When people think about what they want in a romantic partner, they tend to name more positive qualities than negative ones, yet they agree far less on which flaws to avoid. New research published in European Journal of Social Psychology suggests that desirable partner traits cluster tightly around a shared ideal, whereas undesirable traits span a much wider range of meanings. Because people vary so much in the flaws they reject, negative traits turned out to be much more informative than positive ones for predicting hypothetical online dating decisions.
Online dating platforms require users to form quick impressions of strangers from brief text descriptions and a handful of photos. Most psychological research on partner preferences has focused on the specific meaning, or content, of the personality traits people seek, identifying widely desired qualities such as warmth, kindness, intelligence, and trustworthiness. Rather than looking only at what individual traits mean, psychologists can also examine their broader structural patterns, specifically how frequently positive and negative traits come to mind and how closely traits of the same type resemble one another.
Work in social cognition indicates that good and bad traits are not simple mirror images of each other. For instance, a 2012 study of person descriptions found that adults naturally used a much wider variety of words when describing people they disliked than when describing people they liked. Building on similar patterns across language and memory, researchers introduced the Evaluative Information Ecology model in a 2019 theoretical review co-authored by one of the new study’s authors, Christian Unkelbach. That framework argues that everyday social environments supply an abundance of positive information that tends to be quite alike, alongside rarer negative information that takes many distinct forms.
Relationship scientists have simultaneously begun paying closer attention to undesirable partner qualities rather than treating them merely as the absence of good ones. For example, an earlier survey on romantic red flags grouped common dealbreakers into six broad categories, finding that apathy and being gross or unkempt were rated most off-putting for long-term commitments, whereas being gross or unkempt and clinginess topped the list in short-term mating contexts. Although a 2020 study of courtship decisions found that learning about dealbreakers shifted people’s romantic interest more strongly than learning about positive dealmakers, other work has questioned whether negative qualities always carry more weight when daters evaluate profiles.
Don’t miss the next big psychology finding
Every morning, PsyPost sends a short briefing on the most interesting new peer-reviewed research in psychology and neuroscience.
To bridge these two lines of work, Tabea J. Zorn and Christian Unkelbach examined how the structural properties of traits operate in online dating. Zorn conducted the research at the Social Cognition Center Cologne in the Department of Psychology at the University of Cologne, and Unkelbach is a professor of experimental psychology at the University of Cologne, where he is affiliated with the Social Cognition Center Cologne and the Center for Social and Economic Behavior. Dating offers a demanding test of whether positive traits outnumber negative ones, because active daters are often coming out of unsuccessful relationships that might leave past red flags fresh in memory.
“My colleagues and I have a general model of evaluative information (i.e., good and bad behaviors/traits/people, etc.) and what we call its ‘ecology,'” Unkelbach told PsyPost. “In a nutshell: Positive information is more frequent, but negative information is more diverse (or less similar). We call this the structural properties of positivity prevalence and negativity diversity.”
“The present paper investigated whether these assumptions also hold for the information in dating contexts,” he added. “It thus presents a test of the model, and if the model holds, there are interesting consequences for self-presentation in online dating.”
Across four studies comprising seven experiments with a total of 1,074 participants, the researchers tested how people generate, evaluate, and act upon positive and negative personality traits in dating scenarios. Every experiment was preregistered, and the team posted their data, materials, and analysis code publicly online.
The first three experiments tested whether people bring to mind more positive traits to seek than negative traits to avoid. In Studies 1a and 1b, the researchers surveyed two groups of 100 English-speaking adults from the United Kingdom and the United States via the online platform Prolific, all of whom had prior experience with dating apps. Participants listed as many positive traits to look for and negative traits to avoid when dating online as they could think of, typing their answers into two text boxes. Study 1a displayed the positive box first, whereas Study 1b reversed that order to rule out prompt-order effects.
Study 1c moved the task into a university laboratory in Germany with 98 German-speaking adults (after excluding three participants under age 18) and randomized which text box appeared first. Unlike the online surveys, the lab experiment dropped the requirement of prior online dating experience and instead asked afterward, resulting in 61 participants who had dated online and 37 who had not. Across all three experiments, the team filtered out between 19 and 31 percent of raw entries because they did not describe character or personality traits, such as height, hair color, profile photo habits, or shared music tastes.
