Tag: Raphael Thomadsen

During the 2020 election cycle, presidential candidates spent nearly $3 billion on television, radio and digital ads—shattering records and demonstrating how important advertising is to campaign strategy. Given the amount of resources dedicated to advertising, understanding how messages influence voter behavior is critical to campaigns.

New research from Olin Business School is shedding light on how slant—the extremeness of the message—and consistency with the candidate’s primary campaign messaging in national television advertisements affected voter behavior during the 2016 presidential election, specifically online word-of-mouth chatter and candidate preference in daily polls.

More than 800 national ads

With the help of recent advancements in text analysis methods, researchers conducted an extensive review of more than 800 national television ads that ran from June through November 2016. The study results were published on Jan. 28 in Quantitative Marketing and Economics by Raphael Thomadsen, professor of marketing at Olin; Donggwan Kin, a PhD candidate at Olin; Beth L. Fossen, at Indiana University; and David A. Schweidel, at Emory University.


“Slant and consistency are two vital dimensions related to the branding of political candidates, with slant representing what the candidate stands for, and consistency representing the extent to which the candidate creates a clear and repeated message of what they represent, which creates the branding of the candidate,” Thomadsen said. 

Their findings challenge conventional campaign wisdom, which suggests that candidates should moderate their positions and become more centrist after winning the party’s nomination. 

Looking specifically at Twitter trends, the researchers observed that both candidates, on average, experienced a 30% increase in online word-of-mouth chatter between the five-minute window before an ad was shown and the five-minute window after an ad was shown. However, political ads with messages that were extremely Republican or extremely Democratic decreased the volume of candidate-related word of mouth, especially in earlier stages of the campaign.

There’s a caveat to that finding, though: Centrist messages in political ads did generate more online word-of-mouth and higher daily poll ratings, but the benefit of centrism was lost if those same messages were inconsistent with the candidate’s primary election platform.

“We find that consistency with the primary message, which is generally more partisan, is also important,” Thomadsen said.  “What that means, to me, is that the branding is important and that the benefit of moving to the center can be offset — or even more than offset — by the loss of the consistency of the message that such a move necessitates.

“It also demonstrates why candidates who have stuck with more extreme messaging have not suffered as much as political scientists focused on the median voter theories would believe they would.”

‘This goes against the advice that some consultants give, which is that no one pays attention until late in the race.’

Raphael Thomadsen

The importance of both centrism and message consistency were largest in the early stages of the general election, which seems to suggest that people may be more responsive to a candidate’s messages in political ads early in the campaign, Thomadsen said

“This goes against the advice that some consultants give, which is that no one pays attention until late in the race. Our analysis suggests the opposite — the candidate’s brand is built early in the race, and then at the end things become more of a scrum for voters,” Thomadsen said.

According to Thomadsen, slant and consistency have traditionally been difficult to study on a large scale because the research was labor intensive, but new text analytics tools make the research more efficient. In the present study, authors were able to dig deeper — beyond the traditional focus on tone, source and volume of advertising — for a more nuanced understanding of how political ads impact voter behavior.‘This goes against the advice that some consultants give, which is that no one pays attention until late in the race. Our analysis suggests the opposite — the candidate’s brand is built early in the race, and then at the end things become more of a scrum for voters.’

What makes political ads effective

“This research is among the first in marketing to use text analysis to derive message-related metrics that are linked to the performance of television commercials,” Thomadsen said. “Similar approaches could be used outside of political marketing, by product and service marketers, to assess the importance of message consistency in an efficient and automated way.”

In total, the analysis included 824 ad airings for 60 unique ad creatives aired by 11 political advertisers, including campaigns, political parties and seven PACs. National television advertising buys account for more than 25% of all campaign television spending — a share that is expected to increase as the rising cost of local ad inventory in battleground markets increases, according to Thomadsen.

