Crime Linkage: Can Behavior Connect Crimes to the Same Offender?
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
Ukrainian Psychological Hub · Published October 11, 2026 · Investigative Psychology & Crime Analysis.
Crime linkage is the systematic comparison of offenses to assess whether two or more incidents may have been committed by the same offender. Investigators compare behavior, place, timing, victim selection, and contextual features; researchers test whether those comparisons distinguish known same-offender pairs from different-offender pairs. Crime linkage can identify valuable investigative leads. Its conclusions must remain proportional to the quality of the data, the method’s tested accuracy, and the alternative explanations for similarity.
The short answer to the title’s question is yes: behavior can help connect crimes, but a shared behavioral pattern is neither a unique personal identifier nor proof that one person committed every offense in a proposed series. The latest 2026 meta-analysis of 29 studies supports some crime-type-specific linkage methods while emphasizing substantial gaps in operational accuracy research. Davies and Woodhams’ review of investigative practice likewise distinguishes evidence for the underlying behavioral assumptions from evidence that a linkage decision improves real investigations.
What is crime linkage?
In forensic and investigative psychology, behavioral crime linkage refers to the evaluation of similarities and differences across offense events to decide whether they may share an offender. It is also called behavioral case linkage, crime series linkage, or linkage analysis, although agencies use these labels differently. Its core question is about a relationship between incidents: are incidents A and B plausibly attributable to the same offender?
The unit being compared is typically a crime pair or an incident against an existing series. The information may include how a person approached a target, overcame obstacles, used force or threats, moved through an environment, selected an opportunity, obtained property, exited, or communicated during the offense. The comparison can incorporate spatial and temporal information. Burrell, Costello, and Woodhams’ 2024 methodological review describes a broad range of preprocessing choices, behavioral representations, prediction models, and accuracy measures, underscoring that crime linkage names a family of methods rather than one standardized test.
A successful investigative link can motivate further checks: compare trace evidence, identify overlooked witnesses, exchange information among agencies, or look for a shared suspect. A link remains a hypothesis until independently tested. The mere fact that an analyst found crimes similar does not establish a perpetrator’s identity, a clinical diagnosis, criminal responsibility, or a legal finding.
The two psychological assumptions: consistency and distinctiveness
Behavioral consistency within an offender’s series
Consistency means that aspects of an offender’s behavior are sufficiently recurrent across offenses to be recognizable. Repetition can arise from learned routines, familiarity with environments, favored approaches, access to opportunities, or recurring constraints. Yet offending is adaptive. Victim resistance, security measures, location, accomplices, changing resources, intoxication, and chance can alter visible actions. An offender may change procedures after police publicity or a failed attempt.
Consistency is therefore an empirical question about particular behaviors and crime types, not a promise that a person repeats a fixed script. The foundational review of crime-linking psychology found that some offender-initiated behaviors showed greater consistency than behaviors dictated by circumstances. The strength of consistency depends on what behavior is measured, how reliably it is recorded, and the population sampled.
Behavioral distinctiveness between offenders
Distinctiveness means that an offender’s observable behavior differs enough from other offenders’ behavior for the pattern to help separate linked from unlinked incidents. A common feature such as nighttime offending or wearing gloves contributes little on its own when many offenders do the same. A conjunction of less common contextual choices may convey more information, but unusual behavior still can be imitated, coincidental, or caused by shared circumstances.
Consistency without distinctiveness produces many false links: several people may follow the same common routine. Distinctiveness without consistency misses genuine links: a highly unusual act may appear in only one offense. Sound linkage compares both, including dissimilarities that contradict an initial hypothesis. In a multinational sexual-offense study, some same-offender pairs were distinguishable through behavior, while prediction performance changed when the researchers incorporated harder-to-link, less selectively solved cases.
