• Descriptive
    • Moments
    • Concentration
    • Central Tendency
    • Variability
    • Stem-and-Leaf Plot
    • Histogram & Frequency Table
    • Data Quality Forensics
    • Conditional EDA
    • Quantiles
    • Kernel Density Estimation
    • Normal QQ Plot
    • Bootstrap Plot

    • Multivariate Descriptive Statistics
  • Distributions
    • Binomial Probabilities
    • Geometric Probabilities
    • Negative Binomial Probabilities
    • Hypergeometric Probabilities
    • Multinomial Probabilities
    • Dirichlet
    • Poisson Probabilities

    • Exponential
    • Gamma
    • Erlang
    • Weibull
    • Rayleigh
    • Maxwell-Boltzmann
    • Lognormal
    • Pareto
    • Inverse Gamma
    • Inverse Chi-Square

    • Beta
    • Power
    • Beta Prime (Inv. Beta)
    • Triangular

    • Normal (area)
    • Logistic
    • Laplace
    • Cauchy (standard)
    • Cauchy (location-scale)
    • Gumbel
    • Fréchet
    • Generalized Extreme Value

    • Normal RNG
    • ML Fitting
    • Tukey Lambda PPCC
    • Box-Cox Normality Plot
    • Noncentral t
    • Noncentral F
    • Sample Correlation r

    • Empirical Tests
  • Hypotheses
    • Theoretical Aspects of Hypothesis Testing
    • Bayesian Inference
    • Minimum Sample Size

    • Empirical Tests
    • Multivariate (pair-wise) Testing
  • Models
    • Manual Model Building
    • Guided Model Building
  • Time Series
    • Time Series Plot
    • Decomposition
    • Exponential Smoothing

    • Blocked Bootstrap Plot
    • Mean Plot
    • (P)ACF
    • VRM
    • Standard Deviation-Mean Plot
    • Spectral Analysis
    • ARIMA

    • Cross Correlation Function
    • Granger Causality
  1. Introduction to Probability
  2. 8  Sensitivity and Specificity
  • Preface
  • Getting Started
    • 1  Introduction
    • 2  Why Do We Need Innovative Technology?
    • 3  Basic Definitions
    • 4  The Big Picture: Why We Analyze Data
  • Introduction to Probability
    • 5  Definitions of Probability
    • 6  Jeffreys’ axiom system
    • 7  Bayes’ Theorem
    • 8  Sensitivity and Specificity
    • 9  Naive Bayes Classifier
    • 10  Law of Large Numbers

    • 11  Problems
  • Probability Distributions
    • 12  Bernoulli Distribution
    • 13  Binomial Distribution
    • 14  Geometric Distribution
    • 15  Negative Binomial Distribution
    • 16  Hypergeometric Distribution
    • 17  Multinomial Distribution
    • 18  Poisson Distribution

    • 19  Uniform Distribution (Rectangular Distribution)
    • 20  Normal Distribution (Gaussian Distribution)
    • 21  Gaussian Naive Bayes Classifier
    • 22  Chi Distribution
    • 23  Chi-squared Distribution (1 parameter)
    • 24  Chi-squared Distribution (2 parameters)
    • 25  Student t-Distribution
    • 26  Fisher F-Distribution
    • 27  Exponential Distribution
    • 28  Lognormal Distribution
    • 29  Gamma Distribution
    • 30  Beta Distribution
    • 31  Weibull Distribution
    • 32  Pareto Distribution
    • 33  Inverse Gamma Distribution
    • 34  Rayleigh Distribution
    • 35  Erlang Distribution
    • 36  Logistic Distribution
    • 37  Laplace Distribution
    • 38  Gumbel Distribution
    • 39  Cauchy Distribution
    • 40  Triangular Distribution
    • 41  Power Distribution
    • 42  Beta Prime Distribution
    • 43  Sample Correlation Distribution
    • 44  Dirichlet Distribution
    • 45  Generalized Extreme Value (GEV) Distribution
    • 46  Frechet Distribution
    • 47  Noncentral t Distribution
    • 48  Noncentral F Distribution
    • 49  Inverse Chi-Squared Distribution
    • 50  Maxwell-Boltzmann Distribution
    • 51  Distribution Relationship Map

    • 52  Problems
  • Descriptive Statistics & Exploratory Data Analysis
    • 53  Types of Data
    • 54  Datasheets

