Farmers’ Reactions to the US-China Trade War: Perceptions Versus Behaviors

Analyzing a 2019 Midwest farmer survey, we study how political inclinations affect farmers’ perception of and response to the US-China trade war. Results show that frequent use of conservative (liberal) media is associated with a 2.3% decrease (2.4% increase) in farmers’ expected income loss from the US-China trade war and a 14.3% increase in the probability of perceiving Market Facilitation Program payments as helpful. Viewers of different media sources disagree on facts, including the level of China’s retaliatory tariffs and the share of soybeans exported to China. However, we find little association between media exposure and farmers’ economic decisions that mitigate the risk of a price decline. When substantial financial interest is at stake, partisan bias is present in statements about economic facts and perceptions, but not in economic decisions. The discrepancy shows that retaliatory tariffs have little effects on farmers’ economic behavior but may affect their political behavior.


I. Introduction
Since 2018, the US-China trade war has created waves of global trade tensions, triggered record tariff increases across sectors, and reversed several decades of globalization (Fajgelbaum et al. 2020;Li, Balistreri, and Zhang 2020;Bureau et al. 2018;Elobeid et al. 2021). US farmers, a group with outsized political influence relative to their number (Anderson, Rausser, and Swinnen 2013), became a crucial force in the trade war. Both superpowers seek to influence US farmers through economic incentives, with China imposing several waves of retaliatory tariffs on US agricultural exports (Bown and Kolb 2021) and the United States subsidizing farmers with Market Facilitation Payments (MFP) (Balistreri, Zhang, and Beghin 2020;Glauber 2019).
Although most US farmers support the Republican Party and the Trump administration (Wilson 2020), changing economic incentives puts political alignments to the test. There is a need to better understand farmers' perceptions of the trade war in order to gauge whether China's retaliatory tariff undermines their support for the US trade policy. Other than perceptions, there is also a need to understand whether farmers actual economic behaviors change in response to the trade war.
Farmers' perceptions about the trade war and their behavioral responses are likely to be affected by partisan biases. Longstanding literature has documented how the "partisan screen" filters realities (Campbell et al. 1960, 133). In other words, people's perceptions of realities (Flynn, Nyhan, and Reifler 2017;Jerit and Barabas 2012;Jerit and Zhao, 2020;Taber and Lodge 2006), including economic conditions (Bartels 2002;Evans and Andersen 2006;Evans and Pickup 2010;Stanig 2013;Wlezien, Franklin, and Twiggs 1997), differ depending on whether the party they support presides over the economy. Some studies find the partisan bias can be reduced when economic facts are clear enough (Redlawsk, Civettini, and Emmerson 2010) and when rewards and admonitions are provided for accurate answers (Bullock et al. 2015;Peterson and Iyengar 2021;Prior, Sood, and Khana 2015), suggesting that behavioral responses may differ from stated perceptions. A limited number of studies (Gerber and Huber 2009;Gerber and Huber 2010;McGrath 2017) concerning people's actual economic behaviors find that people do not always behave according to their reported perceptions, suggesting that people reach survey answers and behavioral decisions through different processes.
Because existing studies primarily focus on the perceptions and behavior of the general public related to general economic conditions, it is unclear what happens when people's financial interest is directly affected, like US farmers in the trade war. When substantial financial interest is at stake, it is likely that people's information search and reasoning will be more rigorous and impartial. Would farmers' economic perceptions still be biased by their political views? Would they act upon their stated perceptions? These questions are important because farmers' support determines the sustainability of US agricultural trade policies, and their economic decisions have effects on food supply.
The US-China trade war offers a fortuitous scenario to answer the more general question of whether survey responses to perception questions reflect sincere beliefs (Bartels 2002;Jerit and Barabas 2012;Shapiro and Block-Elkon 2008) or partisan cheerleading (Hamlin and Jennings 2011;Schuessler 2000;Bullock and Lenz 2019). First, China's retaliatory tariffs and the United States' MFP created large and relatively predictable shocks on farmers' incomes. Therefore, one can consider the differences in perceived trade war impacts as the results of different political attitudes after controlling for crop production and other relevant farm characteristics. Second, farmers' perceptions of future price trends have clear behavioral implications (Choi and Helmberger 1993;Shonkwiler and Maddala 1985, also, see Appendix): if farmers expected soybean price to decline due to the trade war, they should reduce planting acres (i.e., the law of supply), sell the product early by decreasing storage and increasing pre-harvest sales (Kadjo et al. 2018), and use non-spot market tools (such as futures and options) to hedge against downward price risk (MacDonald 2020). These expected responses allow us to examine the consistency between stated perceptions and behavior.
This study focuses on how farmers' economic perceptions and behavior differ by consuming liberal or conservative media. Given the increasingly close connection between political alignment and media consumption, we interpret the effects of media exposure as evidence for how political alignment affects perceptions and behavior. This interpretation is bolstered by the fact that most farmers either use liberal or conservative media exclusively as their most frequent sources in our data. 1 We do not attempt to distinguish whether the source of political bias is media influence or audience with pre-existing biases sorting into different media outlets. Instead, we focusing quantifying the result of such partisan bias.
We collected our data from a 2019 survey of 472 crop farmers in the Midwestern states of Iowa, Illinois, and Minnesota. These three states are the top three soybean-producing states in the 1 In recent years, the media landscape in the United States has become increasingly polarized (Prior 2013), with audiences selecting media sources that conform with their views (Peterson and Iyengar 2021). As a result, people with similar political inclinations increasingly congregate around the same media sources. At the same time, liberal and conservative media cast the same events, including the trade war, in drastically different lights, potentially strengthening the partisan divide (Hoewe and Peacock 2020;Jamieson and Cappella 2008). country, accounting for 16.4% of US agricultural cash receipts and 11.1% of US corn and soybean exports in 2019 (NASS 2020). To examine how farmers' perceptions of and responses to the trade war differ by media exposure, we first classify all self-reported media outlets into three categories-conservative (e.g., Fox News), liberal (e.g., MSNBC), and neutral (e.g.,