Crime linkage, offender profiling, geographic profiling, and forensic identification
These activities answer different questions. Crime linkage asks whether offenses share an offender. Offender profiling attempts to infer characteristics of an unknown perpetrator from crime information. Geographic profiling estimates likely activity or anchor areas from spatial patterns. Forensic identification uses independently validated physical, biological, or digital evidence to evaluate associations between people, objects, and events. One investigation may use several approaches, but the evidentiary strength of one cannot be transferred automatically to another.
Crime linkage does not require an elaborate personality profile. It can function as a comparative, data-centered decision-support process that ranks incidents for review. Conversely, a dramatic narrative about an offender’s motives does not make a linkage more accurate. The National Crime Agency’s major-crime investigative support description separately describes behavioral investigative advice and geographic profiling, reflecting their different investigative tasks.
DNA associations, ballistic comparisons, fingerprints, communications, or other independently assessed information can corroborate or challenge behavioral links. Their own evidentiary limits remain applicable. An investigative system may flag a cluster based on similarities long before evidence supports formally accusing any person.
How analysts compare crimes in practice
1. Define the investigation and relevant comparison population
The starting point is a precise task: searching a database for possible related cases, checking whether a specified group forms one series, or deciding whether a new incident fits an established series. Analysts must also decide which offense types, locations, times, and records belong in the searchable universe. Search rules determine which alternatives are visible. A narrow search can miss cross-jurisdiction connections; an excessively broad search can generate large volumes of superficially similar events.
2. Reconstruct each incident from attributable records
Analysts collect descriptions from crime-scene documentation, investigative reports, interviews, timestamps, location records, and available forensic findings. A field should distinguish observed behavior from witness inference and from a later investigator’s interpretation. Descriptions such as “used a vehicle” may be reliable; descriptions such as “was meticulous and calculating” are psychological interpretations that require independent support. For a separate treatment of evidence-producing interview procedures, see Investigative Interviewing vs. Interrogation.
Missing information requires explicit codes such as unknown, not recorded, genuinely absent, or inapplicable. Those categories are not interchangeable. A detail absent from a police report may have occurred but gone undocumented. Quality checks, consistent coding rules, version histories, and blinded recoding where practical reduce the risk that an analyst manufactures similarity from ambiguous records. The UK Serious Crime Analysis Section specifically describes systematic case coding and comparison in its Violent Crime Linkage Analysis System.
3. Compare behavior, location, time, and combinations of features
Relevant domains may include entry method, victim approach, degree and type of control, verbal interaction, property selection, spatial range, time of day, sequence of actions, and escape. Analysts should examine the evidentiary meaning of each feature and its frequency among plausible alternative offenders. The same observation can carry different information in commercial burglary, robbery, and sexual assault.
The analyst must also examine discordant details. A candidate link becomes weaker if the acts differ in ways that are inconsistent with ordinary situational adaptation, although a single difference does not automatically disprove common authorship. Analysts should avoid treating every resemblance as a confirmation while explaining away every dissimilarity.
4. Generate a provisional linkage hypothesis and test alternatives
A defensible report states the comparison set, variables, exclusions, observed patterns, plausible environmental explanations, limitations, and confidence justified by validation research. Other hypotheses should remain live: two offenders copying a known practice; related offenders acting together; similarities created by media publicity; common features caused by local security, housing layout, or opportunity; or a genuinely linked offender changing methods.
Follow-up investigation checks for independent support. A behavioral link is most useful when it directs investigators toward discriminating evidence rather than reinforcing a preferred narrative. Investigative notes should preserve the distinction between a possible link, a provisionally supported link, and an independently established attribution.
What behaviors can help link crimes?
Modus operandi and combinations of actions
Modus operandi (MO) describes how an offense is carried out. It includes actions needed to gain access, control an event, achieve a goal, and leave. Some MO elements are common solutions to common problems; others reflect relatively stable preferences or opportunities. Analysts increasingly examine sets of MO behaviors rather than seeking a single supposedly unmistakable clue.
