Matt Baker Says Personality Explains Atheism. His Own Dissertation Tells a Different Story.

Matt Baker has an appealing explanation for why Christians become atheists: their personalities make them unusually willing to question what they were taught.

In his UsefulCharts episode, Are Atheists Immoral, Arrogant, and Angry?, Baker challenges the familiar accusations that atheists reject religion because they are selfish, arrogant, angry with God, or damaged by poor relationships with their fathers.

His alternative is more flattering. People who become atheists, he argues, tend to prefer logical decision-making and are less afraid to change their minds.

The evidence comes from his doctoral research at the University of Warwick, where he compared the personalities of thousands of former churchgoers with those who remained religious.

Baker's central finding is that people with a Thinking–Perceiving personality preference, represented by the letters TP, have approximately twice the adjusted odds of belonging to his atheist group.

He calls personality the strongest predictor of atheism.

I examined the original dissertation. The association is real within his sample. But the claim that personality is the strongest explanation rests on a questionable interpretation of his own statistical results.

And the more interesting evidence may be what Baker's presentation leaves in the background: the influence of religious experiences, the people who abandoned church without abandoning God, and the limits of a survey recruited heavily through atheist communities.

1. The Strongest Predictor May Not Be Personality

The most revealing evidence appears on page 193 of Baker's dissertation, in Table 29.

There, he presents a statistical model examining five variables associated with becoming an atheist church-leaver.

Baker identifies personality as the strongest predictor because its odds ratio, 2.00, is larger than the other reported values.

That comparison is not enough to establish his conclusion.

An odds ratio describes the association associated with a specified change in a variable. For TP, the comparison is between two categories: having that preference or not having it.

For church experience, the reported odds ratio describes a change of just one point on a scale constructed from ten survey questions.

Those are not equivalent changes.

Consider what happens when we examine a more substantial difference in church experience.

Baker reports an odds ratio of 0.87 for each one-point improvement in the church-experience score.

A ten-point improvement therefore corresponds to:

0.87¹⁰ ≈ 0.25.

In the opposite direction, a ten-point deterioration corresponds to approximately four times the odds of belonging to the atheist group, holding the model's other variables constant.

That is twice the odds ratio reported for TP personality preference.

This calculation uses Baker's published coefficient. It does not establish that church experience is causally more important, nor does it prove that a ten-point difference is the ideal comparison.

It establishes something narrower and more damaging to his ranking: the apparent superiority of personality depends on comparing changes measured in different units.

A meaningful assessment would compare standardized effects, changes across realistic ranges, predictive performance, or the loss of explanatory power when each variable is removed from the model.

Baker's presentation does not establish that hierarchy.

The dissertation contains another revealing number.

The Wald statistic, a measure used to test whether a regression coefficient differs from zero, is 1,095 for church experience and 66 for personality.

A larger Wald statistic does not automatically establish greater practical importance. But these results certainly do not justify dismissing church experience as a distant secondary factor.

There is also an arithmetic inconsistency in the published table. The years-of-attendance coefficient is −0.055, its standard error is listed as 0.250, and its Wald statistic is 107.9.

Those figures cannot all be correct under the conventional calculation of a one-coefficient Wald statistic. The listed coefficient and standard error would produce approximately 0.048, not 107.9.

A misplaced decimal is a plausible explanation. It is not evidence of fabricated results.

It is another reason the statistical reporting deserves scrutiny before the conclusions are presented as settled.

The problem is not that Baker found an association between personality and atheism.

The problem is that he declared a winner without demonstrating that the competitors had been measured on comparable terms.

2. His 23,697 Respondents Were Not a Representative Population

Baker's dissertation reports 23,697 completed questionnaires, collected between September 2011 and March 2012.

His principal comparison involved 10,515 atheist church-leavers and 2,326 continuing churchgoers.

These are substantial numbers. But sample size cannot repair a recruitment problem.

Baker used an online snowball survey. Participants were recruited through social networks, bloggers, forums, religious communities, and targeted advertisements.

The dissertation identifies the origins of website traffic:

• 40% came through Facebook posts shared by friends.
• 24% came through Reddit's atheist category.
• 14% came through atheist blogger P. Z. Myers's Pharyngula.

Together, those sources accounted for 78% of website visitors.

Baker explicitly cautioned that visitor statistics were not the same as completed-survey statistics. That distinction matters.

But the recruitment pattern raises an obvious question.

Are these results representative of former Christians generally, or do they disproportionately describe people active in online communities where intellectual skepticism, religious criticism, and particular personality styles may already be common?

We cannot answer that question from the published recruitment figures alone.

And that is precisely the difficulty.

A self-selected survey can identify real associations among respondents without establishing that those associations hold across the wider population.