Slant was measured by analyzing the language used in the ad, including topics and specific word choice. Ads that featured language primarily used by one party were labeled extreme, while ads that included language used by both parties were labeled centrist. Examples of extreme messaging included ads that focused on national security, immigration, gender equality and health care. The technology also captured other dimensions, such as attack phrases frequently used by a candidate and other phases that set the tenor of their campaign.

To study message consistency, the team compared ad content to primary campaign speeches. Daily poll data on voter preferences and online chatter about the candidates on Twitter provided measures for voter behavior impact.

Clinton and Trump

Altogether, Hillary Clinton had more ad airings, while ads supporting Donald Trump had larger audience sizes. Ads supporting the two candidates were comparable in tone, length and ad position.

While most of the national ads were fairly centrist, Trump’s ads tended to be somewhat more centrist than Clinton’s ads. Some of Trump’s ads leaned toward Democratic ideology, such as his promised support for gender equality, including equal pay and support for child care.

Likewise, some of Clinton’s leaned toward Republican ideology, including ads that discussed threats from nuclear weapons or the Islamic state. Overall, Clinton’s ads had a higher level of consistency than Trump’s.

“Trump’s presidency was more conservative than his 2016 campaign, so we remember those aspects of his messaging more,” Thomadsen said. “His first campaign had a lot of messages for the center, or even the left, of the country. He advocated for family leave, for example.”

Taken together, “the results suggest that it would be advantageous for candidates to adhere close to their primary campaign messages in the early stages of the general election and emphasize moderate or time-specific messages as the election further develops,” the authors wrote.  

“Further, the results suggest that the rising use of extremist messages in political advertising may be a flawed strategy for candidates that could decrease candidate-related word-of-mouth volume and voter preference for the candidate.”

Even before COVID-19 and resulting shutdowns created gridlock for some global supply chains, the assortment at many neighborhood supermarkets was dwindling. The cause was not a lack of supply, though, but rather a lack of demand created by a widening income gap in the United States, according to a new study involving an Olin researcher.

In a paper forthcoming in Management ScienceRaphael Thomadsen, professor of marketing at Olin, along with the University of South Carolina’s Rafael Becerril-Arreola and UCLA’s Randolph E. Bucklin find that the amount of variety available in the market is highly sensitive to the income shares of the middle and upper-middle classes.


As the middle class has been hollowed out, the assortment on grocery store shelves has, too, as a result.

“We find that as we become a more unequal society, the total set of products we have to choose from is reduced, holding average level of income constant,” Thomadsen said.

“This happens largely because as income inequality grows, the people whose income is growing do not spend much more at supermarkets. But people whose incomes are cut reduce their spending quite a bit, leaving less total spending and, thus, less support for niche products.”

Income distribution disparity

The researchers looked at the level of income distribution disparity in more than 1,700 US counties between 2007 and 2013, using the Gini income index. A Gini index of zero expresses absolute equality, where everyone has the same income. An index of 1 represents maximum inequality.

Then they overlaid that with a look at the product range available in grocery stores within each county across nearly 950 categories of non-perishable household items.

Consumer packaged goods manufacturers consistently supplied many new products over the time of the study. More than two-thirds of the counties had a widening of income inequality. In those counties — and at individual stores in those counties — the available assortment of products tended to grow 18% slower than in the counties where income inequality narrowed.

By comparison, in the counties where income inequality narrowed, there was typically more product variety added to the shelves.

The researchers also established that the effects of higher inequality on assortment could be explained by changes in the incomes earned by households in the middle class, which they defined, according to US Census Bureau data, as the third and fourth income quintiles (the 40th to 60th income percentiles and the 60th to 80th percentiles, respectively). When the middle-class share of income rose, variety went up, but when it fell, variety went down.

“The conventional wisdom is that greater inequality means that we have greater diversity of desires, and that this should increase the number of product offerings in the store. Finding the opposite effect made us think harder about how assortment is driven by income inequality,” Thomadsen said.

