The Five Factor Model of personality and evaluation of drug consumption risk

E. Fehrman, A. K. Muhammad, E. M. Mirkes, V. Egan, Alexander N. GorbanarXiv 2015 · DOI · arXiv

TL;DR

Researchers successfully used machine learning and the "Big Five" personality traits to predict an individual's risk of using 18 different psychoactive substances. By identifying "correlation pleiades"—clusters of drugs with similar consumption patterns—the study achieved over 70% accuracy in classifying users versus non-users across nearly all categories.

What Problem Does This Paper Address?

While we know that certain demographics are more prone to drug use, predicting an individual’s specific risk remains a complex challenge for healthcare providers and policy planners. Most existing research looks at broad statistical trends rather than individual psychological profiles.

The authors tackle the "value of definitions" problem—how do we define a "drug user" when consumption ranges from a one-time experiment a decade ago to daily use? By bridging the gap between psychology (personality theory) and data science (supervised learning), the paper seeks to determine if our inherent psychological makeup—how neurotic, open, or impulsive we are—can reliably signal our risk for substance misuse.

Methods

The study utilized an online survey to collect data on the Five Factor Model (FFM) of personality: Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. They also measured impulsivity (BIS-11) and sensation seeking (ImpSS).

What sets this paper apart is its exhaustive approach to model selection. Rather than picking one algorithm, the researchers ran a "tournament" of classifiers for each drug, including:

  • Decision Trees (DT) and Random Forests (RF)
  • k-Nearest Neighbors (kNN)
  • Linear Discriminant Analysis (LDA)
  • Logistic Regression (LR) and Naïve Bayes (NB)
  • Probability Density Function Estimation (PDFE)

They also employed sophisticated feature engineering, such as CatPCA for quantifying categorical data and Sparse PCA for feature ranking, to identify which specific personality traits were the strongest predictors for which drugs.

Datasets

The researchers compiled a dataset of 1,885 respondents (refined from an initial 2,051).

  • Input Features: 12 attributes including age, gender, education, and the personality scores mentioned above.
  • Target Variables: Consumption records for 18 drugs (e.g., alcohol, cocaine, ecstasy, heroin, "legal highs").
  • The "Semeron" Control: To ensure data integrity, the authors included a fictitious drug called "Semeron." Participants who claimed to use it were flagged as "over-claimers" and excluded from the study.

Key Insights and Ideas

  • The Personality Signature: A universal trend emerged: high Neuroticism and Openness, combined with low Agreeableness and Conscientiousness, are strong indicators of drug consumption risk. However, the influence of Extraversion is "drug-specific"—for example, it's high for ecstasy users but low for heroin users.
  • Correlation Pleiades: Drugs are not used in isolation. The authors identified three "pleiades" (clusters):
  • Ecstasy Pleiad: Includes stimulants and hallucinogens like amphetamines, cannabis, cocaine, LSD, and magic mushrooms.
  • Heroin Pleiad: Centered on crack, cocaine, and methadone.
  • Benzodiazepines Pleiad: Often linked with methadone and cocaine.
  • Openness vs. Conscientiousness: Openness to Experience is a major predictor for "party drugs" and hallucinogens, likely driven by curiosity, while low Conscientiousness is a hallmark of more high-risk, addictive consumption patterns.

Key Results

The classification quality was remarkably high. Using Leave-One-Out Cross-Validation (LOOCV), the models achieved:

  • Sensitivity and Specificity > 70% for almost all 18 drugs.
  • Top Performers: Accuracy exceeded 75% for cannabis, crack, ecstasy, LSD, VSA (volatile substance abuse), and "legal highs."
  • Best Classifiers: Decision Trees were the most effective for the majority of drugs, though LDA performed best for alcohol, and kNN was superior for caffeine and chocolate.
  • Risk Mapping: The authors developed "risk maps" that visualize the probability of drug use based on age and sensation seeking, providing a potential tool for clinical intervention.

Limitations and Future Work

The authors acknowledge a significant sampling bias: the data was collected via an online "snowball" methodology, resulting in a participant pool that was more highly educated and had a higher rate of drug use than the general population.

Furthermore, the study is cross-sectional, meaning it captures a snapshot in time. It cannot determine if personality traits cause drug use or if long-term drug use alters personality. Future work could involve longitudinal studies to track how these risks evolve over an individual's lifespan and expanding the dataset to include more diverse ethnic and geographic backgrounds.

Summary updated

Abstract

The problem of evaluating an individual's risk of drug consumption and misuse is highly important. An online survey methodology was employed to collect data including Big Five personality traits (NEO-FFI-R), impulsivity (BIS-11), sensation seeking (ImpSS), and demographic information. The data set contained information on the consumption of 18 central nervous system psychoactive drugs. Correlation analysis demonstrated the existence of groups of drugs with strongly correlated consumption patterns. Three correlation pleiades were identified, named by the central drug in the pleiade: ecstasy, heroin, and benzodiazepines pleiades. An exhaustive search was performed to select the most effective subset of input features and data mining methods to classify users and non-users for each drug and pleiad. A number of classification methods were employed (decision tree, random forest, $k$-nearest neighbors, linear discriminant analysis, Gaussian mixture, probability density function estimation, logistic regression and na{ï}ve Bayes) and the most effective classifier was selected for each drug. The quality of classification was surprisingly high with sensitivity and specificity (evaluated by leave-one-out cross-validation) being greater than 70\% for almost all classification tasks. The best results with sensitivity and specificity being greater than 75\% were achieved for cannabis, crack, ecstasy, legal highs, LSD, and volatile substance abuse (VSA).
BibTeX
@misc{fehrman2015five,
  author = {E. Fehrman and A. K. Muhammad and E. M. Mirkes and V. Egan and Alexander N. Gorban},
  title = {{The Five Factor Model of personality and evaluation of drug consumption risk}},
  year = {2015},
  eprint = {1506.06297},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1506.06297},
}
Figure