Successful
Farming magazine)-based on the ideological placement of an information outlet's audience by the Pew Research Center (Mitchell et al. 2014) and the expert opinions of extension farm management specialists. We examine the role of exposure to these media channels with varying ideological leanings in three sets of outcomes: (a) expected loss of income from the trade war and perceptions of the helpfulness of the MFP; (b) knowledge about China's tariff rate on US soybeans, the share of soybeans exported to China, and MFP payment rate for soybean producers; and, (c) economic decisions about soybean storage, planting, and marketing.
We report three findings. First, political bias does affect economic perceptions. Frequent exposure to conservative media is associated with a 2.3% decrease in farmers' expected income loss, and frequent exposure to liberal media is weakly (p<0.1) associated with a 2.4% increase in farmers' expected income loss. The implied gap in expected income loss is 4.7% between farmers who only list liberal and those who only list conservative media in their top three sources. Given that farmers on average estimate a 14.4% expected income loss, the equivalent of $94,445 per year, the gap of 4.7% equals $30,810 per year. USDA estimates show that the heartland region's average net farm cash income is $110,800 in 2018. Therefore, the estimated income loss gap of $30,810 is economically large in magnitude. 2 Also, frequent exposure to 2 We only sampled farms with more than 250 acres of land in operation, which accounted for around 66% of farms and 90% of harvested cropland in the three sampled states. conservative media is associated with a 14.3% increase in the probability of farmers perceiving MFP payments as helpful, and exposure to liberal media is associated with a 7.4% (statistically insignificant) decrease in the probability of farmers perceiving MFP payments as helpful. In other words, exclusive conservative media consumers are 21.7% more likely to find MFP payments helpful than exclusive liberal media consumers.
Second, the differences in economic perceptions extend to disagreements on basic facts.
When asked about China's tariff rate on US soybeans (25%) and the percent of US soybean exports exported to China (60%), farmers who are frequently exposed to conservative media give answers that are on average 1.4% and 2.5% lower than others (p<0.1), and are 8.4% less likely to overestimate either the tariff rate or export share (p<0.05). In other words, the consumers of conservative media believe in facts that diminish the impacts of the trade war.
Third, there is little association between media exposure and farming and marketing behavior.
For soybeans, the most affected commodity, neither liberal nor conservative media has any statistically significant impact on surveyed farmers' storage, planting, pre-harvest versus postharvest marketing decisions, and their utilization of spot versus non-spot markets. Though imprecise estimation may cause the lack of behavioral responses in individual outcomes, the null results in all outcomes are evidence that behavioral responses are at least small, if not nonexistent.
This article relates and contributes to two significant lines of literature. First, it contributes to the literature on the determinants of trade policy preference (Blonigen, 2011;Rho and Tomz, 2017;Fordham and Kleinberg, 2012), in particular the impact of economic interest on trade preference (Fordham and Kleinberg, 2012;Rho and Tomz, 2017), by providing evidence that media also affects individuals' trade policy preference. Second, we add to the previous literature on partisan bias in economic perceptions (Bartels 2002;Evans and Andersen 2006;Evans and Pickup 2010;Stanig 2013;Wlezien, Franklin, and Twiggs 1997) by showing that the partisan screen exists even when a policy directly affects individuals' economic conditions. These partisan differences in economic perceptions partially stem from disagreements in basic facts.
Third, our article contributes to the literature on political bias and perception-behavior (in)consistency (Gerber and Huber 2009;Gerber and Huber 2010;McGrath 2017). We find no statistically significant evidence that political bias extends to actual behavior, suggesting that cheerleading does not extend to actual behaviors.