The 2026 meta-analysis found that aggregating MO-related behavioral domains could support linkage in sexual offenses. This result is conditional on the crime class and data representation. It is not a warrant to assume that the same behavioral variables perform equally well for homicide, arson, robbery, and vehicle crime.
Geography and travel behavior
Travel distance, proximity between crime locations, and repeated use of particular areas can help separate some same-offender incidents from unrelated events. Spatial information can be comparatively reliable when incident locations are accurately known, but distance alone reflects opportunity structures, transport, population density, and investigation boundaries. An offender may also travel for work or intentionally change operating areas. For the related spatial-location question, see Geographic Profiling: Psychology, Spatial Behavior, and Investigative Use.
Research synthesized in Kelkar, Jones, and Woodhams (2026) found relatively strong support for geographical predictors in certain acquisitive crimes. Whether a feature performs well remains specific to the crime category, region, and validation sample.
Timing, targets, victim interaction, and sequence
Time of day, intervals between incidents, days of the week, target characteristics, and the sequence of actions may add discriminative information. Yet they can encode environmental regularities rather than offender preferences: store hours, transport routes, public events, or the composition of a neighborhood. Target selection needs particular care because the people affected are more than analytical variables, and apparent similarities can reflect where opportunities were available rather than any stable offender trait.
In offenses involving direct interaction, specific observable actions may be relevant, but witness reporting, stress, memory, and inconsistent documentation complicate coding. These factors determine what researchers can actually establish from an event record. Behavioral description must be kept separate from attributing motives or psychological diagnoses.
The ‘signature’ idea and ritualized behavior
In popular accounts, a signature is portrayed as a recurring, psychologically revealing act that uniquely belongs to an offender. That metaphor is easy to misuse. Empirical work has questioned the reliability of specific individual behaviors as repeatable markers. A study of 450 serial homicide cases involving 90 offenders reported difficulties linking crimes through isolated acts and challenged sweeping signature assumptions. Repeated behavior can be analytically interesting without constituting a behavioral fingerprint.
The term “signature” may have a specialized operational history in some agencies, but its use in a report should be defined. A claim that a particular act reveals a stable inner need or personality trait is an additional inference, one that requires evidence beyond case resemblance. Crime linkage can be assessed without endorsing speculative psychological narratives.
What does scientific research actually show?
Foundational studies provide qualified support
Research has found that linked pairs can, on average, be more behaviorally similar than unrelated pairs in certain sampled crime categories. This supports the two necessary principles to a degree, but it does not itself show that every serial offender has a recognizable pattern or that working analysts can reliably identify links in unknown cases. A method can detect statistical differences between groups while making many incorrect case-level decisions.
The Davies and Woodhams 2019 practice review was explicit that evidence supporting consistency and distinctiveness does not automatically validate the operational use of linkage. Research on the tasks, analysts, records, and decision rules used in the field is a separate requirement.
A multinational sexual-offense study illustrates both promise and difficulty
One influential study evaluated 3,364 sexual offenses across five countries, including 668 series. It incorporated solved series, unsolved cases linked by DNA, and solved single incidents. Its discrimination performance was approximately AUC .86 in one analysis, .85 after including more representative categories of cases, and .79 among unsolved DNA-linked series alone. The change matters: cases that are unusually easy to solve may also be unusually easy to link. Read the original study.
These AUC values describe the ranking of linked and unlinked pairs in the specified study designs. They do not mean that 86% of real case links will be correct, that a new suspect is 86% likely to be guilty, or that a particular database search has that accuracy. AUC must be translated into the decisions and base rates actually encountered by a service.
The 2026 meta-analysis changes the evidence map
Published June 29, 2026, the Kelkar–Jones–Woodhams meta-analysis synthesized 29 studies evaluating statistical models that predict behavioral crime links. The evidence was concentrated in sexual offenses and burglary. The most effective predictors and modeling approaches depended on crime type; geographical domains were comparatively useful for acquisitive offending, and aggregated MO domains for sexual offenses.
