By: Eliana Nowlis, FCRH ’23

Abstract

The framing of a public health recommendation can dramatically impact how many people comply with it, which in turn affects public health overall. Framing is a psychological effect that causes people to have different reactions to the same information depending on how it is presented to them. This research investigated which framing techniques are the most successful in getting college students to comply with a public health recommendation. Subjects were presented with information about two hypothetical public health recommendations in an online survey and asked about their willingness to comply with them. The framing of the recommendations emphasized either positive or negative outcomes and was either community or individually focused. The individual frame elicited more compliance among college students than the community frame in one scenario according to a regression analysis. Colleges can use this information to tailor public health recommendations and improve student compliance and health. 

1. Introduction 

People typically like to believe that they are in control of their opinions, but in reality, there are many external factors that can have a hidden influence on what they do or think. For example, small changes in the presentation of an event or issue can have a large impact on people’s opinion about it (Chong & Druckman, 2007). This phenomenon is called framing and is caused by the way our brains process information. Humans interpret information by fitting the words or numbers with which they are presented into an existing conceptual system, which gives that information meaning (Kahneman, 2013, p. 324). When the same information has different wording or a different frame, our brains fit it differently into our existing framework. This is when framing effects become apparent. In this paper, framing is discussed in the context of public health. One of the most crucial and difficult jobs public health organizations have is getting people to listen to their recommendations. The difference between someone following their advice or not could be something as small as one of the words used in an announcement. The question I researched is as follows: “Which frames are the most successful in getting college students to comply with public health recommendations?” The frames included in the study were positive versus negative and community versus individual. This paper provides background information on the concept of public health framing, presents hypotheses and experimental methods, and discusses the findings. It also provides a look into the future for what may be next in this field. 

1.i. Literature on Framing

The first experiment that dealt with the subject of framing was one done by Amos Tversky and Daniel Kahneman (1981). The researchers asked their subjects to choose between two programs that each gave a different way to combat a fantasy disease that had been discovered in America and was expected to kill 600 people. These two plans were framed positively or negatively, with half of the subjects receiving each frame. If a subject received the positive frame, they were asked to choose between these two programs: 

“If Program A is adopted, 200 people will be saved,” or “If Program B is adopted, there is 1/3 probability that 600 people will be saved, and 2/3 probability that no people will be saved.”

If a subject received the negative frame, they were asked to choose between these two programs: 

“If Program C is adopted, 400 people will die,” or “If Program D is adopted, there is 1/3 probability that nobody will die, and 2/3 probability that 600 people will die.”

The difference in framing had a drastic effect on people’s choices. In the positively framed scenario (Program A versus Program B), people tended to choose Program A, the risk adverse plan, because “the prospect of certainly saving 200 lives is more attractive than a risky prospect of equal expected value” (Tversky & Kahneman, 1981). In contrast, people tended to choose Program D, the risk-taking option, in the negatively framed scenario because the certain death of 400 people seemed unacceptable (Tversky & Kahneman, 1981). Even public health officials were susceptible to framing effects when given this problem, which proves that this is not an issue of education or previous knowledge. Tversky and Kahneman’s results have been replicated numerous times in more recent experiments and researchers continue to find that different frames have a measurable effect on opinion. Abhijit Banerjee has done research on framing effects and the COVID-19 virus in India and has shown that there is an increase in compliance to public health mandates when the information is framed as a way to keep the community healthy, rather than the individual (Banerjee et al., 2020). Research done by other groups has also shown that reframing an issue as a community problem rather than an individual problem has led to an increase in public awareness and compliance (Dorfman et al., 2005). It is embarrassing to think that the choices we make can be controlled by such a seemingly insignificant factor, but we must acknowledge that important decisions can be influenced by the way information is framed (Kahneman, 2013). 

1.ii. Public Health Framing 

This experiment focuses on the role framing plays in public health messaging. When releasing information to the public, public health officials must think about how they should best frame the information to get the desired effect. For example, we can examine how the Center for Disease Control (CDC) presents information on their website to attempt to influence people to wash their hands more often. One of the strategies the CDC uses is framing hand washing as a family activity. At the top of the web page, it says “Wash hands. Teach kids to do the same” (CDC, n.d.). They also use both individual and community frames. This can be seen in the headline “Regular hand washing is one of the best ways to remove germs, avoid getting sick, and prevent the spread of germs to others,” with the individual frame at the beginning of the sentence and the community frame at the end (CDC, n.d.). The deliberate framing the CDC displays on their web page shows that, in the field of public health, the way information is presented is very carefully thought out so that it can have the best response possible from the public. Governments, hospitals, non-profits, and many other organizations all keep the framing of information in mind when they are releasing messages about health to the public. 

