How MBTI and Big Five differ, and why psychology uses Big Five

Short answer: where do MBTI and Big Five differ?

MBTI and Big Five are not two entirely opposed systems. In direct comparisons, the four MBTI scales correlate strongly with four of the five Big Five factors, while the fifth factor, neuroticism, has no corresponding MBTI scale¹. The core difference lies in how each divides people: MBTI cuts each dimension into two discrete poles and combines them into 16 groups, while Big Five keeps scores on a continuous range and assigns no type label to anyone²³.

In other words, the two tools measure nearly the same content but differ in measurement model. This model difference explains why academic psychology research almost exclusively uses Big Five.

Comparison table: MBTI scales and their corresponding Big Five factors

The table reflects the conclusions of two direct comparison studies. On an adult sample, Furnham (1996) placed MBTI alongside the NEO-PI and recorded strong correlations between MBTI scales and the four Big Five factors of extraversion, openness, agreeableness and conscientiousness; this result strengthens the view of MBTI as a "Big Four" missing the neuroticism dimension¹. With 468 adults, McCrae and Costa (1989) matched MBTI against the five-factor model, collecting both self-ratings and ratings from acquaintances, and reached a similar conclusion: those four scales are ultimately four continuous dimensions lying within four Big Five factors².

Why does psychology choose Big Five over MBTI?

The answer lies in three points: factor structure, continuity of data, and predictive validity.

First, Big Five is the result of factor analysis on lexical lists describing personality, so its structure replicates across many samples and cultures. MBTI was not built that way: it originates from Carl Jung's theory of psychological types and was developed into a questionnaire by Katherine Briggs and Isabel Myers. This is a difference in methodological origin, not in popularity.

Second, psychometric data show personality scores are continuously distributed, not clustered into two discrete groups. If there really were separate "types" of introverts and extraverts, the score distribution would have two peaks. In reality, most people fall in the middle range. In a review published in Review of Educational Research, Pittenger (1993) examined MBTI through Messick's lens of construct validity and concluded that existing evidence was insufficient to support MBTI's underlying assumptions, including the assumption that each person belongs to one discrete type³.

Third, Big Five has a system of predictive validity widely verified in hiring, education and behavioural research. MBTI lacks that foundation. Boyle (1995), in an article in Australian Psychologist, reviewed MBTI's psychometric limitations in reliability, validity and norms, and did not recommend routine use of MBTI in the Australian context⁴.

What are MBTI's psychometric limitations?

The limitations noted in the psychometric literature centre on four points.

• Dichotomising a continuous dimension. MBTI takes a cut point on each dimension and assigns a binary label. Someone scoring near the cut point may be placed in one group or the other because of a small difference, while two people at opposite extremes are treated as the same type. This is a direct consequence of ignoring the continuity of the data³.
• Test-retest reliability. Boyle (1995) raised the issue of MBTI's reliability in the Australian context⁴. This issue is also discussed in Why retaking the MBTI test gives a different group, so it is only mentioned here as a known limitation, not re-analysed.
• Missing neuroticism dimension. MBTI has no scale corresponding to Big Five's neuroticism¹². This is an important dimension in research on mental health and stress coping, so its absence is a gap in measured content.
• Norms and validity. Boyle (1995) also raised the norms issue⁴. Pittenger (1993) concluded there was insufficient evidence for claims about MBTI's practical utility³.

How is Big Five used in research?

A concrete example of how Big Five operates in research practice is the machine-learning study by Kachur and colleagues (2020) in Scientific Reports. The authors trained an artificial neural network on a large dataset to predict self-reported Big Five scores from static portrait images. Participants were 12,447 volunteers who provided 31,367 facial images and completed a Big Five self-assessment⁵.

The result: the highest correlation between observed and predicted scores was on the conscientiousness factor (0.360 for men and 0.335 for women), and the mean effect size was 0.243, higher than results of previous studies using selfie images⁵. The authors concluded the results support the feasibility of predicting multidimensional personality profiles from static facial images using an artificial neural network trained on a large labelled dataset⁵.

This result should be read at the right level: it is level B evidence, meaning a single study with a modest effect size, not yet independently replicated at comparable scale. The figure 0.243 does not mean you can tell someone's personality from a photo; it means the model performs significantly better than chance. In addition, the data were collected in Russia, with participants voluntarily submitting photos, so it cannot be generalised to other populations without replication⁵.

A notable methodological point: precisely because Big Five gives continuous scores on each factor, the authors could use it as a target variable for the prediction model and compute Pearson correlations. With a discrete type system, training and evaluating the model would be far harder to interpret. This is an example of how the choice of measurement framework directly affects the kinds of research questions that can be asked.

What can be said about the belief effect?

One point needs to be clear: MBTI can produce real effects, but those effects arise from belief rather than from the tool correctly detecting an essence. When a person believes they belong to the introvert group, they may actively seek quieter environments, read more, or prepare more carefully before speaking. Those behaviours, over time, can produce exactly the kind of behaviour that group is described as having. This is a form of self-fulfilling prophecy noted in research on the relationship between appearance and personality⁵.

In other words, this should not be called "MBTI guessing correctly", nor should it be denied that the experience is real for the user. What is being discussed here is testing and data, not the believer.

Where evidence is still missing

One gap should be stated clearly: the source list for this article contains no study comparing MBTI and Big Five on a Vietnamese sample. The results cited above come from Western adult samples (Furnham 1996; McCrae and Costa 1989), an Australian sample (Boyle 1995), or a Russian sample (Kachur and colleagues 2020)¹²⁴⁵. So it cannot be inferred that the factor structure or correlation coefficients found in those samples will replicate exactly among Vietnamese people. This is an open question for domestic research.

Is MBTI considered wrong?

No. This article does not score MBTI as "wrong". MBTI is a self-discovery tool with a clear structure, used by millions and with its own utility in opening dialogue about personality differences. The issue lies elsewhere: MBTI is not a diagnostic tool, and it was not designed as a foundation for comparative psychology research.

If you want to understand what MBTI is and how the 16 groups are determined, see What is MBTI?. If you are interested in why retaking the test can change your group, Why retaking the MBTI test gives a different group has analysed the test-retest reliability aspect. The article you are reading focuses on the remaining aspect: measurement structure and the place of the two systems in research.

The conclusion stands: MBTI measures nearly the same four Big Five factors but in discrete type form, missing the neuroticism dimension; Big Five is the research standard because of its stable factor structure, continuous scales and more widely verified predictive validity. The two tools serve two different purposes, and which one to choose depends on the question you want to answer.