Intelligence is simply talking to many people

How to Change Someone's Mind


Notes on the recent research done on ChangeMyView

Papers cited:

[1] Designing for Civil Conversations: Lessons Learned from ChangeMyView (2017); Shagun Jhaver, Pranil Vora, and Amy Bruckman

[2] Winning Arguments: Interaction Dynamics and Persuasion Strategies in Good-faith Online Discussions (2016); Chenhao Tan et al.

COMMUNITY - different settings, different conversations

[In CMV] "Users tend to conform to standards of thoughtfulness in commenting behavior set by others on online news discussion sites [1]

It may be difficult to influence civil behavior in a new or existing community if the existing participants don't display such behavior. [1]

on most other subreddits, users are more committed to defending their own views rather than understanding others' perspectives. [1]

such communities are not suitable for civil disagreements. [1]

creating example content that suggests the desired norms and use gamification to encourage civil contributions. [1]

Participants said that CMV users are intelligent, knowledgeable and active on the community. This makes posting on CMV rewarding. [1]

Echoroom effects, forming relationship and relationship expectations

when users share a common social identity on an online community, they are more susceptible to group influence and stereotyping, despite participant anonymity. [1]

discussion quality is monitored by moderators, and posters commit to an openness to changing their minds [2]

CMV rules also encourage commenters not to post any low effort comments [1]

CMV rules explicitly forbid users from being rude or hostile to other members or accuse others of being unwilling to change their views. Many participants felt that a strict enforcement of these rules has been critical in maintaining the civil nature of conversations. Participant P-4, who is a CMV moderator, said that the moderation on CMV tends to be stricter than on other subreddits [1] ... it makes people mroe cautious in how they are posting and encourages polite posts. [1]

making sure that OPs explain why they hold their beliefs and do so at reasonable length (500 characters or more), and that OPs engage in conversation with challengers in a timely fashion. [2]

(advantages of using CMV as a data source:) The OP has, in principle, expressed an openness to other points of view, so that we might hope to extract a sufficient number of view-changing examples [2]

"I've seen threads go ugly so fast [on other subreddits] and I think that having active mods helps CMV not get bogged down by trolls." P-5 [1]

to "soapbox" their point of view instead of engaging in a genuine inquiry about the topic. [1]


The use of delta and deltaboards provides strong competitive cues to CMV users and motivate them to adopt behaviors that earn them reputation [1]

gamification cannot help a system if the entity being gamified does not already have some intrinsic value for the users [1]

weaving gamification into this process deepens their engagement and desire to participate [1]

Competition-based incentives that are central to gamification can undermine participants' motivation [1]

demotivating effects of gamification [1]

META - how we find out how persuasion works

an ideal setting for the study of persuasion would allow access to the reasoning behind people's views in addition to the full interations [2]

predicting which of the two similar counterarguments will succeed in changing the same view. [2]

stylistic choices in presentation of an argument [2]


Information doesn't necesarily lead to understanding or change. It is a revealing, not a transforming, medium. [1]

multiple factors are at play in persuasion. Beyond the characteristics of arguments themselves. [2]

in the majority of cases, however, opinions are not changed [2]

possibly depending on how strongly the OP holds them and how the OP acquired and maintained them [2]

Changing someone's opinion is a complex process, often involving repeated interactions between the participants [2]

because people can see time and time again what works [1]

topics that are the most malleable (food, eat, eating, thing, meat, read, book, lot, books, women) and the most resistent (government, state, world, country, countries, sex, women, fat, person, weight)


First person pronouns are strong indicators of malleability, but first person plural pronouns correlate with resistence. In psychology, self-affirmation has been found to indicate open-mindedness and make beliefs more likely to yield ... first person plurals can indicate a diluted sense of responsibility for the view [2]

The use of more paragraphs, bold formatting, and bulleted lists are all higher when a malleable view is expressed. Taking more time and presenting the reasons behind an opinion in a more elaborated form can indicate more engagement. [2]

Dominance is the most predictive of malleability: the average amount of control expressed through the words used is higher when describing a malleable view... The same holds for happiness (captured by valence). In terms of arousal, malleable opinions are expressed significantly more serenely, ending on a particularly calm note in the final quarter, while stubborn opinions are expressed with relatively more excitement. [2]

resistant views tend to be expressed using more decisive words such as anyone, certain, ever, nothing, and wrong, while help and please are malleable words [2]

comparative adjectives and adverbs are signs of malleability, while superlative adjectives suggest stubbornness. [2]

The overall low scores suggest that this is indeed a challenging task for both humans and machines. [2]

(Not yet studied): sequence of discourse connectors (first-but-because; now-then-instead) [2]


The language used in arguing, however, is under one's complete control [2]

