By Cecilia Smith ‘25 and Emmett Zamore ‘24
Russia’s full scale invasion of Ukraine has now been raging for almost two years, with no end in sight. The conflict has huge implications for the well being of Ukrainian civilians, as well as geopolitics more broadly. U.S. assistance has been critical to the Ukrainian war effort, accounting for much of the aid the country has received and allowing Ukraine to replenish rapidly dwindling supplies. Whether that assistance continues is an open question, however, with Congress currently debating whether to send Ukraine a new tranche of aid. Part of the flagging political will for aiding Ukraine can likely be laid at the feet of declining popular support for this aid, with Americans–particularly Republicans–becoming increasingly opposed to it.
With this backdrop, we attempt to analyze how media framing of the war impacts popular support for both additional humanitarian and additional military aid to Ukraine. We concentrate on two commonly discussed implications of the Russian invasion: the humanitarian impact and the international security impact.
Using a survey experiment, we find that a newspaper article with a paragraph focusing on the international security impact of the war makes respondents more supportive of humanitarian aid to Ukraine at a statistically significant level as compared to a control article that does not include a paragraph focusing on the international security impact of the war. Additionally, Democratic respondents who received the international security framing were statistically significantly more likely to support providing additional humanitarian and military aid to Ukraine.
Expectations and Hypotheses
Foreign policy is by definition somewhat removed from most people’s everyday life. As a result, media coverage is generally the source of the American public’s foreign policy knowledge. Studies have found that in addition to mass media significantly influencing the salience of foreign policy in public opinion, media framing can meaningfully influence the direction of their opinions.
For this reason, we are interested in seeing how media framing of the war between Russia and Ukraine influences the support respondents express for providing either humanitarian or military aid to Ukraine. Our survey provides participants with one of three articles. One article includes no treatment, and the other two articles either frame the war to emphasize the ongoing humanitarian crisis or the potential international security threat. We expect the following:
- The international security frame will lead individuals to express increased support for military aid as compared to the humanitarian frame or control.
- The humanitarian frame will lead individuals to express increased support for humanitarian aid as compared to the international security frame or control.
Design, Data & Measures
To examine how media framing of the war in Ukraine impacts support for U.S. aid to Ukraine, we conducted a survey experiment fielded by Prolific between November 30 and December 1 on a sample of 1,415 American adults. Our experiment was based around a mock newspaper article excerpt that included some basic information about the war, the high level of U.S. assistance to Ukraine, and the likelihood of a protracted and indefinite conflict. Noting that there is little hope for a quick resolution was important, as recent media coverage has emphasized the stagnation of Ukrainian advances—a point that could serve to reduce support for continued U.S. aid. The article was presented to all respondents. The control group received no additional information while the two treatment groups each received an additional paragraph emphasizing either the international security or humanitarian implications of the war. The base article used for both the control and treatments is below, with the two potential treatment paragraphs noted:
In 2022, Russia began a full-scale invasion of neighboring Ukraine, intensifying a conflict that began in 2014. Since the Russian invasion began, tens of thousands of Ukrainian and Russian soldiers have been killed.
The international community strongly condemned Russia’s actions, and Western countries, led by the United States and the European Union, imposed economic sanctions against Russia. The U.S. also has led Western nations in giving Ukraine billions of dollars worth of humanitarian and military assistance.
[HUMANITARIAN TREATMENT] The invasion in 2022 has led to a severe humanitarian crisis. Ukrainian and U.S. officials have reported that thousands of civilian lives have been lost and the number continues to rise. Experts also say that more than thirteen million Ukrainians have been forced to leave their homes, and that adults and children alike are seeking safety and relief from the ongoing danger.
[INTERNATIONAL SECURITY TREATMENT] The invasion in 2022 has led to a severe global security crisis. Ukrainian and U.S. officials have warned that Russia’s invasion undermines the sanctity of country borders, and if successful could lead to further invasions by Russia or other countries. Experts also warn that if Russia is allowed to take over Ukraine it could attack a NATO country, which would bring the U.S. into a direct war with Russia.
Polling and statements from Ukrainian leaders show that Ukrainians do not support becoming part of Russia even if they have to fight indefinitely. Politicians and experts agree that the war is likely to continue without a resolution in the foreseeable future.
By having our control article include most of the information in our two treatments, we were able to more precisely isolate the impact of the two frames we were attempting to account for. If the treatments impacted responses as compared to the control, we can be confident that the shift is a result of a framing decision within a broader article, not simply the fact that respondents received information about the war. Including the treatments in a more wide ranging article also provides a tougher test of whether the national security or humanitarian frame impact support for aid to Ukraine, better simulating the real world press environment and giving our experiment more external validity.