Based on the screened trait lists, individuals in all three experiments named more desirable qualities than undesirable ones, averaging between 3.61 and 4.26 positive traits per person compared to between 2.44 and 2.88 negative traits. In the German laboratory sample, neither prior experience with online dating nor the order of the text boxes had a statistically significant effect on how many traits participants generated. Even when the researchers reanalyzed all raw entries without filtering out non-personality items, the higher count of positive entries held in Studies 1a and 1c, though the difference in Study 1b was no longer statistically significant.
Although the researchers had not originally set out to test trait variety in these first experiments, a different picture emerged at the group level when they pooled all distinct English trait words from Studies 1a and 1b. Even though each individual wrote down more positive traits, the combined pool contained 181 unique negative traits versus 152 unique positive traits, alongside four words (“educated,” “flirtatious,” “needy,” and “political”) that some participants listed as desirable and others listed as flaws.
While some negative words were straightforward opposites of positive ideals, such as “unmotivated” versus “motivated,” participants also named flaws with no true positive counterpart, such as “manipulative” and “abusive,” as well as excesses of desirable qualities, such as “overconfident.”
Counting distinct words alone cannot establish how much traits differ in meaning, because separate labels like “smart” and “intelligent” overlap closely. In Study 2, 175 English-speaking adults in the United Kingdom and the United States each rated one of seven lists drawn from the 337 sampled English traits, yielding roughly 25 raters per trait. Participants scored each word on 9-point scales across three foundational dimensions of emotional meaning: pleasantness (from unhappy to happy), emotional intensity (from calm to excited), and sense of control (from feeling controlled to being in control). Before the main task, everyone rated the same 12 reference words spanning the full range of each scale to calibrate their use of the endpoints.
After dropping the 12 reference words and the four traits listed as both good and bad, the researchers plotted the remaining 333 traits in a three-dimensional map. They then calculated each word’s geometric distance from every other trait of the same category, where a smaller score meant higher similarity. Positive traits clustered much closer to one another, averaging a distance score of 1.96 on a scale from 0 for identical ratings to roughly 13.9 for maximum dissimilarity, whereas negative traits spread out more widely with an average score of 2.33. Follow-up checks that were not preregistered removed the most extreme positive and negative words and left this gap intact, indicating that outliers did not account for the wider spread of flaws.
“Most people agree on what positive traits are in online dating contexts, but there is more variety on what people consider ‘negative,'” Unkelbach explained.
Study 3 examined how a trait’s structural similarity relates to its perceived importance in dating. Another 200 English-speaking adults on Prolific evaluated subsets of the 333 analyzed traits, providing about 50 ratings per word. For each trait, participants first indicated whether it was something to look for or avoid when dating, and then rated how important that was on a 7-point scale from 1 (not at all important) to 7 (very important). Responses in which a participant disagreed with a word’s original positive or negative classification, making up about 11 percent of ratings, were excluded as preregistered.
On average, participants rated avoiding negative traits as modestly more important (5.51 out of 7) than seeking positive traits (5.37 out of 7). For positive traits, qualities that sat closer in meaning to other desirable traits tended to receive higher importance ratings, fitting classic impression-formation theories that “central” traits closely linked to many good qualities carry extra weight in shaping an overall impression.
Unexpectedly, the same link between closeness in meaning and higher importance also appeared for negative traits. A direct statistical comparison indicated that this relationship was stronger for flaws than for positive qualities. Rather than unusual, isolated flaws standing out as most important, negative traits that sat closer in emotional meaning to other flaws were rated as more important to avoid.
Looking at individual words, trustworthy topped the list of positive dealmakers at 6.73 out of 7, with kind, loyal, genuine, and understanding next in rank, while abusive led the negative dealbreakers at 6.98, followed in order by unfaithful, dishonest, deceiving, and untrustworthy. Yet those content-based rankings barely overlapped with the traits that sat closest to the center of their respective clusters in Study 2, which included clever, natural, cultured, helpful, and open among positive traits, and crude, stuck-up, unhumble, disloyal, and shallow among negative traits. Across the top 10 lists for both categories, “fake” was the lone trait to rank among both the most important dealbreakers and the most structurally similar flaws.
“Of interest, the diversity of information leads to different lists (i.e., the most similar traits and thereby the most ‘central’ are not the most negative),” Unkelbach noted. “One might derive, but we did not test this, that it is not the ‘content’ alone of a positive trait one displays, but how similar it is to other traits.”
Studies 4a and 4b tested whether structural similarity could predict hypothetical dating app choices above and beyond a trait’s stated importance. After excluding words that 50 percent or more of Study 3 raters had categorized inconsistently, 308 traits remained in the pool (146 positive and 162 negative). In Study 4a, 200 Prolific participants imagined viewing a potential partner described by a single word and chose whether to “swipe left (no option to date)” or “swipe right (option to date).”