The authors also identified a practical reporting weakness: many papers supplied AUC but not enough hit-rate and false-alarm information to reconstruct decision outcomes. For homicide, robbery, car theft, and arson, the evidence base remained less developed. This matters for operational transfer. A model that performs adequately in one offense population cannot be assumed valid in another simply because both are called crime linkage.
The 2025 decision-support study confronts real base rates
A 2025 study in the Journal of Quantitative Criminology examined crime-linkage modeling in a setting with an extremely large number of candidate unlinked pairs. The authors discuss precision–recall evaluation because a model may appear discriminative under ROC analysis yet produce a poor practical yield when true links are very rare. In their described pairwise dataset, 1,907 linked pairs were compared with approximately 59.6 million unlinked pairs. That imbalance illustrates why ranking quality alone is insufficient.
There is no single universal accuracy percentage for behavioral crime linkage. The scientific answer depends on the comparison task, the base rate of true links, the offense category, data quality, analytical model, independent validation, and decision threshold. Evidence needs to be read at exactly that level.
How accurate is a crime-linkage prediction?
Sensitivity, specificity, and the two kinds of error
A true positive is a genuinely linked pair classified as linked; a false positive is a pair attributed to different offenders that the method flags as linked. A true negative is an unrelated pair correctly separated; a false negative is a real link that goes unrecognized. Sensitivity concerns the share of true links detected. Specificity concerns the share of unrelated pairs rejected. A method tuned to detect more genuine links may generate more false alerts, depending on its threshold.
False positives can encourage tunnel vision, connect an innocent person to unrelated offenses, waste resources, or contaminate investigative narratives. False negatives can allow serial offenses to remain fragmented across investigators and jurisdictions. The consequences are asymmetrical in different settings: a database triage flag and an expert claim offered in court demand different decision standards.
ROC-AUC is useful, but it is not the probability that a proposed link is correct
Receiver operating characteristic area under the curve (ROC-AUC) quantifies how often a model ranks a randomly selected linked pair above a randomly selected unlinked pair. It measures discrimination across thresholds. It does not directly provide the positive predictive value of a flagged pair or the accuracy of a particular final decision. Ewanation and colleagues’ methodological paper analyzes why ROC curves can obscure practically important performance in crime-linkage settings.
Base rates, precision, and false discoveries
Most pairs in a large crime database may be unrelated. This creates a severe base-rate problem. Imagine that only 10 of 10,000 possible pairs are truly linked. Even a system detecting eight of the ten true links while wrongly flagging only one percent of unrelated pairs would generate roughly 100 false alerts alongside eight correct ones. Its flagged results would mostly be false links. The numbers are illustrative, not a published estimate of field performance.
Precision, or positive predictive value, asks what fraction of flagged links are actually linked in the tested data. Precision–recall curves can reveal trade-offs more directly when positive cases are rare. Analysts need to report precision at operationally realistic prevalence, along with the volume of cases investigators must review. A high ranking score may be useful for prioritization even when individual alerts have limited credibility, but the report must say which use has been validated.
Calibration and meaningful probabilities
Calibration asks whether events assigned a given predicted probability occur at approximately that frequency in comparable cases. A model with good AUC can be badly calibrated; a score of 0.80 may be only a ranking number rather than an 80% probability of a true link. Calibration depends on sampling, prevalence, and the data environment. Recalibration, external validation, and careful explanation of the model output are essential before presenting numbers as actual linkage probabilities.
An analyst should identify the reference population and sampling scheme for any numerical likelihood statement. Comparisons among crimes already known to belong to solved series may not approximate the probabilities encountered across all new unsolved reports.
Statistical approaches: similarities, classification, and series membership
Similarity coefficients
One approach encodes behavioral features as present, absent, or unknown and calculates similarity between cases. The Jaccard coefficient, for example, weights shared presences among binary variables rather than counting shared absences in the same way. This can be useful where behavior categories are sparse. But a similarity score needs empirically justified feature definitions, missing-data rules, and a method for deciding whether the score is informative.