1.iii. Context and Importance 

Public health officials put so much thought into the framing of their recommendations because it is usually in the best interest of everyone that they are followed. We are currently seeing numerous examples of public health framing in recommendations during the COVID-19 pandemic. This experiment was born out of a curiosity about what can be done to make people listen to authorities on COVID-19 and other public health issues. No matter what the recommendation is, there are frequently people who oppose and disregard it. The correct framing of a message can limit these dissenters because it can increase popular support and compliance (Dorfman et al., 2005, p. 323). This is why it is so important that we study framing. By finding the most effective framing for these issues, more people will follow the guidelines, and public health will be improved overall. 

1.iv. Frames Studied 

I studied the impact of two framing effects on the willingness of college students to comply with a public health recommendation. The first is the widely established positive versus negative framing effect, which was discovered by Tversky and Kahneman in their “live” versus “die” experiment (Tversky & Kahneman, 1981). They were the first to prove that this difference in framing influences opinion, as we process positive and negative messages differently. In my experiment, information is presented either in the framework of likeliness to stay healthy (positive) or likeliness to get sick (negative). The second framing effect is the community versus individual frame. Research has been done on this framing effect and the COVID-19 pandemic and has shown that when the information is framed as a way to keep the community healthy, it elicits a different response than focusing on individual health (Banerjee e al., 2020). My experiment follows this example and frames the information in terms of either the effect on individual or community health. These two framing effects are combined in a two-by-two matrix and tested to see how they affect college students’ willingness to comply to a public health recommendation. 

2. Hypotheses 

Hypothesis 1: When looking at college students’ willingness to comply with a public health recommendation, the positive (staying healthy) frame will be more persuasive than the negative (getting sick) frame. Positively framed messages have been shown to be more effective than negatively framed messages when trying to have people do something (Lee et al., 2018). In my experiment, this means that people will be more likely to comply with the recommendation under a positive framing. 

Hypothesis 2: When looking at college students’ willingness to comply with a public health recommendation, the community frame will be more persuasive than the individual frame. In experiments done by Banerjee et al. (2020) and Dorfman et al. (2005), the community frame has been shown to increase public compliance to a public health recommendation. Based on these previous experiments, I predict that in my experiment people will be more likely to comply with a recommendation if it is delivered in a community frame. 

3. Experimental Methods 

To test these hypotheses, I conducted an experiment through an anonymous online survey. No names, email addresses, or other identifying information were linked to the subjects’ responses to ensure privacy. The survey began with demographic questions about gender, age, ethnicity, and education. I studied college students, so only subjects who answered both “under 18” or “18–30” for age and “some college” or “undergraduate degree” for education were included. This population includes current college students and people who recently graduated college and will be referred to as simply “college students” throughout the paper. After the demographic questions, there was an attention check modeled after one appearing in Haaland et al. (2020, p. 32). Subjects were presented with an informative paragraph about the problem of lack of attention in online surveys and asked to answer “yellow” to the next question. At the end of the paragraph, they were asked “What is your favorite color?” The responses separate those who read the questions carefully from those who did not, and by filtering out the 13 subjects who answered with a color other than yellow the data became more reliable. Next, subjects were presented with two hypothetical scenarios (the order was randomized). I chose to invent a scenario rather than use a real example to avoid previous biases the subjects may have. Hypothetical scenarios are “highly predictive of real-world behaviors,” so my experiment has a real-world application even though it is fictional (Haaland et al., 2020, p. 15). One of the scenarios consisted of a viral eye infection that could be prevented by wearing goggles. The other was about a bacterial infection that could be prevented by not eating dairy products. After reading the background information for their first scenario, subjects received the framed information—a statistic about how effective the preventative measure (either wearing goggles or not eating dairy) was. Subjects randomly received exactly one of the following four sentences: 

They have discovered that by wearing safety goggles in public . . .

