Getting a thumbs down on social media doesn’t push people away from the conversation. Instead, they tend to soften the tone and encourage you to post more. New research published in Marketing Research Journal We provide evidence that negative feedback from colleagues encourages users to stay engaged rather than retreating into isolated communities. These findings suggest that allowing downvotes on social platforms could help defuse extreme debates without silencing individual voices.
Social media platforms are continually experimenting with how users interact with and rate the posts of other users. Almost all platforms have a button to express approval, but few allow you to express negative feedback explicitly. Recently, major networks such as YouTube and X have considered adding dislike or downvote features to help regulate content.
“I started thinking about this after hearing that YouTube was hiding dislikes and that Twitter (now X) and TikTok were testing downvote features,” said Jessica Fung, an assistant professor of marketing at the University of Michigan’s Ross School of Business who will be moving to the University of Maryland, College Park. “Our review of the literature reveals that there is not much empirical evidence about the impact of downvotes on UGC. At the same time, there is a lot of discussion about polarization and social media, which has some very interesting implications, such as whether haters push users into echo chambers.”
UGC refers to user-generated content. This includes comments and posts made by the platform’s community. Managers often worry that negative feedback will push these users into isolated groups of like-minded individuals. These isolated groups are often called echo chambers and occur when people only interact with others who share their exact opinions. When individuals exist entirely within echo chambers, political and social divisions tend to increase.
To understand how negative feedback actually changes behavior in a real online environment, scientists looked at Reddit. Reddit is a large online bulletin board organized into thousands of specific communities. These communities, known as subreddits, are where users discuss everything from the latest news to niche personal hobbies. On Reddit, you can upvote a post if you like it, and downvote it if you don’t like it.
The platform calculates a visible score for every comment by subtracting downvotes from upvotes. Additionally, each user has a public reputation score called Karma. A person’s karma increases when they receive upvotes across the site and decreases when they receive downvotes. The authors chose this environment because the discussion is highly opinion-based and the feedback system is public.
By analyzing how people react when their scores drop, the researchers aimed to see if negative feedback changes the frequency of posts, the location of posts, and the emotional intensity of the writing. To investigate these questions, researchers tracked a sample of 17,525 Reddit users over 61 days. They monitored these people daily to understand their ongoing habits.
The team collected data on nearly 2 million comments across more than 32,000 different subreddits. For every comment a user made, the scientists recorded the text, the community the comment belonged to, and how its upvote and downvote scores changed over the next two weeks. Studying the direct effects of downvotes is complicated because people who post highly controversial opinions may naturally attract more negative feedback and may also post more frequently in general.
To isolate the specific effects of receiving negative feedback, scientists used a psychological concept known as left-digit bias. Left digit bias is the human tendency to pay most attention to the first digit in a sequence. For example, a price drop from $10 to $9 feels much more important than a price drop from $11 to $10.
The authors applied this concept to Reddit’s Karma. They assumed that users would notice a drop in karma more if the first digit changed, such as dropping from 101 to 99, rather than dropping from 102 to 100. Both examples represent a loss of 2 points, but the change in the first digit makes the penalty seem much larger. By comparing users who experienced this very noticeable karma drop to users who experienced a less obvious karma drop of exactly the same size, the researchers were able to measure how noticeable negative feedback changes behavior.
Scientists have found that experiencing a significant drop in karma actually makes users more likely to post again. “A downvote doesn’t silence a user. On Reddit, users who vote down actually post more afterward, not less,” Fung said. Rather than leaving the platform in dissatisfaction, users increased their overall content creation.
“We were initially surprised to find that as the number of downvotes increases, the average number of posts by a user also increases. This prompted us to look deeper into the mechanism,” Fung said. “We find that people tend to post more because they are trying to restore their reputation. On Reddit, that manifests itself in the form of karma. After they get a downvote, they tend to post more until their karma recovers to the level it was before they got the downvote.”
The researchers then looked at where these individuals chose to post after being downvoted. A common concern is that people leave communities that reject their views and seek echo chambers where everyone agrees with them. The data proves that this does not happen. The user continued to comment in the very same community where they received the negative feedback.
“A no vote doesn’t seem to create or encourage an echo chamber,” Fong said. “We found no evidence that users are abandoning communities where they receive downvotes. Users continue to engage with them, while expanding into new spaces.” This detail is very important to social media companies because it means that downvoting doesn’t automatically separate people into divided groups.
Finally, the authors analyzed how negative feedback changed the words people actually used. They wanted to know whether voting no would make people believe in more extreme views or soften their words. To measure this, scientists used machine learning language tools to scan the text of comments. This tool identifies the main topic of a sentence and assigns a strength score based on how emotional or extreme its language is.
In this part of the study, the researchers looked at what happened when a particular comment’s score dropped from a positive number to a negative number. Researchers found that if a strongly worded comment was downvoted by less than zero, users tended to soften their tone the next time they mentioned the same topic. “Downvoting tends to soften the tone of what users say next, especially if the original post was emotionally charged,” Fung said.
Although these findings present interesting perspectives on platform design, there are some limitations that should be considered. “Our research is a case study on Reddit, so caution should be used when interpreting these effects for other social media platforms,” Fung said. “We expect these effects to generalize to environments where there is some type of reputation, such as Reddit, but examining whether these effects are replicated in other environments would be an interesting avenue for future research.”
Looking to the future, researchers are already expanding the scope of this study. “My co-authors Varad Deolankar and S. Sriram are working on another project related to user-driven content and polarization. However, this time we were interested in focusing on content consumption rather than production,” Fong said. “We’re thinking about how content platforms serve their users, how they interpret that content, and how they drive polarization of beliefs.”
The research team is currently investigating how individual biases interact with the platform’s algorithms. “This is an ongoing study, but what we know so far is that people tend to place less weight on information that contradicts their prior beliefs,” Fung said. “And this bias contributes almost as much to polarization as algorithms (which are commonly blamed for contributing to polarization).”
The authors also investigate how different site metrics influence these results. “We also ask whether the engagement metrics that platforms choose to optimize for (likes and dwell time) matter. We found that they do,” Fung said. “An algorithm that maximizes dwell time is less polarizing than an algorithm that maximizes likes. One reason for this is that users dislike content that goes against their views, but spend just as much time reading that content as content that they agree with, so an algorithm that maximizes dwell time will serve more articles that they disagree with than articles that maximize likes.”
The study, “The Impact of Down Voting on Content Creation: Evidence from Social Media,” was authored by Varad Deolankar, Jessica Fong, and S. Sriram.