II. Theory and Hypotheses
The literature has attributed partisan bias to three sources. First, people with different political affiliations selectively consume and absorb (Jerit and Barabas 2012, Kim and Kim 2021, Mitchell et al. 2014 information, especially under a polarized media environment. Second, people may process the same information differently based on their motives. The theory of motivated reasoning (Kunda 1990;Bullock and Lenz 2019) suggests that reasoning may be driven by "accuracy goal" or "directional goal." Reasoning with the accuracy goal seeks to find objectively accurate answers. In contrast, reasoning with a directional goal tries to reach conclusions in a particular direction. Third, people may engage in "cheerleading," which means they provide answers favorable to their political party but not based on any reasoning process (Bullock et al. 2015;Peterson and Iyengar 2021). When cheerleading, survey respondents do not believe in their answers, unlike motivated reasoning with directional goals.
This study considers two sets of survey statements: 1) perceptions about the trade war, including the impacts of tariffs on income and the helpfulness of the MFP program; 2) knowledge on basic facts, namely the share of soybean exported to China, the level of retaliatory tariffs on soybeans applied by China, and the level of MFP payments. If there is partisan bias in information, directionally motivated reasoning, or cheerleading, we hypothesize that consumers of conservative (liberal) media should report lower (higher) negative impacts from tariffs (H1a), more (less) helpfulness of MFP (H1b), lower (higher) soybean export share to China (H1c), lower (higher) tariff rates on soybeans (H1d), and higher MFP payment rates (H1e). The corresponding null hypotheses (H0a~H0e) are for conservative and liberal media consumers to have the same perceptions on these issues relative to non-consumers.
If farmers expect soybean price to decline due to the U.S.-China trade war, they should take measures to mitigate the risk of the price decline. These measures range from the straightforward reaction of decreasing soybean acres to the more sophisticated methods of hedging using futures and options (see Appendix for an introduction to these measures). First, if farmers believe that the trade war has a larger negative impact on the profitability of soybean production (combining tariff and MFP impacts), they should decrease soybean planting according to the law of supply (H1f, Choi and Helmberger 1993). Second, if farmers are pessimistic about the price of soybeans, they should sell the products early by reducing storage (H1g) and increasing preharvest sales (H1h) as opposed to post-harvest sales (Kadjo et al. 2018). Finally, if farmers expect the trade war to create downward price risks, they should reduce the risk through preharvest sales and other non-spot market tools (including futures and options) (H1i, MacDonald 2020). The null hypotheses (H0f~H0i) are for conservative and liberal media consumers to exhibit the same behavior in the above aspects relative to non-consumers.
Rejections of any null hypotheses (H0a~H0i) would indicate that at least one of the three sources of partisan bias is at work, but it is unclear which one. Failures to reject would indicate a lack of support for partisan bias from all sources. Inconsistencies in partisan bias in perceptions, knowledge, and economic decisions are evidence for partisan cheerleading and against motivated reasoning.
A limitation of this study is the reliance on media consumption to classify political alignments. However, we believe this limitation is inconsequential to our key results. First, because of the highly polarized media landscape in the United States, media consumption is a good proxy for political alignment. We find empirical support for that in our data. Second, on average, people aligned with republican and democratic parties also have different media exposures (Mitchell et al. 2014). Therefore, even if we have explicit measures of political alignments, the observed partisan bias would not be free from media effects. For these reasons, we do not expect qualitatively different results if we use direct measures of political attitudes.

Survey
From March to June 2019, we sent both mail and online surveys to 3,000 crop farmers over the age of 18 with at least 250 acres of cropland in Iowa, Illinois, and Minnesota. 3 We selected 3 The average farm size in the US is 444 acres in 2020. Our sample selection of farms with at least 250 acres might limit the generalizability of the results to medium to large farms and we should be cautious extending our findings to small farms. We choose this sampling criteria to respondents through stratified sampling. Forty-four percent of our sample came from Iowa, 32% from Illinois, and 23% from Minnesota. The survey asked about farmers' demographic and farm characteristics, most frequently used media sources, expected farm income loss from the trade war, perceived helpfulness of the first round of MFP payments in 2018, and various farming and marketing decisions. We received 722 responses (a 25.8% response rate). After dropping respondents who did not provide expected income loss from the trade war (a main outcome of interest) and other important farm characteristics, 472 usable observations remained. 4 Figure 1 shows the county locations of surveyed farmers' primary farm operations and county-level soybean planted acres in 2018. 5 exclude recreational and part-time farmers for whom economic considerations may be less important.  The survey allows respondents to answer less than three sources. We classify the reported media outlets into three categories-conservative, liberal, and neutral. We first classify conservative and liberal media sources based on the ideological placement of each media outlet's audience from a study by the Pew Research Center (Mitchell et al. 2014). For local and farm-related media sources not covered by the Pew Research Center study, we determine the liberal or conservative classification based on the expert opinions of farm management specialists from Iowa State University Extension. We categorize media sources not covered by the Pew Research Center study and not recognized by experts as partisan as "neutral" media sources. 6 We conduct robustness checks using Fox News as the only conservative outlet and reached similar conclusions. Table A1 in the Appendix shows farmers' most frequently used conservative news source is Farm Bureau publications (32.6% of all farmers), closely followed by Fox News (28.6%). CNN is the most frequently used liberal source (10.8%), and Successful Farming magazine is the most frequently used neutral source (31.6%).