Regression, classification, Bayesian methods, and clustering
Predictive models may combine features through logistic regression, classification trees, Bayesian analysis, or more complex methods. Clustering groups incidents into candidate series. A pairwise model assesses two incidents; a series-membership model asks whether a new incident belongs with an existing group. These tasks have different error structures, especially when pairwise mistakes spread into larger clusters.
A National Institute of Justice-sponsored statistical approach demonstrated possibilities for Bayes-factor-based linkage and clustering using burglary data. Reported performance within its study cannot be treated as a general field-wide success rate. Bayes factors can express relative support under model assumptions, whereas a posterior probability additionally requires relevant prior odds. Neither expression is equivalent to proof of criminal guilt.
AI and partially automated systems
Modern models can rank thousands of comparisons and reveal patterns that manual search would struggle to locate. Research summarized in the 2024 methods review and 2026 meta-analysis documents expanding analytical approaches. The proper operational claim is that validated systems may support triage and human review under defined conditions. Their adoption requires attention to data provenance, subgroup performance, explainability, security, audit trails, model drift, and error costs.
Predictive systems also inherit the reporting and enforcement processes that create their data. Uneven policing, differing access to forensic processing, and differential case closure can affect which crimes enter training sets and which labels are considered ground truth. A model may learn the administrative structure of the dataset as readily as offender behavior. For a broader treatment of construct validity and automated judgment, see the English Hub’s Psychological Assessment in the Age of AI.
Why linkage models fail or generalize poorly
Selection bias and solved-case bias
Studies based mainly on convictions or cleared incidents may overrepresent easy-to-detect patterns. Cases remain unsolved for many reasons, including missing information, interrupted investigations, and limits on laboratory resources. Testing only on cases with unusually complete evidence can make a method look stronger than it will be in a database of new, incomplete reports. The 2019 multinational study directly explored consequences of this issue.
Uncertain ground truth and circular validation
Validation requires independently established knowledge of which offenses share an offender. If cases are labeled linked because their behavior already seemed alike, and a model is then rewarded for recovering those labels, its evaluation can become circular. DNA-supported series, well-documented admissions corroborated independently, and other established links can improve ground-truth quality, although every source has limits and missingness. Labels should be graded by the strength of independent support.
Information leakage and dependent pairs
Pairs from the same offender may recur across both model training and testing; shared cases can allow the algorithm to recognize incident-specific information rather than generalizable patterns. Repeated observations are not independent. Validation should separate appropriate offenders, series, locations, or time periods where the operational task requires generalization beyond observed cases. The decision to split data by pairs, offenders, series, or geography can dramatically alter apparent performance.
Crime mix, place, culture, and jurisdiction
A model developed in one policing environment may encounter different housing forms, transport patterns, reporting practices, victim populations, or crime definitions elsewhere. Linkage research is historically concentrated in selected offense categories and countries. External validation is needed for every materially different target population. Analysts should distinguish differential measurement quality from genuine behavioral differences between groups.
Observer effects, coding disagreement, and cognitive bias
Investigators who know the suspected series or a favored suspect may unconsciously code ambiguous acts as more similar than a blinded analyst would. Shared narratives among agencies, media coverage, and hindsight can compound this problem. Clear coding manuals, independent recoding, structured consideration of nonlinks, and documentation of evidence chronology are important quality safeguards.
When similarity claims are challenged, a transparent record of how behaviors were defined and compared permits independent review. A linkage conclusion that relies on inaccessible expert intuition or undocumented revisions cannot receive the same scientific weight as a replicable, externally tested analysis.
Operational systems: the FBI’s ViCAP and the UK’s SCAS
In the United States, the FBI describes the Violent Criminal Apprehension Program (ViCAP) as a centralized repository and analytical service for violent incidents that may be serial. Participating agencies can compare homicide, sexual assault, missing-person, and related cases and use analytical support for timelines, mapping, crime-series matrices, and leads. The existence of an operational program establishes that linkage is part of investigative practice; it does not, by itself, validate every individual inference.
