Attachment style test accuracy: what research says and why your result changes

You were anxious last time and secure today, your result in a relationship differed from the one after a break-up, or every site gave you a different style. Stories like these are common, and they naturally raise two questions: “Can I trust an attachment style test?” and “What is my real attachment style?”

The short answer is that a changing result is not unusual, and there are reasons for it. The questionnaires used in research give fairly similar scores when people take them again a few weeks later, but each test has its own items and scoring. The four names, secure, anxious, avoidant and fearful-avoidant, are labels given to two scores, anxiety and avoidance, cut at a line, so near that line a small difference is enough to change the name. And an attachment score holds both a lasting tendency and your current relationship and mood.

This guide covers how researchers have checked the accuracy of attachment questionnaires, why different ways of measuring give different results, whether people really fall into four styles, why your style can change on mind.io, and how stable attachment scores are. It is less about handing you the right style than about reading your result with less confusion and more use. Why MBTI and Enneagram results change is covered in the guide to MBTI accuracy and the guide to Enneagram accuracy.

When people talk about how accurate a psychological test is, they usually mean two things. Reliability is whether the same person gets a similar result on a retest (test-retest reliability) and whether the items measuring one scale move together (internal consistency). Validity is whether the test measures what it sets out to measure, and whether its results connect in meaningful ways with what people feel in real relationships or with other tests.

The adult attachment questionnaires widely used in research were refined against these standards. In 1998, Kelly Brennan and colleagues pooled 323 items from 14 existing attachment questionnaires, gave them to more than 1,000 university students, found that the answers grouped into two scales, anxiety and avoidance, and built the ECR (Experiences in Close Relationships), with 18 items per scale (Brennan et al., 1998, as cited in Fraley et al., 2000, and Nakao & Kato, 2004). In 2000, the same item pool was reanalysed with item response theory and 18 items per scale were selected afresh for the ECR-R; 13 of the 18 anxiety items and 7 of the 18 avoidance items had also been in the ECR (Fraley et al., 2000).

The internal consistency of both questionnaires is usually reported at around .91 (Fraley et al., 2011a). The ECR-R’s authors noted, however, that the low end of both scales, the secure end, is measured less precisely than the insecure end (Fraley et al., 2000). Stability on a retest has been checked too. In a study that gave the ECR-R again three weeks later, the two sets of scores shared 85% of their variance once measurement error was set aside, and when people kept a diary of the attachment-related feelings they had with their partner, ECR-R scores explained 30% to 40% of the differences between people. With family and friends it was only 5% to 15%, which fits a questionnaire built to measure attachment in romantic relationships (Sibley et al., 2005, from the abstract).

Translations have been checked in similar ways. The Japanese ECR kept 26 of its 36 items, with internal consistency of .91 for avoidance and .87 for anxiety, though its authors wrote that its test-retest reliability still had to be examined (Nakao & Kato, 2004). The French ECR showed the same two-scale structure (Lafontaine & Lussier, 2003, from the abstract), and the French ECR-R, given to 600 people, had satisfactory reliability on both scales and fitted the original two-factor structure best (Favez et al., 2016, from the abstract). A study that gave the Korean ECR-R to 143 university students found internal consistency of .92 for anxiety and .93 for avoidance (Lee et al., 2023); we found no retest data for the Korean version.

The mind.io attachment style test is none of these. It is not an adaptation of the ECR or ECR-R items but mind.io’s own test: 28 statements written by mind.io on the same two scales of anxiety and avoidance (14 per scale, 5 of them scored in reverse), answered from 1 to 5. The test’s retest reliability and validity have not been studied. So the figures above describe research questionnaires, and none of them tells you how accurate a mind.io result is.