Variables
Question Answer choices Media exposure

Media exposure
When seeking information about the trade disruption, what are your three most frequently used media sources?

Expected income loss
Before receiving trade assistance from the USDA, to what extent do you think your farm's net income in 2018 was affected by the trade disruptions? Categorical variable from 1 (Up more than 20%) to 9 (Down more than 20%) Belief that the Trump administration's $12 billion trade relief plan will be beneficial to your farm How helpful do you think President Trump's $12 billion trade relief plan will be to your farm?
Categorical variable from 1 (Not at all helpful) to 5 (Very helpful)

Knowledge of the US soybean market
To the best of your knowledge, what percent of tariff did the Chinese government impose on US soybean exports in July 2018? The first set of outcomes we examine is farmers' expected income loss and perceived benefits from the MFP payments. The expected income loss from the 2018 trade war is a categorical variable scaling from 1 (Down more than 20%) to 9 (Up more than 20%). To gauge the accuracy of farmers' expected income loss, we also estimate actual income loss using two alternative specifications proposed by Janzen and Hendricks (2020) and calculate the gap between estimated loss and self-reported expected loss. We measure the benefit of MFP payments on a five-point scale, from "Not helpful at all" to "Very helpful." The second set of outcomes includes farmers' knowledge of Chinese tariffs on US soybeans, China's share in US soybean exports, and the level of MFP payments for soybeans. The knowledge questions have choices that range from too low to too high. The third set of outcomes involves farmers' decisions regarding soybean storage, planting, and marketing. Marketing includes the timing of sales (pre-, at-, and postharvest) and the use of spot vs. non-spot marketing tools (e.g., futures, options, and other grain contracts). 7 We measure whether farmers increase their soybean storage on a five-point scale, from "Decrease a lot" to "Increase a lot." We measure farmers' soybean planting behavior using their share of soybeans in total planted acreages. Marketing behavior include the shares of soybeans marketed in spot and non-spot markets, and pre-, at-, or post-harvest. Table 2 presents summary statistics of the main variables used in our analysis. Farmers sought information about the US-China trade war mainly from neutral media (55.5%), followed by conservative (53.0%) and liberal (24.2%) media. 8 Among the 472 participants, only 49 (10.4%) use liberal and conservative media simultaneously, and 66.7% use liberal or conservative media exclusively as their most frequent sources. The segregation of audiences supports our interpretation of media exposure as a proxy for political bias. The survey asks for average expected income loss as a categorical variable, from 1 (Up more than 20%) to 9 (Down more than 20%). When we convert the scale variable to the mean of the upper and lower bounds that define each category, 9 the average expected income loss is 14.4%.  8 Farmers' use of media with a particular ideological leaning may not be exclusive. They can get information from conservative, liberal, and neutral media or any combination of the three. 9 We code scale 1 (up more than 20%) as -25%, scale 2 (up 10-20%) as -15%, scale 3 (up 5-10%) as -7.5%, scale 4 (up less than 5%) as -2.5%, scale 5 (no change) as 0%, scale 6 (down less than 5%) as 2.5%, scale 7 (down 5-10%) as 7.5%, scale 8 (down 10-20%) as 15%, and scale 9 (down more than 20%) as 25%. These conversions are not accurate reflections of farmers' expected income loss and would likely bias our estimates of the impacts of media. To compare farmers' expected income loss with actual income loss, we follow Janzen and Hendricks (2020) and estimate farmers' income loss using two methods. Depending on the calculation method used, the estimated average actual loss is 11.2% or 16.7%. The average perception of whether MFP is helpful is 3.6 on a scale from 1 (Not at all helpful) to 5 (Very helpful), with 39.6% of farmers saying it is somewhat helpful, 21.2% of farmers saying it is quite helpful, and 27.8% of farmers saying it is very helpful. In the remaining analysis, we aggregate these three categories as helpful and the remaining two categories (Not at all helpful and Not sure) as not helpful.

Summary Statistics
We also study farmers' beliefs about basic underlying facts that may affect their perception of trade war impacts. Regarding China's retaliatory tariff rate on US soybeans, 64.2% of farmers answered the question correctly, and 21.3% and 14.5% chose numbers that are too low and too high, respectively. Regarding what percentage of US soybean exports go to China, 34.3% of farmers answer the question correctly, and 58.0% and 7.7% of farmers underestimated and overestimated, respectively. On the MFP payment rate for soybean producers, 92.3% of farmers answered the question correctly ($1.65/bushel), and 7.7% of farmers underestimated it.
In 2018, respondents planted an average of 497 acres of soybeans and 594 acres of corn. On average, the amount of soybeans respondents stored increased in 2018. Farmers sold averages of 46.4% of their soybeans pre-and at-harvest and 53.6% post-harvest. Furthermore, respondents sold 53.8% of soybeans in the spot market and 46.2% in the non-spot market, including futures, options, and other grain contracts.