. . . the people you come in contact with have a 30% chance of staying healthy. (Positive/Community)

. . . you have a 30% chance of staying healthy. (Positive/Individual)

. . . the people you come in contact with have a 70% chance of getting sick. (Negative/Community)

. . . you have a 70% chance of getting sick. (Negative/Individual) 

These framings were mirrored in the dairy scenario, but with the probabilities switched (70% chance of health, 30% chance of sickness). Subjects were then asked for a quantitative point belief to measure willingness to comply. I chose this method for its simplicity as well as its historical sensitivity to question framing (Haaland et al., 2020). For the goggles scenario, subjects were asked if they left their house 5 times a day, how many of those times (0–5) they would wear goggles. For the dairy scenario, they were asked how many days a week (0–7) they would give up dairy products. After the subjects completed their first scenario, the process was repeated with the second. The survey was distributed on social media and through flyers in the Bronx and around Fordham. Data was collected from 180 college students.

3.i. Methods of Analysis 

I chose to analyze my data using a linear regression model, which is a process used to estimate the relationship between a dependent variable and one or more independent variables. First, the independent variables were checked for correlations between them in order to avoid negative effects on the regression. No two variables were highly correlated, so the analysis could proceed. For the goggles scenario, I began with a linear regression equation used to estimate the response of any given subject (i) based on their gender, ethnicity, and the framing they received. The equation is as follows:

y_i = α + (β_g)(g_i) + (β_e)(e_i) + (β_f1)(f1_i) + (β_f2)(f2_i) + ϵ_i

where the variables are defined as:

y_i – The number of times per day subject i reported willingness to wear goggles in public.

α – The baseline response of an average Asian female with a positive and individual frame. These baseline traits were arbitrarily chosen by the statistical software package (R) and have no effect on the results, only on how they are interpreted. This is the intercept of the equation.

β_g,e,f1,f2 – Each of the β coefficients represent a numerical deviation from the baseline response if a subject does not have the baseline traits (i.e., male, non-Asian, negative or community frame).

g,e,f1,f2 – Binomial variables that take a value of 0 if the subject has the baseline trait for that variable and 1 if they do not.

ϵ – The residual error term representing any variation subject i showed from the average for their group.

An identical equation was used for the dairy scenario, but the values of the response variable yi and each of the β coefficients were in terms of the number of days per week the subject reported willingness to give up dairy products. I then used a stepwise approach for each of the two scenarios to produce equations that only included β variables that had statistical significance. In statistics, we can consider something to be significant if its p value (a measure of how likely it is that the results are due to random chance,) is ≤ 0.05. For each step, the variable with the least statistical significance (highest p value) was removed from the equation. This process was repeated until the two equations only included β variables with statistical significance.

4. Results

The main finding from the goggles scenario was that males reported willingness to wear goggles 0.44 fewer times per day than females. The main finding from the dairy scenario was that the individual framing of information elicited 0.67 more days per week of compliance from subjects than the community framing. The statistical analysis is detailed below:

Goggles: Only one predictive variable in the goggles scenario was shown to be significant with its p value at the ≤ 0.05 level, which was Gender(Male). The final model was derived from the initial equation using the steps detailed above and follows the equation:

y_i = α + (β_g)(g_i) + ϵ_i

where α is the baseline response for females and β_g is the deviation from the baseline of the average male. The β_g coefficient (estimate) and statistical significance (p value) are displayed below:

Factor	Estimate	Std. Error	t statistic	p value
Intercept(Female)	α = 4.6731	0.1141	40.948	<2e-16
Gender(Male)	βg = −0.4378	0.1815	−2.412	0.0169

Table 1: Goggles Statistics

The estimate of -0.44 tells us that the average male willingness to wear goggles in public was 0.44 outings less per day than the average female willingness of 4.7 outings per day. The model was statistically significant (F = 5.828, DF = 2, 178, p = 0.00356). However, only 6.38% of the variability in stated willingness to wear goggles is explained by the predictive variable in the model (R² = 0.0638). 