There is more than one way to measure attachment. Early studies asked people to pick, from three short descriptions, the one that sounded most like them. When several data sets gathered this way were looked at again, about 30% of people picked a different style between one week and several months later, and most of those who had described themselves as anxious-ambivalent changed their answer. The authors suggested that this may reflect the relational schema that happens to be active at the moment of answering rather than poor measurement (Baldwin & Fehr, 1995, from the abstract). Questionnaires such as the ECR, now widely used in research, combine many items to show your position on the two scales as continuous scores.

Which relationship you have in mind also changes the score. In a study that asked the same nine questions separately about mother, father, partner and best friend, avoidance toward the mother and avoidance toward the partner correlated only .17 among 21,838 online respondents who were dating or married, and even the most similar pair, anxiety toward mother and father, correlated .41. The authors concluded that people who are secure with their parents may be insecure with their partners (Fraley et al., 2011a). A Japanese study also found correlations between relationships of .22 to .39 for avoidance and .37 to .57 for anxiety (Komura et al., 2016). The mind.io test asks people without a partner to think of their most recent relationship or of how they usually are in close relationships, so if you pictured your current partner one time and a past relationship the next, different scores are only natural.

Attachment can also be assessed by interview. The Adult Attachment Interview (AAI) listens to how people talk about their childhood experiences with their caregivers and classifies their current state of mind. A meta-analysis of 10 studies (961 people in all) that used both the interview and questionnaires found a mean correlation of .09 between security on the interview and questionnaire attachment scores, a trivial to small overlap. In the studies that also looked separately at people the interview classed as insecure (dismissing or preoccupied), questionnaire avoidance correlated only .15 with the dismissing (versus preoccupied) classification. The authors concluded that the two methods predict somewhat different aspects of romantic relationships (Roisman et al., 2007). So if an interview you had in therapy and an online questionnaire disagree, that does not mean one of them is wrong. The shares found with the interview are in the guide to how common each attachment style is.

Attachment researchers have used statistical methods to ask whether people fall into a few clearly separate groups or spread along continuous scores. Observations of 1,139 fifteen-month-old babies already showed differences in attachment that were largely continuous (Fraley & Spieker, 2003, from the abstract), and a study of adults’ self-report questionnaires found that both general attachment and attachment in specific relationships, such as with parents and peers, fitted dimensions better than categories (Fraley et al., 2015, from the abstract). An analysis of Adult Attachment Interview data from 40 studies and 3,218 people pointed the same way (Raby et al., 2022, from the abstract).

Chris Fraley, one of the ECR-R’s developers, writes on his web page that there is no “natural” or “correct” way to assign people to attachment styles and that there are no real types; since classifying people by their scores reduces measurement precision and statistical power, he advises researchers not to sort people into styles. He adds that a label such as “relatively secure” can still help when explaining what a score means to one person (Fraley, n.d.).

The mind.io test also scores anxiety and avoidance from 0 to 100 and then gives one of four names depending on whether each score is above 50. That 50 is simply the middle of the 1-to-5 answer scale, not an average or a norm drawn from other people. So, with the other score the same, an anxious result with anxiety at 52 and a secure result with anxiety at 48 carry different names but almost the same amount of anxiety. The style name is shorthand for a position; the two scores themselves say more.

The mind.io attachment style test turns the average of each scale’s 14 items into a score from 0 to 100. So if you answered all 14, moving one answer by one step moves that scale by about 1.8 points. A scale counts as “high” only above 50, and 50 itself goes to the low side, so someone with anxiety at 52 can get a different style name next time by moving just one answer one step in the less anxious direction and landing on 50. It only takes one of the two scales crossing the line to change the name, from anxious to fearful-avoidant or from secure to avoidant, for example.

That is why the result page shows not only the style name but both scores and your position on the two scales. If either scale is between 45 and 55, the top of the result says it sits close to the line and names the nearest neighbouring style; if a scale is between 40 and 60, “The texture of your scores” links to the style just across the line so you can read it too. These notes do not mean the test got it wrong; they tell you that this time, that scale landed close to the line.