“A large factor is that consumers spend a smaller percentage of their income at supermarkets as their income rises, since they fulfill their needs at some point, or even move to consumption at restaurants or other options. However, we also see that greater inequality comes from the hollowing out of the middle class, so greater income inequality does not necessarily lead to greater heterogeneity — it leads to a large group of consumers with lower incomes that become more similar to each other.”

Unintended consequences

The analysis sheds light on the potentially unintended consequences of policy interventions and impact on consumer welfare.

“For example, the US government recently increased the minimum number of offerings per category required for retailers to participate in the food stamp program,” Thomadsen said. “A long-term implication of our results is that consumers’ ability to use food stamps may decline in areas where falling average income and rising income dispersion leads retailers to reduce their assortments and possibly stop participating in the food stamp program.”

Additionally, the increase or decrease in assortment that arises from changes in income distribution can impact consumer welfare. “In particular, consumers who lose access to their preferred items may be most adversely affected,” he said.

Additionally, the findings can help manufacturers understand trends. In particular, adding Gini tracking to one’s operational dashboard may be a useful insight to drive marketing efforts.

“Our results show that markets with rising income inequality experience assortment pruning by the grocery channel and markets with falling inequality see expansion,” the researchers wrote. “With data on local changes in the Gini index, managers may be better able to guide either defensive actions to retain shelf space or offensive actions to acquire it.”

President-elect Joe Biden has signaled that fighting the COVID-19 pandemic will be an immediate priority for his administration. He recently announced a coronavirus advisory board of infectious disease researchers and former public health advisers along with an updated strategy that will include increases in testing and contact tracing, as well as transparent communication.

But Inauguration Day is still a month away. The number of confirmed COVID-19 cases is likely to increase to 20 million by the end of January, nearly doubling the levels around the time of his election, an Olin COVID-19 forecasting model predicts.

The model, which accurately forecasted the rate of COVID-19 growth over the summer of 2020, was developed by Olin’s  Meng LiuRaphael Thomadsen and Song Yao. Their paper—presenting the model and its forecasts—was published November 23 by Scientific Reports.

“One of the key reasons for the increased accuracy of this model over other COVID-19 forecasts is that this model accounts for the fact that people live in interconnected social networks rather than interacting mostly with random groups of strangers,” said Thomadsen, professor of marketing. “This allows the model to forecast that growth will not continue at exponential rates for long periods of time, as classic COVID-19 forecasts predict.”

The evolution of COVID-19 depends on how much we, as a country, continue to social distance or return back to normal levels of interaction. This chart shows forecasted cases in the U.S. through the end of January 2021 based on our current social distancing levels, as well as less and more social distancing.

An interactive online version of the model also allows users to observe the impact different levels of social distancing will have on the spread of COVID-19. The current social distancing reflects an approximate 60% return to normalcy, as compared with the level of social distancing before the pandemic. If we continue, as a nation, at the current level of social distancing, the model forecasts that we are likely to reach 20 million cases before the end of January 2021.

“Even small increases in social distancing can have a large effect on the number of cases we observe in the next two and a half months,” Thomadsen said. “Going back to a 50% return to normalcy, which was the average level of distancing in early August, would likely result in 5 million fewer cases by the end of January.

“We could effectively squash out the COVID growth within a few weeks if we went back to the levels of social distancing we experienced in April.”

Raphael Thomadsen

“We could effectively squash out the COVID growth within a few weeks if we went back to the levels of social distancing we experienced in April,” he added.

However, the researchers caution that this is likely a conservative estimate due to increased testing and the upcoming holidays.

“In our model, we assume that only 10% of cases are ever diagnosed, meaning that we will start to hit saturation,” said Song Yao, associate professor of marketing and study co-author. “However, more recently, testing has increased, and probably more like 25% of cases are diagnosed. In that case, total COVID cases would increase beyond 20 million in the next few months unless we, as a society, engage in more social distancing.”

“The upcoming holiday seasons will present a great deal of uncertainty to the outlook of the pandemic as people travel more at the end of the year. This will likely make our forecast an optimistic one,” said Meng Liu, assistant professor of marketing and study co-author.