Our two most important dependent variables read as follows:
Do you support the U.S. continuing to provide financial humanitarian assistance of any kind to Ukraine? (4 = Strongly oppose, 3 = Somewhat oppose, 2 = Somewhat support, 1 = Strongly support, 5 = Don’t Know)
Do you support the U.S. continuing to provide financial military assistance of any kind to Ukraine? (4 = Strongly oppose, 3 = Somewhat oppose, 2 = Somewhat support, 1 = Strongly support, 5 = Don’t Know)
To make our analysis more intuitive, we recoded these questions to go from strongly oppose at 1 to strongly support at 4. We set “don’t know” responses to missing in order to avoid making assumptions about respondents’ opinions, but as a robustness check we also tried placing “don’t know” responses in the middle of the scale, and this scaling did not significantly change any of our results.
In addition to our core dependent variables, we asked respondents the following questions:
Do you believe one form of assistance should be of higher priority than the other?
Do you agree with the following statement: “The war in Ukraine directly impacts the United States in a negative way.”
(IF YES TO CONTINUED AID) Do you support additional financial assistance for Ukraine even if this increases the national debt of the United States?
(IF YES TO CONTINUED AID) Do you think that the United States should continue providing financial assistance to Ukraine indefinitely, or do you think there should be a cutoff date?
(IF CUTOFF) When should the U.S. cut off aid to Ukraine?
To analyze the impact of our treatments on our dependent variables, we performed linear OLS regressions analyzing the effect of the treatment on the main dependent variables. None of our secondary dependent variables were impacted by the treatment in a statistically significant way, so we focused only on the two primary dependent variables. Because the covariate of having a military connection did not adequately randomize between experimental groups, we included the military variable in our basic models. The military variable was simplified by setting it to 1 if the respondent was in the military or had a family connection to the military (1-3 in the original survey question), 0 if the respondent had no military connection (5 in the original survey question) and missing if the respondent refused to answer (4 in the original survey question). To increase the robustness of our results as well as examine the impact of demographic variables on support for funding Ukraine, we also estimated linear regression models that included party ID, amount of news consumed, religious service attendance, whether a respondent consumes international news, eastern European connection, age, gender, education, and whether a respondent was White or non-white as covariates.
The party ID variable was a recoded version of the two party ID questions asked to respondents, with strong Republicans, not very strong Republicans, and Republican-leaning independents coded as 1, true independents coded as 2, and strong, not very strong, and independent leaning Democrats coded as 3. The Eastern European connection variable was set to 1 if respondents were from Eastern Europe or had family from Eastern Europe, and 0 otherwise. The news variable was recoded so that the value would increase from 1 to 5 as the respondent consumed more news. Similarly, the frequency of religious services variable was recoded to go from 1 to 5 as a respondent attended religious services more often, with “prefer not to say” responses coded as missing.
It is important to note that the survey our data comes from was not a demographically representative sample of the U.S. population. In particular, the sample was heavily Democratic and young, with self-identified Democrats or Democratic-leaning independents making up 57% of respondents and respondents over 60 making up just 13% of the sample. Because party identification is strongly and increasingly associated with support for U.S. aid to Ukraine, the heavily Democratic nature of the sample is particularly critical to keep in mind.
Results
Overall, there were high levels of support for both humanitarian and military aid. 71.9% of respondents either somewhat or strongly supported humanitarian aid, while 62.4% of respondents either somewhat or strongly supported military aid.
Across the treatment and control conditions the results were somewhat different than expected. The highest levels of support for humanitarian aid were found among the group of respondents who received the international security treatment. Specifically, 45.32% of respondents in this group strongly supported humanitarian aid, compared to 39.58% of respondents who received the humanitarian treatment, and 35.21% of respondents who received the control article.
To test if these differences were significant, we ran several regression models. The first were bivariate models, testing only the effects of the treatment and military connection on support for humanitarian aid or military aid. In the military aid model, although the international security treatment somewhat increased support for military aid, it was not at a statistically significant level.
On the other hand, as shown in Figure 1 the humanitarian aid model revealed that respondents who received the international security treatment were indeed more likely than those in the control group to support humanitarian aid (p=0.027). While statistically significant, the magnitude of the effect was small, with the international security frame causing an increase from 2.98 out of 4 to 3.14 out of 4 on the support for humanitarian aid scale when compared to the control article. Regardless, this effect was the opposite of what we were expecting in our initial hypotheses.
To ensure that the variations in support were the result of the treatment, we created a more complex model and included additional variables that could impact the relationship. Specifically, we tested the effects of gender, education level, frequency of attendance at religious services, news consumption, international news consumption, political party, familial connections to Eastern Europe, and race (White vs non-White), alongside the treatment variables. We first tested these variables in relation to support for military aid, but once again found that the treatments had insignificant effects. At this point, we decided to focus primarily on support for humanitarian aid.