Even for the same person, the moment you answer in moves the score. As above, the relationship you have in mind matters, and so can the mood of your current relationship or a recent argument. One study also found that people report less security after a break-up or separation (Ruvolo et al., 2001, as cited in Fraley et al., 2011b). A study that measured attachment again and again, daily or weekly, showed these moment-to-moment swings appearing alongside a lasting tendency (Fraley et al., 2011b); the next section looks at it more closely.

One more thing: in October 2026, mind.io replaced nine of the attachment test’s statements with new ones, including one that had been scored the wrong way and one that counted a healthy time-out as avoidance. Saved results and score links stayed the same, but if a result from before then differs from the one you get now, the changed items may be part of the reason.

Over a few weeks, scores on research questionnaires come out much the same. As noted above, the ECR-R was highly stable over three weeks (Sibley et al., 2005, from the abstract), and when the Japanese ECR-RS was given to university students again four to five weeks later, participants with a partner showed retest correlations of .80 for avoidance and .72 for anxiety toward that partner. The authors added that simple correlations can run high because of temporary errors such as memory or mood, and that a stricter estimate that removes them gave .61 for avoidance and .79 for anxiety toward the partner (Komura et al., 2016). With the pick-one-description method, by contrast, about 30% changed style even over short periods (Baldwin & Fehr, 1995, from the abstract). Scores can be fairly stable while the name near the line still flips easily.

What about longer periods? In the United States, 203 people who were dating or married answered an attachment questionnaire every day for 30 days, and about 380 answered every week for nearly a year. The study asked them to answer for how they felt at the moment rather than how they usually are. Attachment toward partners moved more than attachment toward parents, and among the weekly participants the test-retest correlation for avoidance toward the partner started high, then settled at about .30 after some 20 weeks and went no lower. The authors concluded that each person has a lasting central value around which they fluctuate from moment to moment, and that attachment at any one time is neither simply a state nor simply a trait but a combination of situational and enduring influences. They also noted as a limitation that most participants were fairly secure, which may have made stability look higher than it is (Fraley et al., 2011b).

Studies that follow people from childhood into adulthood describe attachment security as “moderately” stable. A meta-analysis covering infancy to age 19 reached that conclusion (Fraley, 2002, from the abstract), and a meta-analysis of 127 papers on 21,072 attachments to mothers and fathers reported a correlation of .39 between two time points, though no statistically significant stability was found across intervals longer than 15 years (Pinquart et al., 2013, from the abstract). In adulthood, too, anxiety and avoidance were seen to decline with age on average (Chopik et al., 2019, from the abstract). Which relationship experiences move attachment toward security is covered in the guide on becoming more securely attached.

First, look at the two scores rather than the name. Check how far anxiety and avoidance each sit from 50, and which scale comes out high in both results. If avoidance was low both times and only anxiety moved around 50, the question “Am I secure or anxious?” turns into a more accurate description: “I don’t tend to keep my distance, and my anxiety rises in some situations.”

Second, retake it on the same basis. Note which relationship you had in mind this time and answer with the same relationship in mind next time, so the two results can be compared. You can also take it once for your current partner and once for your close relationships in general, and set the two results side by side.

Third, give it time. Retaking it again and again right away because you dislike the result makes it easy to remember the items and pick the answers you want. Try again in a few weeks or months, in a similar situation. If an earlier result is still saved in the same browser, “Compare with last time” on the result page shows how much your anxiety and avoidance scores changed. That difference may reflect your recent state, the context or the way you answered more than a change in who you are.

Fourth, check your scores against real moments. Jotting down the moments when you felt badly shaken shows your pattern more clearly than any test score. How to keep that record is in the guide on how to read your attachment result.

Finally, here is what not to use the result for. An attachment style is a reference for understanding yourself, not a diagnosis, and it is not a label for judging a partner by their result or for deciding to break up. A fearful-avoidant result does not necessarily mean you have been through trauma either; the guide on coping with fearful-avoidant attachment explains why. If a partner controls or threatens you, that is not an attachment style problem but a safety problem: reach out first to people you trust or to a specialist service. Warning signs and helplines are listed under “When it is not about attachment” in how to talk to an avoidant partner. If low or anxious mood lasts more than two weeks, the depression check-in (PHQ-9) or the anxiety check-in (GAD-7) can help you see where you are before you talk to a professional. These check-ins are not diagnoses either.

