Big Five test accuracy: what research says and why your scores change

Last time your extraversion came out at 63, “Mid”, and today it is 67, “High”; your result’s title switched from “An evenly spread profile rather than one standout trait” to the name of one trait; or another site put the same trait in some top percentage. Stories like these are common, and they naturally raise two questions: “Can I trust a Big Five test?” and “What is my real score?”

The short answer is that scores moving up or down by a few points is not unusual, and there are reasons for it. Research Big Five tests keep people in much the same order when they are retaken within a few weeks, yet things that vary from one occasion to the next, such as the day’s mood, still mix into the scores a little. The fewer items a test has, the less those items hang together, and once scores are cut into bands such as low, mid and high, a small difference near an edge changes the name. And over the years, personality itself moves a little.

This guide covers how researchers have checked the accuracy of Big Five tests, why tests differ in length and items, how similar scores stay over weeks and over years, how well scores match real life and other people’s view of you, why your scores move on mind.io, and how to read scores that changed. How the MBTI and the Big Five differ is covered in the guide to MBTI vs the Big Five, and whether personality itself changes with age and effort in the guide Can personality change?. Why MBTI, Enneagram, attachment and HSP results change is covered in the guide to MBTI accuracy, the guide to Enneagram accuracy, the guide to attachment test accuracy and the guide to HSP test accuracy.

When people talk about how accurate a psychological test is, they usually mean two things. Reliability asks whether the same person gets a similar result on a retest (retest reliability) and whether the items move together consistently (internal consistency). Validity asks whether the test measures what it claims to measure, and whether its results connect meaningfully with other tests and with how people actually are. The two kinds of reliability are different properties. A mood scale can have items that agree well with each other yet give a different score on another day, and a set of unrelated items can give the same value every time. That is why internal consistency should not stand in for retest reliability (McCrae et al., 2011). Retest reliability is also a correlation: it shows how well people keep their order across two sittings, not that one person gets the same score again.

The Big Five does not sort people into types. It scores how much of each of five traits someone has: openness, conscientiousness, extraversion, agreeableness and emotional sensitivity (neuroticism in research terms). The researchers behind the BFI-2 wrote that over the past quarter century, scientific consensus on the structure and basic definitions of the five traits led to an explosion of research (Soto & John, 2017a). Still, there is more than one Big Five test, so each one’s reliability has to be checked on its own.

One meta-analysis looked across tests. It pooled 682 retest correlations from 74 samples, 14,923 people in all, retested within two months (intervals of 1 to 8 weeks, about 3.7 weeks on average). The median across the five traits was .816, with extraversion highest at .851 and agreeableness lowest at .778. About 10% of the differences in scores between people was estimated to come from things that vary with the moment of measurement, such as current mood or feelings (Gnambs, 2014, author’s manuscript).

Test by test: the 240-item NEO-PI-R gave correlations of .91 to .93 across the five traits when 132 college students were retested 7 to 14 days later (McCrae et al., 2011). The 60-item BFI-2 showed internal consistency of .83 to .90 per trait in 1,000 adults from English-speaking countries and 470 students, and retest correlations of .76 to .84 when 110 students answered again about eight weeks later (Soto & John, 2017a). Translations report similar levels. The Korean BFI-2 gave .80 to .92 when 421 of 856 employees were retested a month later (Choi et al., 2026), the Japanese version .85 to .90 in 171 students after three weeks (Yoshino et al., 2022), and the French version .84 to .90 in 139 students after four weeks (Lignier et al., 2023).

The mind.io Big Five personality test is none of these. It does not reuse any research test’s items: it is mind.io’s own test of 30 statements written by mind.io (6 per trait, 3 of them worded the opposite way), answered from 1 to 5. The test’s retest reliability and validity have not been studied. The figures above describe research tests; none of them tells you how accurate a mind.io result is.