The United Kingdom’s National Crime Agency Serious Crime Analysis Section analyzes serious sexual offenses and certain homicides using structured case data in ViCLAS. Its remit, reporting thresholds, and investigative workflow are UK-specific. It illustrates organized comparative case analysis rather than a universal procedure mandated by law.
Professional practice differs internationally and across US jurisdictions. Agency databases, retention rules, privacy arrangements, professional qualifications, evidentiary disclosure obligations, and courtroom admissibility standards can all vary. A research finding from a national policing program should be described as evidence about that program and its tested context.
Can crime linkage be used as evidence in court?
Investigative utility and legal admissibility are distinct questions. A crime-linkage report may help police decide what to investigate. Whether a behavioral opinion may be presented as expert evidence, the foundation it needs, the method’s error information, and the permissible conclusions depend on the applicable court and jurisdiction. US federal evidence law, the laws of individual states, and procedures outside the United States are not identical.
A court-facing report requires a precisely defined proposition. “These crimes share several unusual features” is an observable comparison, assuming the source descriptions are reliable. “The same person probably committed these crimes” is a model-dependent inference. “The defendant committed them” is an identification and legal proposition requiring further evidence. Treating the first as if it proves the third skips the essential evidentiary steps.
A sound expert should disclose feature selection, case inclusion, competing hypotheses, the extent of independent testing, known error rates, population limits, and the effect of contextual information. The APA Specialty Guidelines for Forensic Psychology address professional competence, reliance on appropriate methods, clarity of roles, and communication of limitations in psycholegal work. These professional guidelines are not statutes, court holdings, or a substitute for local procedural law.
Similarity between multiple allegations can influence jurors or investigators even when each link remains uncertain. The strongest safeguard is an evidence hierarchy that separates behavioral screening, corroborated forensic associations, formal expert opinions, and decisions made under applicable legal standards. A risk score, behavior score, or cluster label is never itself a finding of criminal responsibility.
A worked example: three burglaries that look related
Consider a fictional city where three burglaries occur within a small area over two weeks. Each involves a rear entry point, unattended properties, and removal of small electronics. An initial analyst flags the three incidents. Those shared facts warrant checking but may be common in that location and period; they cannot yet separate a single series from a local trend involving unrelated offenders.
A closer review discovers that two reports independently document the same unusual sequence of actions and a matching direction of departure. The third report lacks those details because the witness never observed the exit. That missing observation is unknown, not evidence that the pattern did not occur. The analyst checks other nearby burglaries with similar entry opportunities and counts how common the combination is.
Suppose the first two incidents then share an independently recorded vehicle identifier while the third does not. The identifier may substantially strengthen the lead for the first pair, depending on how it was obtained and verified. It should not retroactively be credited to the third incident merely because all three were grouped earlier. Investigators preserve the alternative that some offenses are linked and others are not.
This example demonstrates the core workflow: search, compare, test distinctiveness against a relevant background, investigate dissimilarities and missingness, obtain independent corroboration, and keep the inference narrower than the evidence. It is an illustration of reasoning, not a validated field protocol.
Ethics, fairness, privacy, and harm prevention
False links can expose people and communities to unjustified suspicion; missed links can delay protection and accountability. Agencies therefore need proportionate access controls, restricted retention, accurate correction procedures, auditable decision logs, and rules for separating leads from confirmed findings. These safeguards are especially important when multiple departments share behavioral labels.
Fairness requires examining whether data completeness, response times, police activity, case-selection practices, or feature coding differ across populations and neighborhoods. Subgroup auditing should measure performance where valid ground truth is available, rather than inferring individual behavior from demographic categories. Geographic features can reflect unequal policing and infrastructure as well as repeated offender movement.
Public communication is also an evidence problem. Announcing an apparent serial offender prematurely can amplify fear, encourage imitators, pressure witnesses, and bias later reports. Investigative necessity, the rights of affected people, and public safety require case-specific assessment rather than sensational framing.