Q. Is the attachment style test scientifically reliable? A. Research questionnaires such as the ECR and ECR-R usually report internal consistency of around .91 (Fraley et al., 2011a), and scores stayed stable when the ECR-R was retaken three weeks later (Sibley et al., 2005, from the abstract). But each test has its own items and scoring, and the mind.io test uses its own items, so there are no reliability data for it yet. Read your anxiety and avoidance scores rather than the style name.

Q. I get a different style every time. Which one is real? A. If one of the two scores is near 50, you may show traits of both styles. Read the note about the line and the neighbouring style on the result page, then retake it in a few weeks with the same relationship in mind. The scale that comes out on the same side each time is closer to your tendency.

Q. I seem to be a different style with my partner than with friends. A. That can happen. In a study that asked the same questions about each relationship, scores toward the partner and toward the best friend correlated only .14 for avoidance and .25 for anxiety (Fraley et al., 2011a). Noting which relationship you had in mind makes results easier to compare.

Q. My result differs from the attachment interview I had in therapy. A. The Adult Attachment Interview and questionnaires measure different things. In an analysis of 10 studies, the mean correlation between security on the interview and questionnaire scores was .09 (Roisman et al., 2007). Read them as two angles rather than one being wrong.

Q. Does my attachment style stay the same for life? A. No. Attachment security in childhood carries over moderately into adulthood (Fraley, 2002, from the abstract), and in adulthood anxiety and avoidance were seen to decline with age on average (Chopik et al., 2019, from the abstract). Which experiences help change is covered in the guide on becoming more securely attached.

Q. Can I ask my partner to take the test to find out their style? A. It can be a good starting point for understanding each other’s reactions, but using it to judge your partner or to end the relationship tends to shut the conversation down. How to read two results together is in the guide to attachment styles in dating.

Q. Do MBTI and Enneagram results change for the same reason? A. In one respect, yes: near a dividing line, a small difference changes the name. More is in the guide to MBTI accuracy and the guide to Enneagram accuracy. If you would rather see your personality as scores than as a type, the Big Five test fits.