Big Five tests vary a lot in length. The NEO-PI-R has 240 items (48 per trait) and also measures 30 facets beneath the five traits (McCrae et al., 2011); the BFI-2 has 60 items and measures 15 facets (Soto & John, 2017a). Its short forms from the same developers, the BFI-2-S and BFI-2-XS, have 30 and 15 items (Soto & John, 2017b). The Mini-IPIP, drawn from the public IPIP item pool, has 20 items, 4 per trait (Donnellan et al., 2006, from the abstract), and the TIPI has 10 items, only 2 per trait (Gnambs, 2014, author’s manuscript).

Fewer items mean items that hang together less. In data from 2,000 adults from English-speaking countries recruited online and 423 students, average internal consistency per trait was .86 and .85 for the 60-item BFI-2, .78 and .77 for the 30-item BFI-2-S, and .62 and .59 for the 15-item BFI-2-XS. Agreement with ratings of 230 students by someone who knew them well, mostly friends or partners, also slipped slightly, averaging .55, .53 and .50. The 30-item BFI-2-S’s retest correlations in two student samples (137 and 121 people, about two and three months apart) averaged .76 and .83 (the 15-item BFI-2-XS, .70 and .76), but the authors noted that these came from the same data used to pick the items and may be somewhat flattering. The developers judged that the short forms keep much of the full test’s reliability and validity at the level of the five traits, yet still recommended the full form for most research (Soto & John, 2017b). In the meta-analysis too, the NEO-PI-R’s retest correlations were the highest, reaching .918, while the TIPI, with 2 items per trait, was lower at .664 to .807. The number of items itself showed no separate significant effect in that analysis, though, and the author put part of the gap between tests down to how consistent their items were (random error) (Gnambs, 2014, author’s manuscript).

How a test is built has been studied too. A study that used BFI-2 items to build scales of different lengths and make-ups, and compared them in data from 536 students, found that with fewer than 6 items per trait, links with outside criteria such as behaviour and life outcomes dropped quickly, while beyond 9 items they grew only slowly, so it suggested 6 to 9 items as an efficient range. It also found that balancing items worded in the “agree” direction with items worded in the “disagree” direction reduces how much a habit of agreeing regardless of content can inflate or suppress links with other measures. For high-stakes decisions such as personnel or diagnostic ones, though, the authors added that longer scales with even slightly higher validity can be worth it (Soto & John, 2019).

Translation and samples change the numbers as well. In the meta-analysis, studies outside the United States, which often used tests adapted from another language, had significantly lower retest correlations, and the author speculated that adaptation may have cost some precision (Gnambs, 2014, author’s manuscript). The team behind the Korean BFI-2 noted that an earlier Korean version of the BFI, tested in older adults, had traits with low internal consistency, such as .52 for extraversion and .61 for agreeableness (Choi et al., 2026). In a Turkish online sample where 200 people were retested within four weeks, a short Big Five measure used alongside gave retest correlations as low as .292 to .501 (Yanartaş et al., 2022); that study is covered in the guide to Enneagram accuracy.

The mind.io test has 30 items, 6 per trait, with 3 worded in each direction. In length and balance it sits at the lower end of the range suggested above and has as many items as the BFI-2-S, but its items are different, so the reliability of the BFI-2-S or any other test cannot be carried over to mind.io scores. Nor does it measure facets the way the BFI-2 or NEO-PI-R do. Each trait’s facets are introduced on the trait pages, such as emotional sensitivity.

The longer the gap, the lower the correlation, even for the same test. As the meta-analysis author summarises earlier studies, the same BFI gave retest correlations of .81 to .84 over two weeks and .79 to .89 over two months, but .62 to .70 over three years. Even within eight weeks, longer gaps came with slightly lower correlations for openness, extraversion and emotional sensitivity (Gnambs, 2014, author’s manuscript). Researchers read the extra drop over longer gaps mainly as real change.