The Pew Research Center's investigation of online nonprobability surveys found substantial differences in accuracy between sampling methods and providers. Statistical sophistication cannot automatically eliminate the consequences of unknown selection probabilities.

Baker's two principal groups also differed substantially in composition.

Approximately 75% of his atheist church-leavers were male, compared with about 51% of the continuing churchgoers.

The atheist group was also somewhat younger, more heavily white, and more European.

Baker recognized the sex imbalance and analyzed several personality comparisons separately for men and women.

But his headline logistic regression combined the sexes, and sex does not appear among the five reported predictors.

That matters because Thinking–Feeling preferences also vary by sex in the personality framework Baker used.

It does not prove that the personality association disappears after demographic adjustment.

It means the published headline model leaves an important potential source of confounding insufficiently addressed.

A study can contain thousands of statistically significant observations and still leave us uncertain about the people it represents.

3. People Who Left Church but Kept Believing in God Complicate the Story

Perhaps the most useful neglected group in Baker's dissertation consists of 1,977 people who stopped attending church but continued believing in God.

These people matter because they separate two different decisions.

One is abandoning a religious institution.

The other is abandoning religious belief.

Those decisions are not interchangeable.

Baker's principal comparison places atheist church-leavers against believers who continue attending church. Consequently, it combines differences in religious belief with differences in religious participation.

But what happens when people leave the institution without becoming atheists?

Baker actually examined that question.

Among female theist church-leavers, the TP combination was overrepresented by a factor of 1.46 compared with continuing churchgoers.

Among male theist church-leavers, the corresponding factor was 1.50.

The associations were stronger among atheist church-leavers: 2.14 for women and 1.89 for men.

Baker interpreted the theist church-leavers as occupying an intermediate position between continuing churchgoers and atheist church-leavers.

That is an important qualification.

It indicates that the personality pattern is associated not only with disbelief but also with leaving church while retaining belief.

The stronger association among atheists means the findings cannot simply be reduced to church attendance.

Still, several explanations remain possible.

Perhaps TP individuals are more comfortable questioning religious doctrines.

Perhaps they have greater difficulty fitting into church communities whose social and organizational cultures favor other personality preferences.

Perhaps both mechanisms operate together.

Baker himself discusses the possibility that certain personalities may feel uncomfortable or isolated in religious settings.

This is a considerably more complicated story than personality producing atheism.

And it raises a question his headline cannot answer:

Are TP individuals especially inclined to reject God, or especially inclined to reject the institutions claiming to represent him?

His dissertation contains evidence relevant to both explanations. It does not decisively separate them.

4. Myers-Briggs Categories Are a Fragile Foundation for Explaining Religious Change

Baker describes his findings using the familiar Myers-Briggs personality categories.

However, his dissertation did not administer the proprietary Myers-Briggs Type Indicator.

It used the Francis Psychological Type Scales, a separate instrument built around the same four personality dimensions.

Baker acknowledged this distinction in the dissertation.

The instrument's reported internal-consistency coefficients in his study were:

• Extraversion–Introversion: 0.80
• Sensing–Intuition: 0.65
• Thinking–Feeling: 0.70
• Judging–Perceiving: 0.74

These results show that the questionnaire was not devoid of measurement consistency.

They also do not settle whether dividing human personalities into sixteen discrete categories is the best way to explain complex religious decisions.

That criticism predates Baker's research.

In a 1989 study published in the Journal of Personality, psychologists Robert McCrae and Paul Costa examined the Myers-Briggs framework alongside the Five-Factor Model.

Their findings did not support the idea that personality consists of sharply separated types. Instead, the Myers-Briggs dimensions behaved more like continuous personality traits.

They also questioned the interpretation of the Judging–Perceiving dimension.

Importantly, they found meaningful correlations with established personality dimensions, meaning the instrument was not simply measuring nothing.

The Myers-Briggs Foundation disputes sweeping claims that its assessments lack scientific value. It cites studies reporting acceptable reliability and argues that critics sometimes confuse the validity of dimensional scores with disagreements over personality-type theory.

That defense deserves recognition.

But it does not establish that someone classified as INTP possesses a distinct psychological mechanism that makes atheism more likely.

And Baker's research did not independently demonstrate such a mechanism.

He found an association between a personality classification and religious identification.

He then interpreted that association through ideas about rational decision-making and intellectual independence.

Those interpretations are plausible hypotheses.

They are not direct observations of why respondents changed their beliefs.

5. Baker Measured Personalities After the Religious Decisions Had Already Happened

This is the most fundamental limitation.

Baker surveyed adults about their current beliefs, current personality preferences, and earlier religious experiences.

He did not measure their personalities before they became atheists and then follow them through the process of losing faith.

That distinction is crucial.

