Splitting and Black-and-White Thinking: What Research Actually Shows

Why the label needs precision

"Splitting" is one of the most widely repeated—and least carefully defined—terms in online discussions of relationships. A person describes yesterday's intensely positive encounter and today's devastating criticism; someone replies that a partner is "splitting." Sometimes that shorthand captures a genuinely polarized way of evaluating others. Often it pretends to know a cognitive mechanism from a two-message snapshot. The science does not support diagnosing someone else that way. [1,2]

The more measurable term is dichotomous thinking: a tendency to organize information into stark categories rather than recognize intermediate positions. Idealization is an unusually all-positive assessment; devaluation is an unusually all-negative one. Yet people also change their assessments for perfectly good reasons. Someone who discovers dishonesty, becomes unsafe, notices new evidence, or changes a value judgment is not thereby demonstrating a disorder. [1,2]

This review synthesizes published experiments, a computational model, clinical guidance, and systematic psychotherapy studies. It is not a new pooled meta-analysis and cannot determine what happened within an individual relationship. The mechanisms and treatment studies answer different questions, and combining them loosely would exaggerate what is known. [1–8]

Black-and-white thinking is not an all-or-nothing diagnosis

Everyone occasionally thinks in categories. "That was a disaster" may be ordinary figurative language rather than an enduring trait. During conflict, fear or exhaustion, judgments can temporarily become more extreme. Researchers instead ask whether the person repeatedly has difficulty integrating mixed evidence: that someone might be caring in one context and unreliable in another, or that a decision can be partly good and partly regrettable. [1,2]

This distinction separates a cognitive style from a diagnosis. Borderline personality disorder has broader diagnostic criteria and requires careful longitudinal assessment. It would be unsound to identify BPD from a single episode of angry language or alternating affection and resentment. Other conditions and circumstances, including mood disorders, trauma-related distress, relationship insecurity and coercion, may influence how a person evaluates situations. [4]

Conversely, refusing to use a diagnostic label does not imply dismissing harmful behavior. Threats, controlling acts, intimidation or repeated boundary violations matter whether or not a person has any psychiatric diagnosis. The appropriate level of accountability depends on observable conduct, not a speculative explanation of an absent person's thinking.

What the 2012 social-cognition experiment found

Arntz and ten Haaf designed an interpersonal evaluation task in which participants interacted with supposed mental health trainees in accepting, rejecting and neutral roles. The authors compared 18 participants with BPD, 18 with Cluster C personality disorders and 18 nonpatients. Structured ratings and interviews were used to assess negativity, dichotomous thinking, splitting and complexity of explanations. [1]

The BPD group showed more dichotomous evaluations across conditions, but the authors specifically reported that the evidence for splitting as a unique BPD characteristic was not convincing. They also found no clear reduction in the complexity of interview-based attributions compared with nonpatients. This mixed outcome matters more than a meme-worthy summary that "research proves people with BPD split." [1]

The study was small, its task was artificial, and statistical differences between groups do not classify individuals reliably. It cannot tell us how often a specific pattern happens in daily life, whether it happens under all relationship conditions, or whether it causes a person to mistreat someone. The finding is evidence about an experimentally measured cognitive tendency under specified conditions—nothing broader.

The 2024 model: how extreme beliefs might persist or reverse

Story and colleagues proposed a Bayesian social-inference model in which people form beliefs about others' dispositions and update those beliefs using observed actions and contextual explanations. The model allows strong categorical beliefs, such as "this person is completely good" or "this person is completely bad," to persist even when contradictory evidence appears. For example, a negative action may be attributed to an unfortunate circumstance while a strongly favorable impression remains; after a threshold is crossed, the global appraisal can flip. [2]

That framework offers a testable account of idealization and devaluation rather than merely naming them. The authors report empirical model-fitting results showing group differences in inferred splitting tendencies. The model nevertheless assumes a particular mathematical description of learning. Alternative social, developmental and clinical models can sometimes explain similar observed judgments. A well-fitting computer model is not proof that the brain literally uses those variables or that an everyday event reveals hidden computational states. [2]

The model also suggests something practical but easy to misunderstand: contextual evidence can matter. A late arrival due to a genuine transit disruption is different from the same behavior without explanation. Understanding context may support more nuanced judgments. But providing endless excuses for genuinely harmful conduct is not the clinical goal; context must be weighed alongside reliability, boundaries and safety. [2]

Why idealization can feel so convincing

In a polarized positive evaluation, a person might emphasize signs of affection, infer unusually dependable intentions, or discount contradictory events. Positive judgments often reflect real positive experiences; the issue arises when the conclusion becomes resistant to proportionate correction. When an event later challenges an idealized view, the emotional result may be abrupt disappointment or a complete reversal in appraisal. [2]

The same framework can describe a persistently negative interpretation. An ambiguous message, ordinary delay or conflict may be read as proof of bad intent, while positive actions are discounted or attributed to ulterior motives. Yet some negative interpretations are accurate: if a partner has repeatedly lied, a careful person may justifiably become less trusting. Nuance does not mean trusting someone who has shown themselves unsafe or unreliable.

Research measures patterns across situations; an outsider observing isolated snapshots has little basis to declare which inference mechanism is operating. In practice, a nonjudgmental inquiry into evidence, alternative explanations, and the consequences of actions is usually safer than accusing someone of splitting.

How does emotion regulation relate to polarized judgments?