What about years? A meta-analysis of 3,217 retest correlations from 152 longitudinal studies found that, with the gap held at 6.7 years, the consistency of people’s order rose from .31 in childhood to .54 in the college years and .64 at age 30, and stopped rising at around .74 between ages 50 and 70 (Roberts & DelVecchio, 2000, from the abstract). A newer meta-analysis that reanalysed those data and added studies published after 2005 (rank-order stability from 189 new samples, 178,503 people) found an average of r = .60: stability rose until young adulthood, with little evidence of further rises after age 25, only .035 more from 25 to 60. It also noted that data from Western countries were overrepresented (Bleidorn et al., 2022). In adulthood, then, people keep their order fairly well, but they are not set in stone.

Day-to-day behaviour moves far more. In studies where people logged their behaviour again and again over two to three weeks, a typical person showed nearly every level of all five traits, yet each person’s average was almost perfectly stable (Fleeson, 2001, from the abstract). A personality test asks about something close to that usual average, but your state while answering still slips into the score a little.

Average change with age, such as conscientiousness rising and emotional sensitivity easing, and studies in which counselling or self-chosen goals shifted personality scores are covered in the guide Can personality change?.

On the validity side, whether Big Five scores connect with real life has been studied a lot. A review of that research lists links between personality traits and happiness, physical and psychological health, relationships with peers, family and partners, occupational choice, satisfaction and performance, community involvement, criminal activity and political ideology (Ozer & Benet-Martínez, 2006, from the abstract). A project that retested 78 such published links in preregistered replications (median sample of 1,504 per replication) found that 87% came out statistically significant in the expected direction, with effects typically about 77% as strong as in the original studies (Soto, 2019). These are group-level tendencies; your score does not decide your life.

Scores have also been compared with how others see people. Among three meta-analyses of ratings by others (263 samples and 44,178 people in all), the one comparing a person’s self-rating with one other person’s rating (about 20,000 to 29,000 people per trait) found that the correlation between the two (uncorrected average) ranged from .29 for agreeableness to .41 for extraversion. The rater mattered a lot: for emotional stability (the opposite end of emotional sensitivity) it was .43 with family members but .08 with strangers, and for extraversion .48 with family and .22 with strangers (Connelly & Ones, 2010). In the BFI-2 research, when 184 students were each rated by someone who knew them well (mostly friends or partners), agreement with their self-ratings ranged from .42 to .69 by trait, averaging .56 (Soto & John, 2017a). Your view and other people’s overlap a good deal, but they are not the same.

Some traits are seen best by you, others by those around you. In a study of 165 people rated by themselves, four friends and strangers, and compared with criteria from behavioural tests, the self was the best judge of traits close to emotional sensitivity (neuroticism), friends of intellect-related traits, and everyone did about equally well on extraversion-related traits (Vazire, 2010, from the abstract). In the meta-analysis, closer relationships made other people’s ratings more accurate, especially for emotional stability, which shows less on the outside (Connelly & Ones, 2010).

The situation you answer in moves scores too. In a meta-analysis of 33 studies, job applicants scored higher than non-applicants on conscientiousness (d = .45) and emotional stability (d = .44) (Birkeland et al., 2006, from the abstract). When the self you want to show slips into your answers, the scores can shift. Comparing yourself with other people or other countries needs care as well. One line of research finds that people rate themselves against similar others around them, which blurs cultural comparisons made with such self-rating scales (Heine et al., 2002, from the abstract), and another found that country averages of self- and peer-rated conscientiousness did not track behavioural and demographic indicators (Heine et al., 2008, from the abstract). The team behind the Korean BFI-2 likewise suggested that the narrower spread of Korean scores, compared with the US samples, may come from Asian respondents choosing middle options more often (Choi et al., 2026).

The mind.io Big Five personality test turns the average of each trait’s 6 items into a score from 0 to 100 (an average of 1 is 0, an average of 5 is 100). Moving a single answer by one step therefore moves that trait’s score by about 4.2 points (4 or 5 after rounding). Scores calculated with the current items are spaced out, too, as in 0, 4, 8, 13 and so on, so 34 to 37 and 64 to 66 never come up. 33 (Low) and 38 (Mid), and 63 (Mid) and 67 (High), are one answer apart.