References: Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998), Self-report measurement of adult attachment: an integrative overview, in J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships, 46-76, New York: Guilford Press (cited in Fraley et al., 2000, and Nakao & Kato, 2004); Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000), An item response theory analysis of self-report measures of adult attachment, Journal of Personality and Social Psychology 78(2), 350-365, https://doi.org/10.1037/0022-3514.78.2.350; Fraley, R. C., Heffernan, M. E., Vicary, A. M., & Brumbaugh, C. C. (2011a), The Experiences in Close Relationships-Relationship Structures questionnaire: a method for assessing attachment orientations across relationships, Psychological Assessment 23(3), 615-625, https://doi.org/10.1037/a0022898; Sibley, C. G., Fischer, R., & Liu, J. H. (2005), Reliability and validity of the Revised Experiences in Close Relationships (ECR-R) self-report measure of adult romantic attachment, Personality and Social Psychology Bulletin 31(11), 1524-1536, https://doi.org/10.1177/0146167205276865; Nakao, T., & Kato, K. (2004), Constructing the Japanese version of the Adult Attachment Style Scale (ECR) [in Japanese], Japanese Journal of Psychology 75(2), 154-159, https://doi.org/10.4992/jjpsy.75.154; Komura, K., Murakami, T., & Toda, K. (2016), Validation of a Japanese version of the Experience in Close Relationship-Relationship Structure [in Japanese], Japanese Journal of Psychology 87(3), 303-313, https://doi.org/10.4992/jjpsy.87.15208; Lafontaine, M.-F., & Lussier, Y. (2003), Structure bidimensionnelle de l’attachement amoureux: anxiété face à l’abandon et évitement de l’intimité, Canadian Journal of Behavioural Science 35(1), 56-60, https://doi.org/10.1037/h0087187; Favez, N., Tissot, H., Ghisletta, P., Golay, P., & Cairo Notari, S. (2016), Validation of the French version of the Experiences in Close Relationships-Revised (ECR-R) adult romantic attachment questionnaire, Swiss Journal of Psychology 75(3), 113-121, https://doi.org/10.1024/1421-0185/a000177; Lee, J.-Y., Kim, Y.-K., & Shin, Y.-J. (2023), Validation of the Korean version of culturally responsive Experiences in Close Relationships-Short Form, International Journal for the Advancement of Counselling 45(1), 57-81, https://doi.org/10.1007/s10447-023-09503-6; Baldwin, M. W., & Fehr, B. (1995), On the instability of attachment style ratings, Personal Relationships 2(3), 247-261, https://doi.org/10.1111/j.1475-6811.1995.tb00090.x; Roisman, G. I., Holland, A., Fortuna, K., Fraley, R. C., Clausell, E., & Clarke, A. (2007), The Adult Attachment Interview and self-reports of attachment style: an empirical rapprochement, Journal of Personality and Social Psychology 92(4), 678-697, https://doi.org/10.1037/0022-3514.92.4.678; Fraley, R. C., & Spieker, S. J. (2003), Are infant attachment patterns continuously or categorically distributed? A taxometric analysis of strange situation behavior, Developmental Psychology 39(3), 387-404, https://doi.org/10.1037/0012-1649.39.3.387; Fraley, R. C., Hudson, N. W., Heffernan, M. E., & Segal, N. (2015), Are adult attachment styles categorical or dimensional? A taxometric analysis of general and relationship-specific attachment orientations, Journal of Personality and Social Psychology 109(2), 354-368, https://doi.org/10.1037/pspp0000027; Raby, K. L., Verhage, M. L., Fearon, R. M. P., Fraley, R. C., Roisman, G. I., van IJzendoorn, M. H., et al. (2022), The latent structure of the Adult Attachment Interview: large sample evidence from the Collaboration on Attachment Transmission Synthesis, Development and Psychopathology 34(1), 307-319, https://doi.org/10.1017/S0954579420000978; Fraley, R. C. (n.d.), Information on the Experiences in Close Relationships-Revised (ECR-R) adult attachment questionnaire, University of Illinois at Urbana-Champaign, https://labs.psychology.illinois.edu/~rcfraley/measures/ecrr.htm (accessed October 2026); Fraley, R. C., Vicary, A. M., Brumbaugh, C. C., & Roisman, G. I. (2011b), Patterns of stability in adult attachment: an empirical test of two models of continuity and change, Journal of Personality and Social Psychology 101(5), 974-992, https://doi.org/10.1037/a0024150; Ruvolo, A. P., Fabin, L. A., & Ruvolo, C. M. (2001), Relationship experiences and change in attachment characteristics of young adults: the role of relationship breakups and conflict avoidance, Personal Relationships 8(3), 265-281 (cited in Fraley et al., 2011b); Fraley, R. C. (2002), Attachment stability from infancy to adulthood: meta-analysis and dynamic modeling of developmental mechanisms, Personality and Social Psychology Review 6(2), 123-151, https://doi.org/10.1207/S15327957PSPR0602_03; Pinquart, M., Feußner, C., & Ahnert, L. (2013), Meta-analytic evidence for stability in attachments from infancy to early adulthood, Attachment & Human Development 15(2), 189-218, https://doi.org/10.1080/14616734.2013.746257; Chopik, W. J., Edelstein, R. S., & Grimm, K. J. (2019), Longitudinal changes in attachment orientation over a 59-year period, Journal of Personality and Social Psychology 116(4), 598-611, https://doi.org/10.1037/pspp0000167