Many participants pointed out that they repeatedly observed that users who were able to change the views of OPs were those who were polite in their posts. This inspired them to be civil in their own comments. [1]

frequency of certain types of words may be associated with persuasion success (Arousal, Concreteness, Dominance, Valence) [2]

successful arguments being by using calmer words [2]

when it comes to normalized counts, though, only italicizing exhibits significance [2]

Definite articles ("the" instead of "a") are preferred, which suggests that specificity is important in persuasive arguments [2]

Persuasive opening arguments use fewer positive words, suggesting more complex patterns of positive emotion in longer arguments

persuasive arguments use a significantly larger absolute number of personal pronouns [2]

Hedging is more common in persuasive arguments under root reply and full path ("It could be the case")... Their presence might signal a weaker argument, but alternatively, they may make an argument easier to accept by softening its tone. [2]

if anything, it seems better to avoid using question marks [2]

One common practice in argumentation is to quote the other party's words. However, this does not seem to be a useful strategy for the root reply. [2]


dissimilarity with the original post in word usage, while existing theories mostly study matching in terms of attitude functions or subject self-discrepency [2]

Both effective and ineffective arguments start off more similar with the original post; effective arguments ramain less similar overall. [2]


a larger number of words is strongly correlated with success [2]

persuasive arguments have significantly more of both number of paragraphs and number of sentences [2]


challenger that enters the fray before another tends to have a higher likelihood of changing the OP's opinion; this holds even for first-time CMV challengers [2]

A late entry might give the challenger time to read attempts by other challengers and better formulate their arguments, while an early entry might give her the first-mover advantage. [2]

Most deltas were given to the first and second people who entered. Challengers who entered later saw a steadily-reducing liklihood of getting a delta.


the more participants there are in the effort to persuade the OP, the larger the liklihood of the OP changing their view [2]

a saturation in how much value each new challenger adds beyond a certain point [2]

-It seems that around 20 challengers makes a post involving enough to produce real results. Between 20 challengers and 130 challengers the delta percentage goes from 25% to 40%, but when you add more than 130 challengers it doesn't go up past 40%.


more than one reply in a rooted path-unit correspond to a higher chance that the OP will be persuaded. So, while the challenger's opening argument should be important, statements made later in the rooted path-unit could be more important [2]

Deltas were awarded, on average, 3% for first reply, 5% for second-deep reply, 6% for third-deep reply, 5% for fourth-deep, and then shot sharply down to 0 percent by the fifth-deep reply, shoing there is no point to adding comments after the fourth-deep in a path-unit.


since single-challenger subtrees consistently outperform multiple-challenger subtrees in terms of conversation rate [2]

might suggest that when talking about the same counterargument, challengers might not be adding value to it, or they might even disagree; alternatively, root replies that attract multiple challengers might be less effective to begin with. [2]

Although both single- and multip-challenger subtrees got more deltas the deeper the conversation went (from 2 to 4 exchanges deep), single-challenger subtrees went from 7% to 8% liklihood of getting a delta, while multiples went from 3% to 4%.

include links as evidence--an interesting contrast to studies of the backfire effect: "When your deepest convictions are challenged by contradictory evidence, your beliefs get stronger." However, it hurts to be too intense in the counterargument. [2]

.com links seem to be the most powerful (compared with .edu and pdf documents) [2]


a member's success rate goes up with the number of attempts made. This observation can be explained by at least two reasons: the success rate of frequent challengers improves over time, and/or frequent challengers are better at persuasion from the beginning. [2]

suggesting that the success rate of frequent challengers does not increase. It is worth noting that this lack of apparent improvement might be explained by a gradual development of a "taste" for original posts that are harder to address. [2]


[some people] felt responsible for posting on CMV to create a dialog with people that they may not agree with beforehand. [1]


is benefit-cost analysis the only way to convince a utilitarian? [2]

Harvard negotiations:

If you like someone you might play lost-win. If you don't care for the relationship you might play win-lose.

Sometimes cooperation is not possible because negotiators are going too fast into solutions. You hear the problem and immediately suggest a solution. All parties are doing this. You end up with compromise. To get to win-win, based on your interest you should first develop criteria that must be fulfilled by a solution you could say yes to. What condititions must a solution you could say yes to fulfil? 'How would you realize that a restaurant is a good restaurant for you?' Of those that fulfil these criteria, there will be lots of restaurants. Before creating a solution, agree on certain principles that must be fulfilled by the solution.

People like to chose. You should have options to chose from. Develop maybe 2, 3 or 4 options, then evaluate your options by the criteria. More sustainable because they have a choice.

Comments: 0

Interested to discuss? Leave a comment.


Your email will not be published nor shared with anyone. In your text you can use markdown for marking up *italic*, links <> and other elements. These comments are moderated and published manually as soon as possible.