The result page shows the five scores with Low, Mid or High, but it shows no separate note when a score sits near the edge of a band. So look past the band name and check yourself how far each score is from 33 or 67. The same goes for the result’s title. A trait’s name goes into the title only when the highest trait is at 67 or more; otherwise the title reads “An evenly spread profile rather than one standout trait”. When two or more traits share the highest score, the one that comes first in the order openness, conscientiousness, extraversion, agreeableness, emotional sensitivity is picked. So if two traits are tied or one step apart, the title may change next time.

It is worth spelling out what the score means. A mind.io score is not a percentile against other people; it is your own answers placed on a 0 to 100 scale. Choosing 3 (neutral) for every item gives 50 on all five traits, so 50 does not mean “average”: it means you mostly answered toward the middle. Do not compare it directly with a percentile or standard score from another site. A percentile depends on the group that site compares you with.

Answering habits and your state also go into the score. Each trait on mind.io has 3 items worded one way and 3 the other, so a habit of agreeing regardless of content does not push a score in only one direction. But the day’s mood, whether you pictured yourself at work or at home, and whether you answered as the self you want to show still change your answers. The result page’s comparison with an earlier result says the same: a difference does not mean your personality has changed and may reflect recent state, context or how you framed your answers.

One more thing. Two changes that affect results were made in October 2026. The first is how emotional sensitivity is scored. From the test’s launch in January 2026, the worry items and the calm items of this trait were scored the opposite way round, so the emotional sensitivity scores in that period actually measured emotional stability. The scoring has been put right, and results saved before the fix and old share links are corrected automatically when opened: emotional sensitivity is shown as 100 minus the stored score. An emotional sensitivity you once saw as 25, “Low”, may therefore now show as 75, “High”; that is the calculation being corrected, not your personality changing. So an older result’s emotional sensitivity may be a value such as 37 that the current scoring never produces, and the result’s title, explanations and comparisons that depend on it may look different too. The other four traits were scored correctly from the start. The second change is to the items. The same month, six items that mixed in other traits or asked about something other than a personality trait were replaced, and each trait now has 3 items worded each way. Saved results are not rescored with the new items, and the comparison with an earlier result does not mention the item change, so allow for the partly different items when you compare with a result from before. Both changes are also listed under “Changes that affected results” on mind.io’s Methodology page.

First, look at the scores and their direction rather than the band names. If your extraversion came out at 63 and then 67, the names differ but you are near the same spot. If instead it jumped from the 30s to the 70s, first think back on whether your state or the situations you pictured were different, or whether something big happened in between. If emotional sensitivity in particular came out close to the opposite of before, also check whether the earlier result predates October 2026; the scoring fix from that time is described above.

Second, retake it under similar conditions after some time. Retaking it several times straight away because you dislike the result makes it easy to remember the items and pick the answers you want. Answer with your usual self in mind rather than today alone, and retake it a few weeks later at a similar time. If an earlier result is still saved in the same browser, the result page’s comparison shows how many points each of the five traits moved. The direction over several results tells you more than a single difference.

Third, borrow someone else’s eyes. Asking family or friends who know you well where you seem high or low on the five traits shows which parts of your scores they share and which they don’t. Emotional sensitivity, which shows less on the outside, was judged far better by close others than by strangers (Connelly & Ones, 2010), and one study found you yourself know it best (Vazire, 2010, from the abstract).

Fourth, set it beside a type test. If your E and I keep switching on the MBTI test, check whether your Big Five extraversion sits around the middle. How to read the two together is in the guide to MBTI vs the Big Five.

And what not to use it for. A mind.io Big Five result is a self-reflection aid, not a psychological diagnosis. A high emotional sensitivity score does not on its own mean anxiety or depression. If worry or low mood has unsettled your daily life for more than two weeks, check how you are with the anxiety check-in (GAD-7) or the depression check-in (PHQ-9) and talk with a professional. These check-ins are not diagnoses either. Sorting people, in hiring, evaluations or choosing a partner, is not what this test is for; research finds higher conscientiousness and emotional stability scores when people apply for a job (Birkeland et al., 2006, from the abstract). Researchers also count it as a limit that much Big Five research comes from Western, especially English-speaking, data (Choi et al., 2026; Bleidorn et al., 2022). And when you compare scores with a friend, remember that you are comparing two self-ratings made the same way.

