https://doi.org/10.1080/00223980.1975.9915803

52.

Sampson

R. J.

Groves

W. B.

(1989). Community structure and crime: Testing social-disorganization theory. American Journal of Sociology, 94(4), 774–802. http://www.jstor.org.ezproxy.bu.edu/sn/2780858

53.

Scheft

Fontenette

D. C.

(2005). Psychiatric barriers to readiness for treatment for Hepatitis C Virus (HCV) infection among injection drug users: Clinical experience of an addiction psychiatrist in the HIV-HCV coinfection clinic of a public health hospital. Clinical Infectious Diseases, 40(Suppl. 5), S292–S296. https://doi.org/10.1086/427443

54.

Schulz

A. J.

Mehdipanah

Chatters

L. M.

Reyes

A. G.

Neblett

E. W.

Jr. Israel

B. A.

(2020). Moving health education and behavior upstream: Lessons from COVID-19 for addressing structural drivers of health inequities. Health Education & Behavior, 47(4), 519–524. https://doi.org/10.1177/1090198120929985

55.

Shim

Kelly

Hornik

(2006). Cancer information scanning and seeking behavior is associated with knowledge, lifestyle choices, and screening. Journal of Health Communication, 11, 157–172. https://doi.org/10.1080/10810730600637475

56.

Tsoy

Godinic

Tong

Obrenovic

Khudaykulov

Kurpayanidi

(2022). Impact of social media, Extended Parallel Process Model (EPPM) on the intention to stay at home during the COVID-19 pandemic. Sustainability, 14(12), 7192. https://www.mdpi.com/2071-1050/14/12/7192

57.

Turner-Musa

Ajayi

Kemp

(2020). Examining social determinants of health, stigma, and COVID-19 disparities. Healthcare, 8(2), 168. https://doi.org/10.3390/healthcare8020168

58.

United Nations. (2020, March 31). UN tackles “infodemic” of misinformation and cybercrime in COVID-19 crisis. The Department of Global Communications. https://www.un.org/en/un-coronavirus-communications-team/un-tackling-‘infodemic’-misinformation-and-cybercrime-covid-19

59.

U.S. Bureau of the Census. (2022). QuickFacts: New York City, New York. https://www.census.gov/quickfacts/newyorkcitynewyork

60.

Van Holle

Van Cauwenberg

Deforche

Van de Weghe

De Bourdeaudhuij

Van Dyck

. (2015). Do psychosocial factors moderate the association between objective neighborhood walkability and older adults’ physical activity? Health & Place, 34, 118–125. https://doi.org/10.1016/j.healthplace.2015.05.004

61.

Vann

R. J.

Tanner

E. C.

Kizilova

(2022). Perceived access, fear, and preventative behavior: Key relationships for positive outcomes during the COVID-19 health crisis. Journal of Consumer Affairs, 56(1), 141–157. https://doi.org/10.1111/joca.12439

62.

Viswanath

Emmons

K. M.

(2006). Message effects and social determinants of health: Its application to cancer disparities. Journal of Communication, 56(Supplement), S238–S264. https://doi.org/10.1111/j.1460-2466.2006.00292.x

63.

von Humboldt

Low

Leal

. (2022). Health service accessibility, mental health, and changes in behavior during the COVID-19 pandemic: A qualitative study of older adults. International Journal of Environmental Research and Public Health, 19(7), 4277. https://www.mdpi.com/1660-4601/19/7/4277

64.

Witte

(1992). Putting the fear back into fear appeals: The Extended Parallel Process Model. Communication Monographs, 59(4), 329–349. https://doi.org/10.1080/03637759209376276

65.

Witte

(2001). Putting it all together: The Extended Parallel Process Model. In Witte

Meyer

Martell

(Eds.), Effective health risk messages (pp. 23–31). Sage.

66.

Witte

Allen

(2000). A meta-analysis of fear appeals: Implications for effective public health campaigns. Health Education & Behavior, 27, 591–615.

67.

Witte

Cameron

K. A.

McKeon

J. K.

Berkowitz

J. M.

(1996). Predicting risk behaviors: Development and validation of a diagnostic scale. Journal of Health Communication, 1(4), 317–341. https://doi.org/10.1080/108107396127988

68.

Yoon

You

Shon

(2022). An application of the Extended Parallel Process Model to protective behaviors against COVID-19 in South Korea. PLOS ONE, 17(3), Article e0261132. https://doi.org/10.1371/journal.pone.0261132

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.40 MB