Imagine someone grows up religious, spends years questioning church doctrine, becomes involved in skeptical communities, and eventually stops believing in God.

When that person completes a personality questionnaire years later, the answers may reflect longstanding personality characteristics.

But they may also reflect changes in outlook, social identity, behavior, or self-understanding.

Baker's survey cannot establish which came first.

He understood this problem when examining paternal relationships.

He acknowledged that damaged relationships with fathers might contribute to atheism, but becoming an atheist might also damage those relationships.

The same caution should apply to personality.

A characteristic measured after deconversion cannot automatically be treated as the cause of deconversion.

Subsequent research makes that limitation particularly important.

In a study published in the Journal of Personality in 2024, Wiebke Bleidorn and colleagues analyzed eleven waves of longitudinal data from approximately 20,000 Dutch adults.

Their analysis included 450 people who left Christianity.

They found that deconversion was associated with small declines in agreeableness, conscientiousness, and openness at the time of religious change.

The researchers did not establish that abandoning religion causes those personality changes.

But their findings demonstrate why timing matters: personality and religious identity can change in connection with one another.

A further longitudinal study by Zhuo Job Chen and Heinz Streib, published online in 2025, examined religious deconversion using a larger cross-sectional dataset and a longitudinal subsample of 502 participants.

It identified weaker prior belief in the truth of religious teachings and lower religious identification as important predictors of subsequent deconversion.

Personality mattered too. Greater openness to experience and lower extraversion were associated with later deconversion, but religious commitment showed more robust predictive effects in that study.

The authors also identified increased need for cognition as an outcome associated with deconversion, rather than simply assuming that an intellectual disposition must precede it.

Neither later study directly replicates Baker's sample or his exact personality measure.

They therefore cannot disprove his specific correlation.

But they demonstrate that the relationship between personality and religious change is more complicated than the explanation offered in his video.

6. His Defense of Atheists Is More Persuasive Than His Explanation of Atheism

Baker's episode takes aim at several unflattering claims about atheists.

Lee Strobel suggested that immoral pleasures can discourage people from embracing Christianity.

Dinesh D'Souza portrayed atheists as unusually arrogant.

Paul Vitz proposed that troubled relationships with fathers help explain atheism.

Baker tested measures related to those claims and argued that they were inadequate explanations for religious disbelief.

Much of that criticism is reasonable.

But his results require careful interpretation.

For example, the dissertation reports that atheist respondents scored somewhat higher on the selfishness measure than continuing churchgoers.

Among women, the average scores were 20.08 for atheists and 17.73 for churchgoers.

Among men, they were 21.81 and 19.81 respectively.

The differences were statistically significant.

Baker nevertheless rejected the claim that selfishness was a major explanation for atheism, pointing out that both groups scored toward the altruistic end of the scale.

That qualification matters.

The study measured self-reported altruistic dispositions. It did not observe actual charitable behavior, verify dishonesty, or independently assess the moral character of its participants.

A difference in questionnaire scores is not proof that one group has superior morals.

Nor is it evidence that an individual's religious position was caused by moral shortcomings.

His findings concerning arrogance were similarly limited.

Atheists scored somewhat higher, but Baker did not regard the differences as sufficient to support the sweeping claim that atheism results from arrogance.

The anger test was narrower still: it measured general anger, not anger specifically directed toward God.

Baker is right to question people who turn disagreement with religion into an accusation against someone's character.

But there is an inconsistency in his broader reasoning.

He correctly resists interpreting modest differences in selfishness and arrogance as evidence of their importance in causing atheism.

Yet he is much more willing to interpret a personality association as the central explanation for why atheists abandoned Christianity.

One set of psychological correlations receives caution.

Another receives a compelling story about rationality, truth-seeking, and intellectual courage.

The evidence does not justify treating those explanations so differently without further testing.

7. His Own Research Contains a More Interesting Religious Explanation

Baker's dissertation provides evidence that former believers' childhood religious environments were not all alike.

Participants who grew up in more conservative denominations often reported stronger religious emphasis but more negative church experiences.

Those from more liberal denominations tended to report less religious emphasis but more positive experiences.

Among atheist church-leavers, Jehovah's Witnesses, Eastern Orthodox churches, Pentecostal churches, Roman Catholic churches, and Baptist churches were among the denominations associated with the least positive reported childhood experiences.

Methodist, Presbyterian, Anglican, Lutheran, and United Church of Canada backgrounds were among those associated with more positive experiences.

These are comparisons within Baker's surveyed former believers, not representative rankings of entire denominations.

Nevertheless, the pattern suggests more than one possible route away from religion.

A person raised in a liberal religious environment might gradually abandon belief because the religion was never especially central to family life.