Q. Are Big Five tests scientifically reliable? A. Research tests retaken within a few weeks had a median retest correlation of .816 (Gnambs, 2014, author’s manuscript), and in preregistered replications 87% of 78 links with life outcomes came out again (Soto, 2019). That is about research tests, though. The mind.io test uses its own items and has no reliability data yet, so read it through the scores and their direction over several results rather than the band names.

Q. Is it more accurate than the MBTI? A. No single figure compares the two on the same terms. But because the Big Five scores each trait instead of cutting people into types, a changed result can be read as how many points moved. How the two tests differ is covered in the guide to MBTI vs the Big Five, and why MBTI results change in the guide linked in the last question below.

Q. Is a score of 50 average? Another site gave me a percentile. A. A mind.io score is not a percentile but your answers placed on 0 to 100, so 50 means you mostly answered toward the middle. A percentile depends on the comparison group, and comparing average self-ratings with people from other countries or cultures needs care too (Heine et al., 2002, from the abstract).

Q. My old emotional sensitivity result is the opposite of my new one. A. If the old result is from before October 2026, the scoring fix may be why. Before it, the worry and calm items of emotional sensitivity were scored the opposite way round; saved results now show emotional sensitivity as 100 minus the stored score when opened. The other four traits were always scored in the right direction. If the gap is still large, retake it after some time as described above.

Q. Does a changed score mean my personality changed? A. Differences over a few weeks are mostly explained by measurement wobble or your state at the time (Gnambs, 2014, author’s manuscript). Differences over years include real change as well (Bleidorn et al., 2022). Average change with age and the effects of trying to change are covered in the guide Can personality change?.

Q. How often can I retake it? A. There is no set interval, but a gap of a few weeks or more under similar conditions makes the results easier to compare than retaking it straight away. Look at the direction over several results rather than at one score.

Q. Do MBTI, Enneagram, attachment and HSP 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, the guide to Enneagram accuracy, the guide to attachment test accuracy and the guide to HSP test accuracy.