    • 55  Frequency Plot (Bar Plot)
    • 56  Frequency Table
    • 57  Contingency Table
    • 58  Binomial Classification Metrics
    • 59  Confusion Matrix
    • 60  ROC Analysis

    • 61  Stem-and-Leaf Plot
    • 62  Histogram
    • 63  Data Quality Forensics
    • 64  Quantiles
    • 65  Central Tendency
    • 66  Variability
    • 67  Skewness & Kurtosis
    • 68  Concentration
    • 69  Notched Boxplot
    • 70  Scatterplot
    • 71  Pearson Correlation
    • 72  Rank Correlation
    • 73  Partial Pearson Correlation
    • 74  Simple Linear Regression
    • 75  Moments
    • 76  Quantile-Quantile Plot (QQ Plot)
    • 77  Normal Probability Plot
    • 78  Probability Plot Correlation Coefficient Plot (PPCC Plot)
    • 79  Box-Cox Normality Plot
    • 80  Kernel Density Estimation
    • 81  Bivariate Kernel Density Plot
    • 82  Conditional EDA: Panel Diagnostics
    • 83  Bootstrap Plot (Central Tendency)
    • 84  Survey Scores Rank Order Comparison
    • 85  Cronbach Alpha

    • 86  Equi-distant Time Series
    • 87  Time Series Plot (Run Sequence Plot)
    • 88  Mean Plot
    • 89  Blocked Bootstrap Plot (Central Tendency)
    • 90  Standard Deviation-Mean Plot
    • 91  Variance Reduction Matrix
    • 92  (Partial) Autocorrelation Function
    • 93  Periodogram & Cumulative Periodogram

    • 94  Problems
  • Hypothesis Testing
    • 95  Normal Distributions revisited
    • 96  The Population
    • 97  The Sample
    • 98  The One-Sided Hypothesis Test
    • 99  The Two-Sided Hypothesis Test
    • 100  When to use a one-sided or two-sided test?
    • 101  What if \(\sigma\) is unknown?
    • 102  The Central Limit Theorem (revisited)
    • 103  Statistical Test of the Population Mean with known Variance
    • 104  Statistical Test of the Population Mean with unknown Variance
    • 105  Statistical Test of the Variance
    • 106  Statistical Test of the Population Proportion
    • 107  Statistical Test of the Standard Deviation \(\sigma\)
    • 108  Statistical Test of the difference between Means -- Independent/Unpaired Samples
    • 109  Statistical Test of the difference between Means -- Dependent/Paired Samples
    • 110  Statistical Test of the difference between Variances -- Independent/Unpaired Samples

    • 111  Hypothesis Testing for Research Purposes
    • 112  Decision Thresholds, Alpha, and Confidence Levels
    • 113  Bayesian Inference for Decision-Making
    • 114  One Sample t-Test
    • 115  Skewness & Kurtosis Tests
    • 116  Paired Two Sample t-Test
    • 117  Wilcoxon Signed-Rank Test
    • 118  Unpaired Two Sample t-Test
    • 119  Unpaired Two Sample Welch Test
    • 120  Two One-Sided Tests (TOST) for Equivalence
    • 121  Mann-Whitney U test (Wilcoxon Rank-Sum Test)
    • 122  Bayesian Two Sample Test
    • 123  Median Test based on Notched Boxplots
    • 124  Chi-Squared Tests for Count Data
    • 125  Kolmogorov-Smirnov Test
    • 126  One Way Analysis of Variance (1-way ANOVA)
    • 127  Kruskal-Wallis Test
    • 128  Two Way Analysis of Variance (2-way ANOVA)
    • 129  Repeated Measures ANOVA
    • 130  Friedman Test
    • 131  Testing Correlations
    • 132  A Note on Causality

    • 133  Problems
  • Regression Models
    • 134  Simple Linear Regression Model (SLRM)
    • 135  Multiple Linear Regression Model (MLRM)
    • 136  Logistic Regression
    • 137  Generalized Linear Models
    • 138  Multinomial and Ordinal Logistic Regression
    • 139  Cox Proportional Hazards Regression
    • 140  Conditional Inference Trees
    • 141  Leaf Diagnostics for Conditional Inference Trees
    • 142  Conditional Random Forests
    • 143  Hypothesis Testing with Linear Regression Models (from a Practical Point of View)