Strong emotions can narrow what feels immediately plausible. Anger may make negative interpretations more salient; fear of rejection can magnify uncertainty; shame may turn one mistake into a global verdict about personal worth. Such processes can interact with a polarized cognitive style, but there is no single established pathway accounting for every instance of dichotomous thinking. [1,2,7]

Psychotherapy research also uses broad outcomes such as BPD severity, depressive symptoms, anger expression, interpersonal functioning and quality of life. None is an interchangeable direct measurement of splitting. A trial showing less anger after DBT cannot prove that a specific social-inference parameter normalized. Outcome definitions matter. [3,5,7,8]

The 2026 meta-analysis by Li and Mao included seven controlled studies of affective outcomes in BPD. Its anger estimate varied in statistical significance depending on adjustment, and heterogeneity was high. That is a reason to describe its result as tentative rather than convert it into an assured cure for extreme judgments. [7]

What psychotherapy evidence can—and cannot—promise

A 2024 systematic review by Crotty and colleagues identified 25 psychotherapy studies involving 2,545 participants. Several approaches showed benefits relative to treatment as usual for broader BPD outcomes; the strength of evidence ranged from moderate to low depending on treatment and comparison. The review did not find sufficiently strong head-to-head evidence to declare one psychotherapy clearly best overall. [3]

An individual-participant-data synthesis published in 2026 found an average psychotherapy advantage on BPD symptom severity, but it also identified limitations from attrition and risk of bias. Its results support meaningful improvement at the group level without guaranteeing specific trajectories. Likewise, the multicenter BOOTS trial directly compared DBT and schema therapy, informing overall treatment choices rather than answering which method eliminates splitting in all cases. [5,6]

A separate participant-level analysis examined interpersonal functioning, where estimates and potential modifiers were constrained by missing data and multiple-comparison issues. It is methodologically incorrect to convert an average interpersonal improvement into a claim that every participant became reliably flexible in romantic interpretations. [8]

Psychotherapy may help people check assumptions, tolerate mixed feelings, slow impulsive reactions, attend to context, communicate more directly and practice emotion regulation. Those are plausible targets consistent with the approaches studied; the literature cited here does not quantify how much a particular strategy reduces an isolated splitting score. [1–8]

What if I am the person who thinks in extremes?

Begin by replacing certainty about motives with specific observations. Rather than "they hate me," consider "they did not reply to two messages; I feel hurt and do not yet know why." Rather than "I always ruin relationships," consider "I handled that disagreement poorly; I can take responsibility and practice responding differently." Such reframing can make room for evidence without declaring feelings invalid.

It may help to separate three columns—what happened, what meaning you assigned, and what other explanations remain possible. Check whether the new explanation is consistent with repeated behavior. If you are in a coercive or unsafe relationship, the objective is not to rationalize threats or disregard danger. Contextualizing a harmless delay and dismissing an act of intimidation are completely different judgments.

If polarized appraisals create substantial distress or disrupt relationships, an individualized assessment can explore emotional regulation, mood symptoms, trauma, anxiety, sleep and relevant diagnostic questions. Professional help can identify appropriate support without turning a research construct into a self-diagnosis. [3,4]

What if someone I care about switches between praise and criticism?

A useful response is to name concrete behavior and its impact. "The insults during arguments hurt, and I will end the conversation when that happens" is clearer than "You are splitting again." Clear boundaries are compatible with recognizing someone else's feelings. Avoid diagnostic arguments when the other person is upset, and do not attempt to win certainty about their private thoughts. [1,4]

If both people are safe and willing, discussing triggers, interpretation differences and mutually workable repair may be helpful. If one person threatens, controls or intimidates the other, safety and outside support take priority. A mental-health label never makes abuse inevitable, and having a diagnosis does not remove responsibility for harm.

Relationship change also requires more than a vivid apology. Look for sustained behavior, independent support, respect for limits and a willingness to consider facts over time. A model of cognition cannot tell someone whether to stay or leave; ethical advice preserves the person's choices.

Limitations and what stronger research would require

The evidence base is uneven. A small three-group social-cognition study, a mathematically explicit computational account and psychotherapy trials with broad clinical outcomes cannot simply be pooled into one numerical answer. They differ in design, populations, measurement and purpose. [1–3]

The 2012 experimental finding is suggestive but underpowered for strong individual classification. The 2024 model offers testable predictions without establishing a diagnostic threshold or causal neuronal mechanism. The 2026 psychotherapy syntheses address interventions and outcomes that are related but not identical to polarized social judgments. Studies may overrepresent treatment-seeking patients and may not generalize to undiagnosed people or every relationship. [1,2,5–8]

Needed future work includes preregistered replication, ecological sampling across ordinary interactions, longitudinal evaluation of whether judgments become more flexible, careful comparison across diagnoses, and clinical trials that directly measure dichotomous thinking rather than substituting global symptom scores. Researchers should report whether tasks predict meaningful functioning without stigmatizing clinical groups.

Bottom line

Splitting is a hypothesis about how some people organize social evaluations, not a shortcut for diagnosing a difficult partner. The direct social-cognition experiment supported greater dichotomous thinking in its small BPD group but did not convincingly establish splitting as unique to BPD. A later computational model made the hypothesis clearer and more testable without resolving every causal or clinical question. [1,2]

Psychotherapy offers grounds for hope about broad BPD symptoms and functioning. But the responsible conclusion remains conditional: evidence supports working toward more flexible appraisals, accountability and safer communication; it does not justify predicting a person's character or treatment outcome from an online label. [3–8]

Related evidence: Emotional dysregulation across disorders, BPD versus complex PTSD, Loving a partner with BPD, and attachment to people who hurt us.

Browse the full mental-health research library for related evidence reviews.