Econometric Model
The econometric model we use to test the role of exposure to conservative, liberal, and neutral media in economic perceptions and farming behavior is: where denotes the outcome of interest; , , and s are the indexes for individuals, counties, and states, respectively; and, , , and represent farmer ′s exposure to conservative, liberal, and neutral media, respectively. In the main analysis, we use dummies to measure media exposure. Specifically, , equals one if a farmer listed at least one conservative (liberal, neutral) media outlet as a frequent information source, zero otherwise. Exposure to conservative, liberal, and neutral media is not mutually exclusive.
We also check the robustness of the results using the share of different media types as an alternative measurement of media use.
To alleviate the concern of omitted-variable bias, we include a rich set of control variables, , that include farmer demographic characteristics and farm characteristics. Demographic variables include farmers' income, age, education, and gender. Farm characteristics include 2018 soybean and corn production (calculated using farmers' 2018 planted acreage and county-level yield), whether the farmer has livestock, whether the farmer has an off-farm job, and the cash rent for that farm. We estimate cash rent by multiplying the county-level cash rent for nonirrigated cropland by the share of rented land. We include state fixed effects, , to capture time-invariant differences across states. We cluster the error term ( ) at the county level to allow for error correlation between observations within a county.
The outcomes include both continuous and categorical variables, and we choose econometric models accordingly. We use Ordinary Least Square (OLS) regression or continuous variables, interval regression (Billard and Diday 2000) for categorical variables with known cutoff points, and the probit and ordered probit models and for ordinal variables. Average marginal effects are reported for probit and ordered probit models.

Farmers' Actual Income Losses
To provide a benchmark for farmers' losses from the trade war, we follow Janzen and Hendricks's (2020) two methods of estimating actual losses from soybean and corn sales. 10 where and denote farmer ′ soybean and corn harvested area in 2018, respectively; and, and denote the soybean and corn yield, respectively, in county .
Yield data is from the USDA National Agricultural Statistics Service (NASS 2020).
The second method uses the decrease in unit export value from before the trade war (2017/18) to after it started (2018/2019) to measure farmers' losses from the trade war. The unit price of US soybean exports to China declined by $1.38/bushel, and that for corn declined by $.01/bushel. We calculate the second measurement of real income loss as: The notations in equation (3) are the same as in equation (2). To investigate whether media is associated with the gap between farmer's expected and actual income loss from the trade war, we construct the following measurement: where denotes farmers' self-reported percentage-of-income impact from the trade war; and, * 100% denotes estimated actual percentage-of-income impacts from the trade war.
USDA provided two rounds of MFP payments to farmers in 2018 and 2019. Our survey questions referred only to the first round (2018). 11 The 2018 MFP payment rates were $0.01/bushel for corn and $1.65/bushel for soybeans. As the estimated actual losses according to the second method are higher than farmers' expected income loss, it is possible that farmers overlooked survey instructions to the contrary and included the MFP payments into their reported expected income loss. Therefore, we also calculate farmers' MFP payments in 2018 using their corn and soybean production and the corresponding payment rate. We present results excluding and including MFP payments when analyzing the gap between expected and actual income loss.

Perceived Economic Conditions
To visualize the relationship between soybean production and farmers' perceived economic conditions, we present the LOWESS (Cleveland 1981)  than farmers who are conservative media consumers. Figure 2.2 shows that given the same level of soybean production, farmers exposed to conservative media are more likely to believe that MFP payments are helpful than farmers exposed to liberal media. Furthermore, farmers with more soybean production express higher levels of expected income loss and more helpfulness of MFP, which is expected since both China's retaliatory tariffs and MFP payments target soybean.  (2) and (3) show the media's association with the gap between farmers' expected and actual income losses as measured by two alternative methods. Columns (4) and (5)  Log of average soybean production in 2013-2017 (Bushel) Exposed to conservative media Exposed to liberal media .8 Probability that MFP payments are helpful 2 4 6 8 10 12 Log of average soybean production in 2013-2017 (Bushel) Exposed to conservative media Exposed to liberal media  (2) and (3) present the OLS results for the gap between expected and actual income loss as measured by equations (2) and (3). Columns (4) and (5) present the OLS results for the gap between expected and actual income loss when we include MFP payments in farmers' actual income loss to account for the possibility that farmers might unconsciously account for the 26 MFP payments when reporting their expected income loss. Column (6) shows the marginal effects from probit estimation results of media exposure on farmers' beliefs as to whether MFP payments are helpful. We include state fixed effects in all specifications and cluster standard errors at the state level when using OLS. Standard errors are in parentheses. *, **, and *** denote significance level at the 0.1, 0.05, and 0.1 level, respectively.
We observe the following results. There is strong evidence that conservative media exposure relates to significantly lower expected income loss, while liberal media exposure has a positive association (p<10%) with expected income loss. In column (1), interval regression results show that farmers frequently exposed to conservative media report a 2.3% lower expected income loss.
Given that survey participants reported an average gross income of $655,551 in 2018, the 2.3% decrease in expected income loss would mean a decrease of $15,005, which is economically significant. 13 Columns (2) and (3) show that when we measure the actual loss using the two methods proposed by Janzen and Hendricks (2020), we find a negative association between conservative media exposure and the gap between expected and actual income loss (ranging from 2.2% to 2.6%, with significance level ranging from 5% to 10%). For liberal media exposure, column (1) shows that farmers who are frequently exposed to liberal media report a higher expected income loss (2.4%) at the significance level of 10%. In our sample, 66.7% of farmers list either liberal or conservative media as their most frequently used sources (with or without neutral sources). The implied gap in expected income loss between liberal-only and conservative-only audiences (with and without consuming neutral media) is 4.7%, or $30,811.
These findings suggest that farmers frequently exposed to conservative and liberal media substantially differ in their expected income loss. Consumers of conservative media are more 13 (1) shows, farmers who produce more soybeans expect more income loss and, as column (6) shows, are more likely to believe that MFP payments are helpful.
These results are expected considering that both China's retaliatory tariffs and the United States' MFP payments target soybeans. Despite the political bias, economic reality still shapes perceptions to some extent.
Overall, the results in table 3 show that conservative media exposure is associated with lower expected income loss and a higher probability of finding MFP payments. Liberal media exposure, on the other hand, is associated with higher expected income loss and a lower probability of finding MFP payments are helpful. These findings support hypotheses H1a and H1b. Given that farmers' media use reflects their political inclination and bias, these findings indicate that political bias is associated with farmers' economic perceptions of their income loss from the trade war and whether MFP payments are helpful. These findings add to previous studies on the impact of political bias on people's economic perceptions by showing that the partisan screen is also at work when a policy directly affects individuals' economic conditions.