Dairy: In the dairy scenario, the predictive variable that was shown to be significant at the ≤0.05 level was Framing(Individual). The final model was derived from the initial equation and follows the equation:

y_i = α + (β_f1)(f1_i) + ϵ_i

where α is the baseline response for subjects with a community frame and β_f1 is the deviation from the baseline for subjects with an individual frame. The β_f1 coefficient (estimate) and statistical significance (p) are displayed below:

Factor	Estimate	Std. Error	t statistic	p value
Intercept(Community)	α = 5.2000	0.2207	23.564	<2e-16
Framing(Individual)	βf1 = -0.6652	0.3086	2.156	0.0325

Table 2: Dairy Statistics 

The estimate of 0.67 means that receiving the individual frame increased the average willingness to give up dairy by 0.67 days per week compared to the average willingness with the community frame of 5.2 days per week. The model was statistically significant (F = 4.647, DF = 2, 178, p = $.0325). However, only 2.63% of the variability in stated willingness to give up dairy products is explained by the predictive variables in the model (R² = 0.0263).

5. Limitations 

Initially, I had ambitions of studying the full population of the US and solicited participants broadly. Due to time and resource constraints however, this ambition turned out to be unrealistic. My protocol for recruiting participants relied primarily on social media and, unsurprisingly, their demographics reflected this. In contrast to my intended population, the population sampled consisted primarily of college students in the US. The population sampled may have been even more specific than that, as the majority of my respondents were likely students at Fordham University. I was unable to examine this consideration further because respondents were not asked what school they attended. The demographics also indicated that the sampling was not balanced. There were a higher percentage of both women and Caucasians. However, with this narrowly focused population, there were enough samples from each demographic category to have some statistical power in looking for effects. The imbalances were not themselves a source of bias in my results because I used regression analysis with demographic information included as factors. This approach produced unbiased results, assuming that the sample was representative of each of the demographic groups.

One of the biggest sources of error in experiments like mine are experimenter demand effects. However, evidence suggests that in online surveys experimenter demand effects have a limited quantitative importance, which is one of the reasons I chose to collect data this way (Haaland et al., 2020, p.24). However, while an online experiment may have reduced experimenter demand effects, it certainly introduced a selection bias for people who chose to take the survey. 

The last source of error is the prior biases people may hold about public health due to the current state of the world. The fact that we are in the midst of a global pandemic means that people have a heightened sensitivity to matters concerning health and personal protection. This may have skewed the numbers upwards and resulted in an inflated view of people’s true willingness to make these behavioral changes. Along these same lines, as this was a survey and not a behavioral study, we cannot be sure how accurate the subjects’ self-reporting was. People are a lot more likely to say they will do something than to actually do it, so this was also a possible source of inflation for the numbers. Although these errors most likely contributed to higher numbers overall, my study of the framing was presumably unaffected because I was studying the differences in responses rather than the responses overall. Due to this distinction, the error associated with previous bias and self-reporting was likely avoided.

6. Discussion 

We can now look at the regression data to discover what effect framing had. In the goggles scenario, gender, not framing, was the significant factor. Males reported willingness to wear goggles 0.44 fewer times per day than females. This may be due to the difference in development between male and female brains. Females begin the process of streamlining their connections, or maturing, about 5 years prior to males and this discrepancy may explain the increased male aversion to following official recommendations (Lim et al., 2015). This male behavioral tendency can also be seen in the current pandemic—one study observing grocery store shoppers found that females were 1.5 times more likely to be wearing a mask than males (Haischer et al., 2020).

In the dairy scenario, we do see evidence of framing effects. The individual framing of information elicited 0.67 more days per week of compliance from subjects than the community framing. This is an interesting result because Banerjee’s study in India found the opposite to be true (Banerjee et al., 2020). Additionally, many of the countries with exceptionally successful responses to the COVID-19 pandemic, such as New Zealand, used community frames to relay messages to their citizens (The Workshop, 2020). Clearly, there is a significant amount of evidence contrary to my result, but there are a few possible explanations. Firstly, the act of giving something up fits into our existing framework differently than the act of doing something (in other experiments subjects were asked to do actions like wash hands or wear a mask). The individual frame may have been more appealing in my experiment because it made the act of giving up dairy feel like a trade for good health rather than a loss for the sake of the community. In contrast, no significance was shown between the community and individual frames in the goggles scenario, where subjects were asked to do something rather than give something up.