Because the researchers initially assumed nearly everyone would reject a partner described by a lone flaw, they preregistered formal predictions only for positive traits in Study 4a and treated the negative-trait analysis as exploratory. After observing wide variation in how people reacted to flaws, they ran Study 4b with 201 participants using a 7-point likelihood scale (from 1, definitely swipe left, to 7, definitely swipe right) and preregistered predictions for both positive and negative traits.
For positive traits, stated importance strongly predicted swiping right in both experiments, with participants in Study 4a accepting an average of 95 percent of partners described by desirable qualities. Structural similarity, however, was not a statistically significant predictor of swiping decisions for positive traits in either Study 4a or Study 4b. Switching from a yes-or-no choice to the 7-point scale in Study 4b spread out participants’ ratings (averaging 5.84 out of 7), suggesting that a ceiling effect in the binary task does not account for the absence of a similarity effect on the positive side.
Responses to negative traits varied considerably more than responses to positive ones. An average of 15 percent of participants in Study 4a were still willing to swipe right across undesirable traits (and ratings averaged 2.18 out of 7 in Study 4b), depending on whether a flaw was mild like “unfashionable” or severe like “abusive.” In both experiments, a negative trait’s importance and its structural similarity each independently predicted hypothetical swiping choices: even after accounting for stated importance, flaws that were closer in emotional meaning to other negative traits were associated with a lower chance of swiping right.
Across all trials in Study 4a, 52.9 percent of choices were right swipes even though negative traits made up slightly more than half of the prompts. The authors suggest this pattern aligns with a “progression bias” toward giving potential matches the benefit of the doubt in early dating decisions.
Asked what surprised the team most about the swiping experiments, Unkelbach pointed to how selectively participants reacted to undesirable traits. “That we found such a differentiated influence on the negative side – we expected that if one has a negative trait, one is ‘out.’ However, people differentiated,” he said.
Across both swiping studies, a trait’s stated importance remained a stronger predictor of hypothetical choices than its structural similarity, though similarity consistently added extra predictive power on the negative side. In hindsight, the authors suggest, negative traits were more informative for distinguishing swiping choices because people agreed on what to seek while varying much more in what they chose to avoid.
These results align with a 2015 study on partner evaluation, which found that people weighed negative dealbreakers more heavily than positive dealmakers when judging potential mates, especially among women and in long-term relationship contexts. Both studies point to the strong role of undesirable traits in romantic decisions, though the earlier study tested how dealbreakers and dealmakers shaped acceptance of described partners, while the new work measured isolated traits’ stated importance and semantic structure.
“The ‘content’ effects are large, but also a little trivial,” Unkelbach pointed out. “The ‘structural’ effects are smaller but also substantial (i.e., people state a much more diverse set of negative traits compared to positive traits).”
“Importantly, we find that negative traits (i.e., traits one should not have) seem to play a substantial role, at least in our hypothetical (this is important – no real dating) dating decisions,” he continued.
There are a few caveats to keep in mind when interpreting these findings. Participants evaluated hypothetical partners described by just one personality word at a time and saw no profile photographs. In real-world dating apps, physical attractiveness in photos heavily influences swiping choices, and personality traits come in bundles alongside a dater’s own traits, allowing people to trade off minor flaws against appealing qualities.
“We must acknowledge that these are self-reports of people in hypothetical scenarios,” Unkelbach cautioned. “The importance for real-life decisions should not be overstated. The model test (i.e., do the structural properties we propose also hold here) is solid, though.”
Future studies could examine how combinations of positive and negative traits interact on fuller profiles, as well as whether initial swiping patterns carry over into face-to-face dates and non-Western cultural settings. Because almost every dater showcases the same cluster of positive qualities, the authors also raise the question of whether admitting a minor flaw on a profile might make a person seem more genuine. In Studies 4a and 4b, however, even relatively mild flaws drew low rates of right swipes when presented on their own.
“Is there a potential gain in reporting negative information in online portals if the goal is to appear as a potential partner (e.g., as a signal of honesty; nobody is perfect)? So far, the answer seems ‘no,’ but it is an open question,” Unkelbach observed.
Unkelbach also credited his co-author with leading the research. “The lead author was my colleague Tabea Zorn, who unfortunately left my team for a (very good) job in the industry. All the best, Tabea!”
The study, “Valence Asymmetries of Traits in Dating Contexts,” was authored by Tabea J. Zorn and Christian Unkelbach.