Knowledge of Trade War-Related Facts
To understand why media exposure is associated with biased expected income loss, we check if media exposure is associated with farmers' knowledge of China's retaliatory soybean tariffs (25%), the share of US soybean exports shipped to China in 2017 (60%), and the MFP payment rate for soybean producers ($1.65/bushel). We use OLS regression and regress farmers' answers in percentage on media exposure and control variables. As table 4 column (1) shows, we find that frequent exposure to conservative media is associated with a 1.4% lower tariff rate estimate and, as column (2) shows, a 2.5% lower export share estimate. Both associations have p-values that are below 10% but above 5%. Because both facts affect the economic impacts of the trade war, we analyze whether farmers overestimate or underestimate either of these two facts.
Marginal effects from probit regressions show that the probability of overestimating either tariff or US export to China is 8.4% (p<0.05) lower for farmers who consume conservative media (table 4 column (3)). These findings support hypotheses H1c and H1d. The effect of conservative media on underestimation of these facts (+) and the effects of liberal media on overestimation (+) and underestimation (-) are also in the expected direction but are not statistically significant (table 4 columns (3) and (4)). Table 4 column (5) shows that exposure to conservative media is associated with a statistically insignificant 3% higher MFP soybean payment rate. Therefore, we fail to reject the null hypothesis H0e. The lack of statistical significance in MFP results is not surprising as the US government had just announced the MFP at the time of the survey, and thus the vast majority of farmers knew the correct answer. As tariff rate and export share to China positively relate to the severity of trade war impacts, these findings indicate that farmers exposed to conservative media perceive facts in a way that diminishes trade war impacts. This is a potential explanation for why media exposure is associated with biased expected income loss. Results in columns (1), (2), and (5) are coefficients from OLS regressions. Columns (3) and (4) are marginal effects from probit regressions. We include state fixed effects in all specifications and cluster standard errors at the state level. Standard errors are in the parenthesis. *, **, and *** denote significance level at the 0.1, 0.05, and 0.01 level, respectively. Table 5 and Table A3 in the Appendix present the role of media exposure on farmers' actual behavior in 2018 and planned behavior regarding soybeans in 2019, respectively. In Table 5 and Table A3, column (1) shows the probit estimation results of whether farmers reduced their soybean storage. Columns (2) and (3) show changes in the share of farmland planted for soybeans and corn, respectively. Columns (4) and (5) show changes in soybeans marketed pre-, at-, and post-harvest. Column (6) shows changes in soybeans marketed on the non-spot markets.  This table presents the impact of media exposure on farmers' soybean storage, soybean and corn planting behavior, and marketing behavior in 2018. Column (1) presents marginal effects estimated with probit, and we estimate columns (2)-(6) with OLS. We also include control variables as specified in equation (3) and omit their coefficients from the table for readability. We include state fixed effects in all specifications and cluster standard errors at the state level. Standard errors are in parentheses. *, **, and *** denote significance level at the 0.1, 0.05, and 0.01 level, respectively a. Coefficient dropped because of perfect collinearity.