Researchers and analysts working with disturbing offense narratives also need supervision, workload controls, and well-being safeguards. The 2026 meta-analysis identifies the cost and emotional burden of linkage work among motivations for research into decision-support tools.
How to judge whether a linkage conclusion is trustworthy
For journalists, lawyers, investigators, researchers, and interested readers, the essential question is not whether the stories sound alike. Ask which observations were independently documented, how many relevant unlinked cases were compared, whether other offenders commonly exhibit the same pattern, how missing data were handled, whether the analyst was blinded to the favored suspect, and whether the method was validated on realistic cases of the same type.
The report should supply enough evidence to distinguish its intended use: exploratory lead generation, ranking cases for review, estimating support for common authorship, or offering an opinion to a legal decision-maker. The threshold and error measures appropriate to one purpose may not satisfy another. Explainable limits strengthen an investigative tool because they allow its information to be used at the level it actually supports.
Readers exploring the broader professional roles and boundaries in this area can continue with the live English Hub article Forensic Psychology: What It Is and What Forensic Psychologists Do.
Common myths about crime linkage
Myth: Every serial offender has a unique behavioral signature
Research supports partial behavioral consistency and distinctiveness for some crimes and samples. Individual features can be inconsistent or widely shared. Serial homicide research cautions against assuming that one repeated act functions as a unique identifier.
Myth: Similar crimes must share an offender
Common targets, social settings, opportunities, local practices, and copycat actions can produce similar cases. A meaningful link requires comparison against alternatives and, whenever available, independent evidence.
Myth: A high AUC means most flagged links are correct
AUC is a discrimination statistic. Precision and practical false-alert burdens depend strongly on the prevalence of true links. Methods research on ROC interpretation explains this distinction.
Myth: Linkage is the same as profiling or diagnosis
Linkage primarily compares incidents. It does not infer an official diagnosis, establish a person’s motives, or determine legal culpability. Its results are investigative inferences with varying levels of support.
Myth: An AI-generated cluster constitutes proof
An algorithm can suggest patterns for a human to investigate. Its output must be validated, calibrated where probability claims are made, checked for source-data bias, and interpreted alongside independent evidence.
Frequently asked questions
Can police link crimes without DNA?
Yes. Analysts can compare observable behavior, geography, timing, and other recorded features when DNA is absent. Such links typically generate or prioritize investigative leads. Their reliability varies with the behaviors studied, the method used, and the population against which it was validated.
What is the difference between crime linkage and linkage analysis?
Crime linkage is the broad research and practice area. Linkage analysis is one of several terms used for the evaluation of proposed common-offender cases. Some agencies distinguish database-based comparative case analysis from expert assessment of a candidate series. Davies and Woodhams (2019) document these terminological variations.
What is comparative case analysis?
Comparative case analysis systematically searches and compares incident records for potentially related offenses. It can start from one index offense or a wider dataset. The UK SCAS uses a form of this approach, while practices in other countries and services may differ.
Which is more useful: the offender’s MO or the crime locations?
It depends on crime type. The 2026 meta-analysis found promising evidence for geographical domains in certain acquisitive offenses and combined MO features in sexual offenses. Neither domain is universally best across all crimes.
Are serial killers’ signatures scientifically reliable?
A recurring behavior may be relevant to comparison, but the strong popular idea of an immutable, individually unique signature is inadequately supported as a general rule. Analysts should evaluate documented behaviors as data rather than infer a fixed identity from cinematic symbolism.
Can a linkage score identify a suspect?
A score can rank candidate case pairs or series and sometimes help prioritize suspect-related inquiries when combined with other data. It does not independently identify a perpetrator or prove that a particular person committed a crime.
How do researchers know which crimes are linked?