References: Birkeland, S. A., Manson, T. M., Kisamore, J. L., Brannick, M. T., & Smith, M. A. (2006), A meta-analytic investigation of job applicant faking on personality measures, International Journal of Selection and Assessment 14(4), 317-335, https://doi.org/10.1111/j.1468-2389.2006.00354.x; Bleidorn, W., Schwaba, T., Zheng, A., Hopwood, C. J., Sosa, S. S., Roberts, B. W., & Briley, D. A. (2022), Personality stability and change: A meta-analysis of longitudinal studies, Psychological Bulletin 148(7-8), 588-619, https://doi.org/10.1037/bul0000365; Choi, J., Kim, N., Zhang, B., Park, S. W., Cho, S., Sohn, Y. W., Soto, C. J., & John, O. P. (2026), The Big Five Inventory-2 in Korea: Validation and cross-cultural comparisons with the U.S. and Chinese versions, Assessment 33(5), 866-889, https://doi.org/10.1177/10731911251357466; Connelly, B. S., & Ones, D. S. (2010), An other perspective on personality: Meta-analytic integration of observers’ accuracy and predictive validity, Psychological Bulletin 136(6), 1092-1122, https://doi.org/10.1037/a0021212; Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2006), The Mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of personality, Psychological Assessment 18(2), 192-203, https://doi.org/10.1037/1040-3590.18.2.192; Fleeson, W. (2001), Toward a structure- and process-integrated view of personality: Traits as density distributions of states, Journal of Personality and Social Psychology 80(6), 1011-1027, https://doi.org/10.1037/0022-3514.80.6.1011; Gnambs, T. (2014), A meta-analysis of dependability coefficients (test-retest reliabilities) for measures of the Big Five, Journal of Research in Personality 52, 20-28, https://doi.org/10.1016/j.jrp.2014.06.003 (author’s accepted manuscript: https://timo.gnambs.de/sites/default/files/gnambstimo/publications/gnambs-2014-dependability.pdf); Heine, S. J., Lehman, D. R., Peng, K., & Greenholtz, J. (2002), What’s wrong with cross-cultural comparisons of subjective Likert scales?: The reference-group effect, Journal of Personality and Social Psychology 82(6), 903-918, https://doi.org/10.1037/0022-3514.82.6.903; Heine, S. J., Buchtel, E. E., & Norenzayan, A. (2008), What do cross-national comparisons of personality traits tell us? The case of conscientiousness, Psychological Science 19(4), 309-313, https://doi.org/10.1111/j.1467-9280.2008.02085.x; Lignier, B., Petot, J.-M., Canada, B., De Oliveira, P., Nicolas, M., Courtois, R., John, O. P., Plaisant, O., & Soto, C. J. (2023), Factor structure, psychometric properties, and validity of the Big Five Inventory-2 facets: Evidence from the French adaptation (BFI-2-Fr), Current Psychology 42(30), 26099-26114, https://doi.org/10.1007/s12144-022-03648-0; McCrae, R. R., Kurtz, J. E., Yamagata, S., & Terracciano, A. (2011), Internal consistency, retest reliability, and their implications for personality scale validity, Personality and Social Psychology Review 15(1), 28-50, https://doi.org/10.1177/1088868310366253; Ozer, D. J., & Benet-Martínez, V. (2006), Personality and the prediction of consequential outcomes, Annual Review of Psychology 57, 401-421, https://doi.org/10.1146/annurev.psych.57.102904.190127; Roberts, B. W., & DelVecchio, W. F. (2000), The rank-order consistency of personality traits from childhood to old age: A quantitative review of longitudinal studies, Psychological Bulletin 126(1), 3-25, https://doi.org/10.1037/0033-2909.126.1.3; Soto, C. J. (2019), How replicable are links between personality traits and consequential life outcomes? The Life Outcomes of Personality Replication Project, Psychological Science 30(5), 711-727, https://doi.org/10.1177/0956797619831612; Soto, C. J., & John, O. P. (2017a), The next Big Five Inventory (BFI-2): Developing and assessing a hierarchical model with 15 facets to enhance bandwidth, fidelity, and predictive power, Journal of Personality and Social Psychology 113(1), 117-143, https://doi.org/10.1037/pspp0000096; Soto, C. J., & John, O. P. (2017b), Short and extra-short forms of the Big Five Inventory-2: The BFI-2-S and BFI-2-XS, Journal of Research in Personality 68, 69-81, https://doi.org/10.1016/j.jrp.2017.02.004; Soto, C. J., & John, O. P. (2019), Optimizing the length, width, and balance of a personality scale: How do internal characteristics affect external validity?, Psychological Assessment 31(4), 444-459, https://doi.org/10.1037/pas0000586; Vazire, S. (2010), Who knows what about a person? The self-other knowledge asymmetry (SOKA) model, Journal of Personality and Social Psychology 98(2), 281-300, https://doi.org/10.1037/a0017908; Yanartaş, Ö., Malakcıoğlu, C., Acarkan, İ., & Akça, E. (2022), Validity and reliability of Enneagram personality types and subtypes inventory in a Turkish sample, Anadolu Kliniği Tıp Bilimleri Dergisi 27(3), 248-259, https://doi.org/10.21673/anadoluklin.1104082; Yoshino, S., Shimotsukasa, T., Oshio, A., Hashimoto, Y., Ueno, Y., Mieda, T., Migiwa, I., Sato, T., Kawamoto, S., Soto, C. J., & John, O. P. (2022), A validation of the Japanese adaptation of the Big Five Inventory-2, Frontiers in Psychology 13, 924351, https://doi.org/10.3389/fpsyg.2022.924351