Media and Behavior
The coefficients for control variables, when statistically significant, have the expected signs. For example, more soybean (corn) production leads to a significantly higher share of land allocated to soybeans (corn) (Columns (2) and (3)), which shows persistency in crop choice across years. We also find that farmers with higher risk tolerance continue to produce affected crops (Column (2)), and farmers exposed to farm-related media are more likely to use non-spot marketing tools (Table A3, Column (6)).
However, we find no statistically significant (P<0.05) association between media exposure and farmers' economic decisions related to soybean production and marketing in 2018.
While it is possible that farmers have committed to some economic decisions in 2018, media exposure also has small and statistically insignificant effect on planned behaviors in 2019. These findings fail to reject the null hypotheses H0f-H0i. Though the lack of statistical power could cause each individual behavior, the absence of statistically significant results in all behavioral aspects suggests that media exposure has a weak, if not non-existent, association with actual economic behavior. When examining the confidence intervals, we find that the effects of exposure to conservative and liberal media are practically small for most of the outcomes. Take the estimated impact of conservative media exposure on the probability of soybean storage increase in 2018 in Table 5 Column (1) as an example. The 95% confidence interval for the probability of soybean storage increase is -0.042-0.004. Thus, the relative effect could be anywhere between a 4.2% decrease and a 0.4% increase with a 95% probability.
Combining the previous findings-that political bias has a significant association with farmers' perceptions of income loss from the trade war and the usefulness of the MFP payments-with the findings from this section-that political bias has a limited role in farmers' actual crop storage, planting, and marketing behavior-our analysis shows that partisan bias is only present in stated perceptions but not in behavior.

Additional Robustness Checks
We check the robustness of our results in several ways. We first check the robustness of the results using the share of conservative, liberal, and neutral media as alternative media exposure measurements. Tables A4 in the Appendix present the estimated effects of alternative media exposure on farmers' beliefs and behavior, respectively. The main findings remain robustmedia exposure is associated with farmers' beliefs but has a limited role in farmers' behavior.

VI. Discussion and Conclusions
Based on a survey of 472 farmers in three Midwestern states, we investigate the correlation between exposure to conservative, liberal, and neutral media and farmers' perceptions, knowledge, and behavior with respect to the trade dispute between the United States and China.
Though we base our results on media exposure, we argue that the relationships between media exposure and perceptions and behavior are indicative of the relationships between political attitudes and these perceptions and behavior.
We find that exposure to conservative (liberal) media is associated with a reduction (increase) in farmers' expected income loss from the trade war and an increase (decrease) in their beliefs that MFP payments are helpful. These findings suggest that perceptions of economic conditions are still subject to the partisan screen even when financial interest is directly involved.
Furthermore, the differences in economic perspective are partially caused by disagreements about fundamental facts. People who control the means of production and whose own financial interest is at stake are more likely to seek accurate information. However, our results suggest that such information-seeking cannot eliminate the effects of partisan bias in shaping economic perceptions.
In contrast to the strong correlation between media exposure and farmers' perceptions and perceived helpfulness of government payments, we find little correlation between media exposure (and, by extension, political attitudes) and economic decisions. Though the lack of statistical power could cause the null results, the absence of statistically significant effects in all behavior aspects under study suggests that political attitudes have weak effects, if any, on economic behavior. The null results on economic behavior suggest that the potentially biased information received by conservative and liberal farmers is not the decisive factor in creating partisan bias, and farmers with different political attitudes make similar economic decisions. The inconsistency between stated perceptions and behavior here adds weight to the argument that survey responses about economic beliefs are subject to cheerleading. Our results go further than the previous literature to show that such cheerleading exists even in expectations regarding one's own economic conditions.
If partisanship does not entirely drive economic perceptions, tariffs applied by China and MFP payments applied by the United States on US farmers would have been effective. We find evidence that farmers' economic perception does depend on economic realities-that is, farmers who produce more soybeans expect heavier loss and perceive MFP subsidies as more helpful.
Therefore, our findings suggest that both economic factors and political bias shape farmers' economic perceptions, and their production and marketing behavior are less likely to be associated with political bias.
This study has several limitations, and thus future studies can improve upon ours. This study relies on the effects of media exposure to infer the effects of political attitudes on economic perceptions and behavior. As a result, the relationships we discover are qualitative. Given the imperfect correlation between media consumption and political attitudes, the magnitude of media effects is likely smaller than the underlying effects of political attitudes. In addition, we base the lack of statistically significant association between media exposure and farming and marketing behavior on a modest sample size, and future studies can use a larger sample or field experiments to explore the perception-behavior link more extensively.

Appendix: An introduction to crop planting, storage, and marketing
In this appendix, we briefly explained farmers' decisions regarding planting, storage, and the usage of crop marketing tools. The purpose of this appendix is to concretely show that these decisions, to various degrees of certainty, can be affected by farmers' outlook on downward price risks.