Research requires a defensible reference classification. This can involve cases corroborated by DNA or other independent evidence and carefully documented investigations. Each labeling source has limits; studies relying only on easy-to-solve convictions may overestimate the performance of methods used on unknown cases.
What is the difference between a false link and a missed link?
A false link connects offenses attributed to different offenders; a missed link fails to connect offenses genuinely committed by the same offender. Both errors have investigative consequences. Their expected frequency depends on decision thresholds and the underlying prevalence of linked cases.
Can crime linkage establish criminal responsibility?
No. Criminal responsibility concerns legally defined culpability under jurisdiction-specific law. Behavioral linkage may contribute to investigative hypotheses or be considered as one form of evidence subject to legal rules; the court determines legally relevant findings. Psychological similarity and legal responsibility are separate questions.
Does the same method work across jurisdictions?
Only after appropriate testing. Data collection, crime definitions, population characteristics, missingness, and law differ among agencies and countries. Transfer requires external validation and, where applicable, recalibration.
Will AI make crime linkage infallible?
No predictive method is error-free. Automation may improve large-scale triage, but can also amplify reporting bias, unknown base-rate effects, spurious correlations, and overconfident interpretation. Real-world testing and accountable human review remain necessary.
Conclusion
Crime linkage is a scientifically investigable method of comparing offenses for possible common authorship. Its best-supported value lies in structured investigative search and prioritization where consistent, sufficiently distinctive patterns can be evaluated against relevant alternatives. Research, including major studies from 2025 and 2026, shows that performance varies by crime type, sampling, behavioral domain, and statistical method. Reliable practice reports operational error rates, considers the low prevalence of true links in large databases, verifies results independently, and preserves the boundary between an investigative lead and a legal conclusion.
Related articles
Read the published English Hub foundations: Forensic Psychology: What It Is and What Forensic Psychologists Do and Psychological Assessment in the Age of AI: Testing, Prediction, Bias, and Professional Judgment. Explore related EN-58 research on geographic profiling and the distinction between investigative interviewing and interrogation. Links to Investigative Psychology and Behavioral Analysis in Criminal Investigations will be added after those separate English articles are published.
Related Articles
References
American Psychological Association. (2013). Specialty Guidelines for Forensic Psychology. American Psychologist.
Bateman, A. L., & Salfati, C. G. (2007). An examination of behavioral consistency using individual behaviors or groups of behaviors in serial homicide. Behavioral Sciences & the Law, 25(4), 527–544.
Burrell, A., Costello, B., & Woodhams, J. (2024). Methods used to link crimes using behaviour: A literature review. Aggression and Violent Behavior, 79, 102014.
Davies, K., & Woodhams, J. (2019). The practice of crime linkage: A review of the literature. Journal of Investigative Psychology and Offender Profiling, 16(3), 169–200.
Ewanation, L., et al. (2023). Receiver operating characteristic curves in the crime linkage context: Benefits, limitations, and recommendations. Applied Cognitive Psychology.
Federal Bureau of Investigation. (n.d.). Behavioral Analysis. FBI.
Kelkar, G. M., Jones, C., & Woodhams, J. (2026). The accuracy of statistical models to link serial crimes: A meta-analysis. Journal of Police and Criminal Psychology.
National Crime Agency. (n.d.). Major crime investigative support. UK National Crime Agency.
National Crime Agency. (n.d.). Serious Crime Analysis Section. UK National Crime Agency.
National Institute of Justice. (2016). A statistical approach to crime linkage. U.S. Department of Justice.
Tonkin, M., Lemeire, J., Woodhams, J., Alrajeh, D., et al. (2025). Building the statistical evidence base for crime linkage decision-support tools with sexual offences. Journal of Quantitative Criminology (online 2025).
Woodhams, J., et al. (2007). The psychology of linking crimes: A review of the evidence. Legal and Criminological Psychology.
Woodhams, J., et al. (2019). Linking serial sexual offences: Moving towards an ecologically valid test of the principles of crime linkage. Legal and Criminological Psychology, 24(1), 123–140.