Skip to content
RelationshipAdvices
DatingModerate evidence

What are actual red flags versus things people just don't like?

The short answer

Two different things share the name. A small set of behaviours genuinely predicts harm: early control over money, movement or contacts, isolation from friends and family, monitoring, and any history of violence. Almost everything else on a typical list is preference, and the research supports treating it that way, since the individual characteristics of a partner predict relationship satisfaction surprisingly poorly compared with how the relationship itself actually goes.

10 min readPublished

What readers think

Looking back at a relationship that went wrong, had you noticed something early?

Anonymous. One vote per reader.

The term arrived from somewhere useful and has been inflated past the point of function. It now covers a partner's job, their music, how quickly they reply, and whether they own furniture, alongside behaviours that reliably precede serious harm. When one word covers both, it stops doing the thing it existed to do, and people end up simultaneously anxious about trivia and slow to recognise the pattern that actually matters.

Screening on traits works badly

The most relevant evidence here comes from a large machine learning study drawing on data from more than eleven thousand couples across dozens of longitudinal datasets. It found that features of the relationship itself, including perceived partner commitment, appreciation, sexual satisfaction and conflict, predicted satisfaction reasonably well, while the individual characteristics of the partners predicted very little in comparison[1]. Personality, background, values and demographics carried far less information than intuition suggests. This is directly relevant to the red flag genre, which is almost entirely a trait-screening exercise: it invites you to assess what kind of person someone is and to infer from that how the relationship will go. The evidence suggests that inference is weak. What predicts satisfaction is something you can only observe by being in the relationship, which is inconvenient advice but better supported than any checklist.

People are also wrong about their own preferences

A second body of work compounds the problem. Studies comparing what people say they want in a partner with what actually attracts them find the two correspond poorly. In speed-dating research where participants stated their ideals beforehand and then met potential partners, their stated preferences did not predict who they were drawn to[2]. A meta-analysis of ideal partner preferences reached a similar conclusion: stated ideals show little predictive validity for evaluations of actual partners once someone is known[3]. So the red flag list is doubly weak. It screens on traits that predict satisfaction poorly, using stated preferences that do not reliably describe what the person applying them actually wants. This does not mean preferences are illegitimate. It means treating them as a diagnostic instrument, rather than as taste, attributes an accuracy to them that they do not have.

The short list that does predict harm

None of the above applies to the behaviours the term was originally for. Systematic reviews of risk factors for intimate partner violence identify a consistent cluster, including prior perpetration of violence, controlling behaviour, and severe jealousy[4]. Alongside that, the pattern described as coercive control has a recognisable early shape: pressure to account for time, discouragement of specific friendships, involvement in decisions about money or work that narrows the other person's independence, and monitoring framed as concern. These differ from preferences in a way that is worth stating explicitly. They are behaviours rather than characteristics, they are directional in that they tend to escalate rather than stabilise, and they concern what someone does to another person rather than what they are like. A useful test is whether the thing you have noticed reduces someone's options. Preferences do not; this pattern does.

Why the inflation is costly

It would be tempting to treat an over-broad category as harmless caution, but it carries two specific costs. The first is that a list containing forty items has no priority ordering, so a genuine warning sign sits alongside a preference for a particular career and receives the same weight. People applying such a list end up dismissing candidates on trivia and, separately, talking themselves past the item that mattered, because everything is a red flag and therefore nothing is. The second cost is that the vocabulary of risk gets applied to ordinary conflict. Describing a partner who is bad at apologising as a red flag borrows the seriousness of the original term for a situation that is common, workable and not dangerous, and it makes the language unavailable when it is needed. Keeping the category small is what allows it to function as a warning.

It depends, on what?

What changes the answer

Preferences are not obliged to justify themselves and nothing here says otherwise. You may decline to date someone because of their politics, their hours, their friends or their handwriting, and you owe no one an evidential case for it. The claim is narrower: that a preference is not a prediction, and that dressing it as one produces false confidence in both directions. There is also a real limit to the risk research. It identifies factors associated with harm at the level of populations, which is not the same as a test that can be applied to one person; most people with any single risk factor are not violent, and applying population-level associations to individuals is exactly the error the red flag genre makes with traits. The honest position is that the short list is worth taking seriously when a pattern appears rather than a single item, and that no list substitutes for noticing whether your own life is getting smaller.

Where people disagree

Both sides of it

Reasonable people land in different places on this. Here is the strongest version of each case, not a straw man of the one we disagree with.

The case for yes

  • Keeping the category small preserves its function, since a warning that applies to everything warns about nothing.
  • The behaviours that predict harm are observable actions rather than personality assessments, which makes them something you can actually check.
  • Relationship-specific experience predicts satisfaction far better than partner traits, so attention is better spent on how it is going than on what someone is like.
  • Naming preferences as preferences is more honest and less anxious than treating every dislike as evidence of hidden danger.

The case for no

  • A short list risks being read as permission to ignore things that genuinely bothered someone, when discomfort is information even where it is not danger.
  • Risk factors are population-level associations and cannot diagnose an individual, so even the short list is a prompt for attention rather than a test.
  • The advice to judge by how the relationship goes requires being in it, which is precisely the exposure people are trying to avoid.
  • Early control can be subtle and is frequently only recognisable in retrospect, which limits how much any list can help at the point it would be useful.

Evidence

What the research says

How we rate evidence strength
Strong evidence
Multiple independent studies, including replications or meta-analyses, point the same way.
Moderate evidence
Several studies support this, but samples are limited or findings vary by population.
Limited evidence
Early, small, or single-study evidence. Treat this as a reasonable hypothesis, not a settled fact.
Contested
Credible researchers disagree, or the evidence points in conflicting directions.

