The five-second answer
Once you fight through the headline and reach the paper, the story becomes much less dramatic.
A 2026 Polish study of 796 adults examined scores on the Autism-Spectrum Quotient (AQ), attachment, and behaviors used to maintain a romantic relationship.
At the zero-order level, higher AQ scores were weakly associated with fewer benefit-provisioning tactics, rho = −0.11, and more cost-inflicting tactics, rho = +0.17. The associations with attachment anxiety (+0.28) and attachment avoidance (+0.33) were larger.
But the pattern changes across structural models, regression models, and age-adjusted analyses. The paper itself also warns that the AQ is not a pure measure of autistic traits: it overlaps with social isolation, anxiety, reduced emotional expression, and communication difficulties.
So the defensible conclusion is not “autistic people are selfish in relationships.” A major part of the story is that a questionnaire containing social-communication difficulties is being related to social-communication behavior inside romantic relationships.
What did the study actually do?
Participants were 796 Polish adults who had been in a romantic relationship for at least three months. Their mean age was 36.4 years and 52% were women.
The researchers used four main families of measures:
- The 50-item AQ for autistic-like traits.
- The short ECR-R for attachment anxiety and avoidance.
- The 38-item MRI-SF for mate-retention behavior.
- Brief measures of mating effort and parental effort.
This was not a diagnostic study of autism spectrum disorder. Participants were not clinically sorted into autistic and non-autistic groups. The analysis treated AQ scores as a continuous self-report variable in the general population.
What does “autistic trait” mean here?
The AQ covers five broad areas: social skill, communication, attention switching, imagination, and attention to detail.
Items include preferences for doing things with other people versus alone, liking to do things the same way repeatedly, and how easily someone can imagine characters in a story.
A high AQ score is therefore not equivalent to a clinical ASD diagnosis.
Turning “high AQ” into “has autism,” then into “only thinks about themselves,” then into “does not care about a partner” is a four-step causal parkour routine that the study did not perform.
Relationship-maintenance tactics are not all the same
The mate-retention inventory groups behaviors into two broad classes.
Benefit-provisioning tactics increase a partner’s rewards or relationship satisfaction: affection, gifts, care, positive signals of commitment, and similar behaviors.
Cost-inflicting tactics impose costs: surveillance, monopolizing time, emotional manipulation, or behaviors that create fear, insecurity, or social restriction.
What people often call a “test” in a relationship can fall into the second category: acting cold to see whether the partner chases, hinting at a breakup to test the response, or provoking jealousy. But “cost-inflicting tactic” is a much broader category than “testing behavior.”
The numbers are much quieter than the headline
The simple correlations were:
| Relationship | rho |
|---|---|
| AQ × attachment anxiety | +0.28 |
| AQ × attachment avoidance | +0.33 |
| AQ × cost-inflicting tactics | +0.17 |
| AQ × benefit-provisioning tactics | −0.11 |
The −0.11 association is small. A rough squared-correlation intuition puts the shared variance near one percent. That is not a clean “AQ determines romantic behavior” result.
The paper also produces different-looking results under different models. AQ negatively predicted benefit-provisioning in the SEM and regression analyses. The cost-inflicting side was more unstable: the simple correlation was positive, some models did not show a direct effect, and adding age as a covariate revealed a positive association again.
That is precisely why a single dramatic sentence is a poor summary.
The real statistical joke: “aren’t you partly measuring similar things?”
This is the most interesting issue.
The AQ contains items about social skills and communication. The romantic-behavior measure asks whether people actively show affection, reward a partner, monitor them, or otherwise interact to maintain a relationship.
So part of the comparison is effectively:
a measure partly sensitive to social-interaction difficulty
versus
how often someone performs social-interaction behaviors with a romantic partner.
In an exaggerated version, it would be like selecting people who report little social interaction and then triumphantly discovering that they interact socially less often.
The present study is not fully circular, because AQ also includes attention to detail, routine-related preferences, attention switching, and imagination. But conceptual overlap remains a serious interpretive issue.
The authors explicitly note that AQ overlaps with negative schizotypy, especially social isolation, anxiety, reduced emotional expression, and communication problems. They also note that AQ factor structure is unsettled: different studies support five-, three-, six-, or hierarchical factor solutions.
A single total AQ score therefore mixes several psychological ingredients into one soup and labels the pot “autistic traits.”
This is not a study of “people who only care about themselves”
A real relationship can contain a pattern like this:
think internally
→ do not communicate it
→ do not seek the partner’s response
→ do not update on the partner’s needs
→ make a private conclusion
→ quietly withdraw
To the other person, that can feel one-sided.
But this study did not measure the motive behind such behavior. It cannot distinguish lack of interest, difficulty noticing needs, discomfort with closeness, low perceived need for explicit maintenance, or affection that is present but not externally expressed.
Behavior and motive are not interchangeable.
Attachment may be the more useful story
AQ correlated more strongly with attachment anxiety (+0.28) and avoidance (+0.33) than with benefit-provisioning (−0.11).
Those attachment dimensions also pull behavior in different directions. Anxiety can increase both positive reassurance and costly attempts to prevent abandonment. Avoidance can reduce benefit-provisioning because closeness itself is uncomfortable.
That makes the more faithful model:
traits captured by AQ → attachment and other pathways → relationship behavior
rather than a one-line “autistic traits cause X in romance.”
What would a stronger study do?
A more informative design would separate AQ components.
Does relationship behavior track mainly with social skill and communication? If so, the result is close to “people with social-interaction difficulty show fewer social-interaction behaviors.”
But if detail focus, preference for routine, or other less-social AQ components predict relationship behavior even after controlling for social-communication items, that would be much more interesting evidence of an autism-specific association.
Better studies would also add partner reports, observed communication behavior, longitudinal data, clinically assessed ASD groups, preregistered models, and cross-cultural replication.
Conclusion: inspect the measure before admiring the headline
The study found several associations between AQ scores, attachment, and relationship-maintenance behavior in a general-population sample.
It did not establish that autistic people are selfish, do not support partners, perform relationship tests, or lack affection.
The cleanest summary is:
People who score higher on a questionnaire that partly captures social-communication difficulty showed small differences in romantic relationship-maintenance behavior. Those differences were entangled with attachment, age, model choice, and overlap between the constructs being measured.
After all the dramatic language, the final boss was not “autism.”
It was construct validity.
References
- Łukasik A, Łaba-Hornecka A. Evolutionary Psychology. 2026. https://doi.org/10.1177/14747049261473640
- Baron-Cohen S et al. Journal of Autism and Developmental Disorders. 2001. https://doi.org/10.1023/A:1005653411471
- English MCW et al. Autism Research. 2020. https://doi.org/10.1002/aur.2198
- Murray AL et al. Psychological Assessment. 2016. https://doi.org/10.1037/pas0000215