A person raised in a conservative environment might maintain intense religious involvement but ultimately reject an institution experienced as restrictive, alienating, or harmful.

Those are interpretations, not causal findings established by the survey.

But they are interpretations grounded in documented differences between religious environments.

And they deserve at least as much attention as the proposition that some people are natural rationalists who simply follow the truth wherever it leads.

Religious deconversion may involve intellectual disagreement, emotional experience, family relationships, institutional culture, and individual personality.

The relative contribution of those factors is an empirical question.

It cannot be settled by assigning a person four letters.

The Verdict: A Real Association, an Unsupported Crown

Matt Baker deserves credit for investigating a subject frequently burdened by religious prejudice.

His survey was extensive. His dissertation contains useful distinctions, substantial statistical work, and more qualifications than his short video could reasonably accommodate.

But the most important claim in the episode does not survive close examination in its strongest form.

Baker establishes that TP personality preferences are associated with atheism in his surveyed population. He does not establish that personality is the strongest explanation for becoming an atheist.

His ranking relies on comparing odds ratios expressed in different units.

His respondents were recruited through a nonprobability online survey, with substantial traffic originating in atheist communities.

His comparison groups differed demographically.

His own data show that people who left church while retaining belief in God also exhibited the personality pattern.

And his cross-sectional research cannot establish whether personality preceded religious deconversion.

Later longitudinal studies reinforce the need to examine religious commitments, experiences, personality changes, and their timing together.

None of this means personality is irrelevant.

It means Baker has offered an interesting piece of the explanation and presented it as something closer to the answer.

The distinction matters because his argument risks replacing one oversimplified account of atheists with another.

Religious critics sometimes portray atheists as people whose moral or emotional deficiencies prevent them from accepting God.

Baker's preferred interpretation risks portraying them as people whose superior willingness to question inherited beliefs allows them to escape religion.

Both stories assign psychological meaning to religious identity that the available evidence cannot fully establish.

An atheist may be thoughtful, careless, charitable, arrogant, intellectually curious, or deeply attached to a community.

So may a Christian.

The challenge for the psychology of religion is not to award one group the more flattering personality.

It is to discover why actual people change their beliefs.

Baker's research contributes to that question. His confidence in personality as the leading answer goes further than his evidence permits.

References

1. Original video — Are Atheists Immoral, Arrogant, and Angry? (Psychology of Atheism Part 3)
UsefulCharts, Matt Baker.
Watch the original video
The work under examination, including Baker's arguments about selfishness, arrogance, anger, paternal relationships, and TP personality preferences.

2. Psychological Type and Atheism: Why Some People Are More Likely Than Others to Give Up God
Matthew James Baker, University of Warwick, February 2015.
Read the complete doctoral dissertation
The primary source for the survey's recruitment methods, group definitions, personality measurements, statistical results, and conclusions; its cross-sectional design prevents causal interpretation.

3. Reinterpreting the Myers-Briggs Type Indicator From the Perspective of the Five-Factor Model of Personality
Robert R. McCrae and Paul T. Costa Jr., Journal of Personality, March 1989.
Read the research
Found evidence for meaningful personality dimensions but not sharply separated psychological types, challenging the theoretical interpretation of Baker's categories.

4. MBTI Reliability and Validity
Myers & Briggs Foundation.
Read the foundation's research assessment
Presents evidence supporting the reliability of Myers-Briggs assessments and qualifications to common criticisms; it does not independently validate Baker's specific causal interpretation.

5. Evaluating Online Nonprobability Surveys
Pew Research Center, May 2, 2016.
Read the methodology report
Explains the representativeness problems associated with self-selected online surveys and documents variation in their accuracy.

6. Psychological Change Before and After Religious Conversion and Deconversion
Wiebke Bleidorn, Madeline R. Lenhausen, Ted Schwaba, and Christopher J. Hopwood, Journal of Personality, published online September 2023; August 2024 issue.
Read the longitudinal study
Tracks religious and psychological changes across eleven survey waves, finding generally small personality changes around conversion and deconversion rather than a simple fixed-personality explanation.

7. Characteristics, Predictors and Outcomes of Religious Deconversion: A Cross-Sectional and Longitudinal Analysis
Zhuo Job Chen and Heinz Streib, The International Journal for the Psychology of Religion, published online June 15, 2025; 2026 volume.
Read the published research
Identifies prior religious commitment and personality traits as predictors of deconversion, while distinguishing characteristics present before religious change from outcomes afterward.

8. Automated Interpretation of Indices of Effect Size
R package effectsize, statistical documentation hosted by ETH Zurich's CRAN mirror.
Read the statistical guidance
Explains that unstandardized logistic-regression odds ratios depend on the units of their predictors, supporting the criticism of ranking Baker's variables by their raw odds ratios.