Planting corn versus soybeans
Corn and soybeans are the major crops for the Midwest region, accounting for 38.3% and 31.3% of the planted acres, respectively, in the three surveyed states in 2019. All farmers plant both corn and soybean at the time of our survey. Corn and soybean are substitutes in production since they compete for the same agricultural land. The decision to produce corn and soybean is driven by profit maximization (profit=total revenue -total cost). Because of the biological lag between planting and harvest, farmers use projected prices to form expectations. The law of supply dictates that lower (expected) price leads to lower planting. In our case, farmers who are more pessimistic about soybean prices should decrease soybean acres.
Practically, the adjustment of the percentages of corn and soybean acres can be achieved through switching between two common crop rotation patterns: continuous corn and corn-soybean rotation (continuous soybean is also possible but may cause decreasing yield, Licht et al. 2021). The corn-soybean rotation saves fertilizer costs since soybean retains Nitrate, but it may deplete organic matters in the soil (Hall, Russell, and Moore 2019). If a farmer decides to decrease soybean planting, they can take acres out from corn-soybean rotation and plant corn for another year. Historically, from 2000 to 2020, the ratio between corn and soybean acres in the three sample states fluctuated between 1.02 to 1.54, which attests to the adjustability of planting decisions.

Storage decisions
Grain storage after harvest is a common strategy adopted by farmers in the Midwest (Edward and Johanns 2018). For example, the grain storage capacity in Iowa in 2018 was 1.45 billion bushels, or 47.5% of the total corn and soybean production in that year. The primary reason for farmers to store grain is to capture price improvement after harvest (Dhuyvetter et al. 2007). In most years, the prices for corn and soybeans are at the lowest after harvest and gradually increase until close to the next harvest season. Other factors, including the trade war, will also affect the price trend, hence farmers' storage decisions.
When deciding how much grains to store, farmers weigh potential price gains against the cost of grain storage. Farmers can choose to store grain on-farm or in commercial grain elevators. Both options come with costs that increase with the quantity and duration of storage. For example, storage would postpone sales and delay loan repayment, creating interest costs. Other things equal, if farmers expect the future price to be lower than in a normal seasonal cycle, they would decrease storage since there is less price gain to justify the storage cost. Therefore, farmers who believe the trade war have a more negative impact use less grain storage. Grain storage facilities usually have redundant capacity (Janzen 2020), which gives farmers the flexibility to change storage levels.

Pre-harvest marketing
As discussed in the storage section, when expecting lower prices in the future, farmers would reduce storage, thus selling more grains earlier at harvest. However, a farmer can sell their crop even earlier through pre-harvest marketing. Essentially, farmers can promise to deliver their stillgrowing crop at a pre-determined price to local grain merchants (Johnson 2018).
The cash price that farmers get when selling grains to a local merchant can be written as:

Cash price=futures price + basis
Where basis is defined as the difference between the futures price (determined by the national market) and the local price, which is determined by transportation costs, local storage availability, and so on. Two most commonly used tools for pre-harvest marketing are cashforward contracts and hedge-to-arrive contracts (Johnson 2020). In cash-forward contracts, the farmers are directly offered a cash price to deliver crop at a later time. In hedge-to-arrive contracts, only the futures price component is fixed, and the basis can still vary according to the market. If farmers expect downward price trends, they should use one of these tools. Which one to choose depends on the expectations about the basis, which has no obvious relationship with the trade war.

Spot vs. Non-spot market
The spot market is where commodities are traded, and payments are made at the time of the transaction. Therefore, the tools for pre-harvest marketing all fall under the umbrella of the nonspot market. In addition to pre-harvest marketing using cash-forward contracts and hedge-toarrive contracts with local grain merchants, common non-spot market tools include futures and options. Farmers commonly use these tools to hedge against downward price risks (Prager et al. 2020, CME 2020. While these instruments can be used for purely speculative purposes, our survey instrument specifically asks for the usage of the non-spot market for grain marketing. Simply put, a futures contract is the commitment to deliver goods at a specific time in the future (say December 2022) in exchange for payment (according to futures price) right now. To hedge against downward price risk, a farmer who has a crop in the field can sell futures now, get paid by the current futures price, and commit to delivering the crop later. When the time comes to deliver, it is possible for the farmer to physically deliver the crop. However, to reduce costs, most farmers would buy futures (releasing them from the duty to deliver). As they buy futures, they would also sell their crop on the cash market. The cost of buying futures and selling the actual crop after harvest will more or less cancel out (up to basis, see formula above). The farmer essentially sells the crop at a fixed earlier price using futures hedging.
Call and put options are derivative products of the futures market. They are the rights (but not obligations) to buy and sell futures at certain prices. Compared to hedging with futures, which lock in a certain price, hedging with options retains the upward potential when the price increases. A crop producer can buy put options, sell call options, or use a combination of the two to protect themselves from price decline (CME 2020). These strategies can be used before harvesting or in the storage stage. A farmer who assesses downward risk to be higher due to the trade war would use non-spot market marketing tools more.  types to measure media exposure. Column (1) presents marginal effects estimated with probit, and we estimate columns (2)-(5) with OLS. We also include control variables specified in equation (3) and omit their coefficients from the table for readability. Standard errors are in the parenthesis. *, **, and *** denote significance at the 10, 5, and 1% levels, respectively.