Another possible explanation is the individualist character of American society (Dorfman et al., 2005). Citizens of the United States tend to care less about community welfare and more about what is best for them (Dorfman et al., 2005). Therefore, the individual framing may have been more persuasive because it emphasized what Americans care about most—themselves. These theories serve as possible explanations for my data, but it is important to note that they are simply conjecture, and further research would be necessary to confirm their validity.

7. Conclusion

Neither of my hypotheses were supported by this study. No significant difference was shown between positive and negative framing, and the community frame was less effective than the individual frame. The goal was to answer the following question: “Which frames are the most successful in getting college students to comply with public health recommendations?” I have found that, in scenarios where you are asking college students to give something up, they are most willing to comply with a recommendation when the information is delivered in an individual frame. 

My experiment is somewhat lacking in the statistical rigor of a more professional project. If it could be redone with more participants, as well as with a broader and deeper set of questions, we may be able to learn more. I am interested in doing a follow-up experiment on the differences in asking someone to do something versus asking someone not to do something, as was mentioned in the discussion of the dairy scenario. There are endless combinations of frames and scenarios to be researched as we attempt to understand the impact frames have on public opinion. The ongoing pandemic has highlighted the importance of doing research in the field of public health messaging, and I hope my results can contribute to the growing knowledge of the impact of framing and inspire future discovery in the area. 

References

Banerjee, A., Alsan, M., Breza, E., Chandrasekhar, A. G., & Chowdhury, A. (2020, July). Messages on COVID-19 prevention in India increased symptoms reporting and adherence to preventive behaviors among 25 million recipients with similar efects on non-recipient members of their communities. National Bureau of Economic Research, 27496. https://www.nber.org/papers/w27496

Centers for Disease Control and Prevention. (n.d.). Handwashing: Clean hands save lives. Retrieved from: https://www.cdc.gov/handwashing/index.html

Chong, D., & Druckman, J. (2007, June 10). Framing theory. Annual Review of Political Science, 10, 103–126. https://www.annualreviews.org/doi/abs/10.1146/annurev.polisci.10.072805.103054

Dorfman, L., Wallack, L., & Woodruff, K. (2005, June). More than a message: framing public health advocacy to change corporate practices. Health Education & Behavior, 32(3), 320–336. http://www.bmsg.org/wp-content/uploads/2005/06/bmsg_report_more_than_a_message_framing_public_health_advocacy_to_change_corporate_practices.pdf

Haaland, I., Roth, C., & Wohlfart, J. (2020, June 26). Designing information provision experiments. Economic and Social Research Council, 1275. http://wrap.warwick.ac.uk/138921/1/WRAP-designing-information-provision-experiments-Roth-2020.pdf

Haischer, M., Beilfuss, R., Hart, M. R., Opielinski, L., Wrucke, D., Zirgaitis, G., Uhrich, T. D., Hunter, S. (2020, October 15). Who is wearing a mask? Gender-, age-, and location-related differences during the COVID-19 pandemic. PLoS ONE, 15(10). https://doi.org/10.1371/journal.pone.0240785 

Kahneman, D. (2013). Thinking, fast and slow. Farrar, Straus and Giroux.

Lee, H. C., Liu, S. F., & Cheng, Y. C. (2018). Positive or negative? The influence of message framing, reglatory focus, and product type. International Journal of Communication, 12, 788–805. https://ijoc.org/index.php/ijoc/article/viewFile/7601/2266

Lim, S., Han, C., Uhlhaas , P., & Kaiser, M. (2015). Preferential detachment during human brain development: age- and sex-specific structural connectivity in diffusion tensor imaging (DTI) data. Cerebral Cortex, 25(6), 1477–1489. https://pubmed.ncbi.nlm.nih.gov/24343892/

Tversky, A., & Kahneman, D. (1981, January 30). The framing of decisions and the psychology of choice. Science, 211(4481), 453–458. https://science.sciencemag.org/content/211/4481/453

The Workshop. (2020, March). How to talk about COVID-19: narratives to support good decision-making and collective action. Retrieved from: https://static1.squarespace.com/static/5e582da2de97e67b190b180c/t/5e964886d84b1b61324f0f4d/1586907349458/The-Workshop-Covid19-2020.pdf