In the humanitarian aid multivariate regression model, the international security treatment remained significant (p=0.0265), leading to higher support for humanitarian aid. However, other variables also proved to have a significant effect. In particular, we found that holding other factors constant, individuals who consumed more news were more likely to support humanitarian aid. In addition, if those news sources were international news sources, the individual was even more likely to support humanitarian aid. Lastly, both race (White vs non-White) and political party affiliation played a role in support for humanitarian aid. Holding other factors constant, White respondents were significantly more likely to support humanitarian aid compared to their non-White counterparts. Similarly, as shown in Figure two, holding other factors constant, both Independents and Democrats were significantly more likely to support humanitarian aid than Republicans.
Democrats were particularly likely to support humanitarian aid to Ukraine (p=0.000). In fact, compared to a Republican respondent, a Democratic respondent landed almost an entire point higher on the scale of support for humanitarian aid while holding the other variables constant. Republican respondents landed at 2.48, between somewhat opposing aid and somewhat supporting aid, while Democrats landed at 3.41, between somewhat supporting aid and strongly supporting aid. Interested in further exploring these results, we decided to return to our basic model, but look only at Democrats or Republicans at one time.
As shown in Figure 3, among Democrats, both the international security and humanitarian treatments increased support for humanitarian aid at a statistically significant level compared to the control. Although the international security treatment increasing support for humanitarian aid was not part of our initial expectations, the significance of the humanitarian treatment’s effect on support for humanitarian aid provides evidence in favor of Hypothesis 2. However, neither treatment had this effect on Republicans (see Figure 3) at the level of statistical significance, which again contradicted our expectations. This suggested to us that the Democrats in the survey may have been more responsive to our treatments and thus potentially to media framing on the Ukraine war.
Wanting to better understand the magnitude of these results, we took our analysis one step further. In Figure 4, we estimated a new model to test the effect of the interaction between our treatment and political party affiliation on support for humanitarian aid.
Our final model showed us that across all treatments, Democrats support humanitarian aid at significantly higher levels than Republicans. Peculiarly, independents appear to have opposite results of both Democrats and Republicans, with support for humanitarian aid highest among those who received no treatment and lowest among those who received the international security treatment. Independents and Republicans only differ significantly from each other within the control group, but otherwise do not. In-line with earlier results, Republicans in the control group have the lowest level of support for humanitarian aid (closest to somewhat opposing) while Democrats in the military group have the highest level of support for humanitarian aid (closest to strongly supporting).
To conclude, our insignificant results regarding support for military aid lead us to reject Hypothesis 1. Meanwhile, Hypothesis 2 requires nuance, as the significant positive effect of the humanitarian treatment on support for humanitarian aid is confirmed only among Democrats. Lastly, we failed to predict the effect that the international security treatment would have on support for humanitarian aid. Among the full sample and among only Democrats, the international security treatment had a significant positive effect on support for humanitarian aid.
Discussion, Limitations, and Next Steps
Although the magnitude of the effects were modest, our experiment indicates that an international security frame is more likely than a humanitarian one to increase support for providing funding to Ukraine–at least among Democrats. It is unintuitive that an international security frame would increase support for humanitarian aid, but it seems possible that the international security frame caused respondents to feel more strongly that the war is dangerous and unacceptable, which translated for support for the Ukrainian cause more generally. Given that both humanitarian and international security concerns about the war are heavily discussed in the media as well as among advocates for supporting Ukraine, it is a notable finding that the international security frame seemed to shift attitudes more than the humanitarian one.
There are a few ways that we or other researchers could improve upon our experiment. First, because the control article contained most of the same information that the treatment articles did, it is possible that the control article had a significant impact on support for aid to Ukraine as opposed to a hypothetical group that received no information. Providing even the control group with information about the war allowed us to more precisely isolate the frames we were interested in, but adding a fourth control group that received no information would allow us to identify the impact of the base article alone. Additionally, a nationally representative sample would have allowed us to be more confident about the external validity of our results. It is likely that a representative sample would have lessened the impact of the international security message, as it had the largest effect on Democrats and Democrats made up far more of the overall sample than they do of the overall population. A representative sample would also make it easier to explore disparate effects of the different frames on different demographics, particularly party identification. Our results suggest that media framing of the war in Ukraine is more impactful in shifting Democrats’ opinions, a result that could be a byproduct of higher media trust among Democrats. A study with more Republican respondents could help shed some light on this interesting finding.
Future research could test different types of media such as video or images. It is possible, for instance, that images illustrating the humanitarian situation caused by the war would have made the humanitarian treatment more impactful in changing opinions. Finally, a similar experiment conducted about an issue much less well known than the war in Ukraine might have very different results, as the salience of the war in Ukraine is an outlier when it comes to foreign policy issues. Analyzing the effect of a humanitarian vs. international security frame on support for providing aid in a less well known conflict could demonstrate a lot about how and whether prior knowledge and opinions moderate the impact of framing when it comes to foreign aid.