Comparative view

Different perspectives

Traditions and disciplines answer this differently. We describe what each one teaches. We are not telling you which is right, and no tradition speaks with a single voice.

What the Evidence Supports

The claim that trait-screening predicts poorly is well supported by two independent lines of work, the machine learning study and the preferences meta-analysis, which is stronger than either alone. The risk factor literature is real but weaker than commonly presented: effect sizes are modest, definitions vary across studies, and the reviews explicitly caution against individual-level application. So the asymmetry in this answer is deliberate. The negative claim, that lists do not work well, rests on better evidence than the positive claim about what to watch for, and readers should treat the short list as informed by clinical and legal understanding of coercive control as much as by quantitative risk research.

Therapist Perspective

Clinicians commonly report that clients who arrive with detailed screening criteria are frequently managing anxiety rather than assessing partners, and that the list grows after each disappointment without improving outcomes. A related observation is that people are often able to describe, in retrospect, the moment something registered and was reasoned away, which suggests the difficulty is less about recognition than about what happens next. Practitioners working with people who repeatedly enter similar relationships also generally treat better screening as the wrong target, on the reasoning that what feels familiar is doing the selecting, and that this is addressed in individual work rather than by a more comprehensive checklist.

Secular Perspective

A secular framing separates two questions that the red flag vocabulary conflates: whether someone is compatible with you, which is a matter of preference and owes no justification, and whether someone is likely to harm you, which is a question about safety and is not a matter of taste. On this view no one is entitled to a relationship with you and declining requires no case. What the framing does insist on is that the second question is not answered by aesthetic judgement, and that the seriousness attached to safety language should not be borrowed to give preferences an authority they do not need and cannot support.

Practical steps

What you can actually do

  1. Ask whether the thing you noticed reduces your options. Preferences do not narrow your life; the pattern that matters does, and that test is easier to apply than any list.

  2. Watch direction rather than instances. A single controlling remark is ambiguous; the same thing escalating over months is not, and escalation is the signal.

  3. Notice whether you are editing yourself to manage someone's reaction, since that is often detectable earlier than any specific behaviour you could point to.

  4. Take a stated history of violence seriously as a fact rather than as a story about a past partner. Prior perpetration is among the most consistent findings in this literature.

  5. Let preferences be preferences. Declining someone because you do not enjoy their company needs no evidential case and is not improved by reframing it as a warning sign.

Worth unlearning

Common misconceptions

  • That a good list lets you screen out bad relationships in advance. Partner traits predict satisfaction poorly compared with how the relationship itself goes.

  • That knowing what you want makes you a good judge of a match. Stated ideals correspond weakly to who people are actually drawn to and to how they evaluate real partners.

  • That more red flags means more safety. A long undifferentiated list has no priority ordering, which reliably buries the items that matter.

  • That intense jealousy early on shows how much someone cares. Severe jealousy and controlling behaviour appear consistently among risk factors for intimate partner violence.

  • That a risk factor identifies a dangerous individual. These are population-level associations; they warrant attention to a developing pattern, not a verdict on a person.

The short version

Key takeaways

  • Individual partner characteristics predict relationship satisfaction poorly compared with relationship-specific experience.

    Research

  • Stated ideal partner preferences show little predictive validity for who people are actually attracted to or how they rate real partners.

    Research

  • Prior perpetration of violence, controlling behaviour and severe jealousy appear consistently as risk factors for intimate partner violence.

    Research

  • The practical test that separates the two categories is whether the behaviour reduces the other person's options.

    By definition

  • Risk factors describe populations and cannot diagnose an individual, so they justify attention to a pattern rather than a conclusion.

    Research

Worth saying

When to get professional help

If a partner controls money or movement, monitors contact, isolates you from friends and family, or you find yourself managing your behaviour to avoid their reaction, that is the pattern described as coercive control and a domestic abuse service is the appropriate contact rather than relationship counselling; couples counselling is generally advised against where one person is afraid of the other, because it can increase risk. If you have noticed yourself repeatedly in relationships with the same shape, individual therapy is more useful than better screening, since the pattern usually reflects what feels familiar rather than a failure to spot warning signs. If you are unsure whether what you are experiencing counts, most domestic abuse services will discuss exactly that question without any expectation that you have decided anything.

Find support services near you

Still wondering

Related questions people ask

On the evidence, that is a preference rather than a risk indicator. Some people are comfortable with it and some are not, and both positions are reasonable. It becomes worth attention only if it comes with concealment or with pressure on you to accept something you have clearly said you are not comfortable with.

Community

What other people say

These are readers describing their own relationships, not professionals. Take them as experience, not advice, and please do not name anyone.

0/4000

    No one has shared their experience yet. If you have been here, you would be the first, and probably the most useful thing on this page.

    Check our work

    Sources

    1. [1]
      Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies(opens in a new tab)

      Joel, S., Eastwick, P. W., Allison, C. J., and colleagues · Proceedings of the National Academy of Sciences · 2020

    2. [2]
      Sex differences in mate preferences revisited: Do people know what they initially desire in a romantic partner?(opens in a new tab)

      Eastwick, P. W., & Finkel, E. J. · Journal of Personality and Social Psychology · 2008

    3. [3]
      The predictive validity of ideal partner preferences: A review and meta-analysis(opens in a new tab)

      Eastwick, P. W., Luchies, L. B., Finkel, E. J., & Hunt, L. L. · Psychological Bulletin · 2014

    4. [4]
      A systematic review of risk factors for intimate partner violence(opens in a new tab)

      Capaldi, D. M., Knoble, N. B., Shortt, J. W., & Kim, H. K. · Partner Abuse · 2012

    Researched and written by the RelationshipAdvices Desk, reviewed before publication, and updated as the evidence changes. We are not licensed therapists. How we work.

    coercive controlcompatibilityred flagssafetydating

    People who read this also asked