    • 144  Problems
  • Introduction to Time Series Analysis
    • 145  Case: the Market of Health and Personal Care Products
    • 146  Decomposition of Time Series
    • 147  Ad hoc Forecasting of Time Series
  • Box-Jenkins Analysis
    • 148  Introduction to Box-Jenkins Analysis
    • 149  Theoretical Concepts
    • 150  Stationarity
    • 151  Identifying ARMA parameters
    • 152  Estimating ARMA Parameters and Residual Diagnostics
    • 153  Forecasting with ARIMA models
    • 154  Intervention Analysis
    • 155  Cross-Correlation Function
    • 156  Transfer Function Noise Models
    • 157  General-to-Specific Modeling
  • Model Building Strategies
    • 158  Introduction to Model Building Strategies
    • 159  Manual Model Building
    • 160  Model Validation
    • 161  Regularization Methods
    • 162  Hyperparameter Optimization Strategies
    • 163  Guided Model Building in Practice
    • 164  Diagnostics, Revision, and Guided Forecasting
    • 165  Leakage, Target Encoding, and Robust Regression
  • References
  • Appendices
    • Appendices
    • A  Method Selection Guide
    • B  Presentations and Teaching Materials
    • C  R Language Concepts for Statistical Computing
    • D  Matrix Algebra
    • E  Standard Normal Table (Gaussian Table)
    • F  Critical values of Student’s \(t\) distribution with \(\nu\) degrees of freedom
    • G  Upper-tail critical values of the \(\chi^2\)-distribution with \(\nu\) degrees of freedom
    • H  Lower-tail critical values of the \(\chi^2\)-distribution with \(\nu\) degrees of freedom
  1. Introduction to Probability
  2. 8  Sensitivity and Specificity

8  Sensitivity and Specificity

Bayes’ Theorem is closely related to the definitions of Sensitivity and Specificity, which can be illustrated for the case of a binary classification problem. Suppose that we are interested in real-time fraud detection for online credit card payments (if \(H_1\) is true, then the financial transaction is fraudulent and \(H_2\) is false). In other words, we wish to design a model which helps us to decide whether or not a financial transaction is allowed to proceed during the process of completing an e-commerce sale.

The default prediction we make (i.e. when there is insufficient evidence to suggest otherwise) is that a transaction is legitimate (\(H_2\) is true). We do this because we know from historical analysis that most transactions are legitimate and only about 0.2% of transactions (i.e. the so-called “prevalence”) involve stolen credit cards or some form of identity fraud. Another reason why we do not want to decide that \(H_1\) is true by default is the legal presumption of innocence. Since we wish to design a real-time fraud detection system (i.e. one that refuses a credit card transaction to proceed before any money is transferred), false negatives are often considered less disruptive to immediate customer experience than false positives. However, undetected fraud still has real costs and may only be recoverable after investigation, if detected later.

Table 8.1: Sensitivity and Specificity
\(H_2\) is true \(H_1\) is true
Accept \(H_2\) True Negative (TN) False Negative (FN)
(type II error)
Reject \(H_2\) False Positive (FP)
(type I error)
True Positive (TP)
True Negative Rate =
TNR = TN / (TN + FP) =
Specificity
True Positive Rate =
TPR = TP / (FN + TP) =
Sensitivity (Recall)

Table 8.1 shows that the Sensitivity or True Positive Rate (TPR) reflects the proportion of correctly identified fraud cases. Fraud detection models with high Sensitivity are good at detecting fraud cases. On the other hand, the table also shows the Specificity or True Negative Rate, which represents the proportion of correctly identified transactions that are legitimate. In other words, fraud detection models with high Specificity are good at detecting legitimate transactions.

A perfect prediction model would have 100% Specificity and Sensitivity. Most of the time, however, these tests will have a non-zero error. In fact, there is a theoretical lower bound on classification error called the “Bayes error rate,” determined by the overlap of class distributions, and it is generally unknown in practice (see Chapter 113).

Furthermore, there is often a trade-off between Sensitivity and Specificity. For instance, a change that improves the Sensitivity of the model will often result in lower Specificity and vice versa. In contrast, improving both (Sensitivity and Specificity) at the same time could be achieved when more or better quality of data becomes available.

Suppose that our fraud detection system has 99% Sensitivity and Specificity and that the prevalence of fraud (based on historical data) is 0.2%. What is the probability that a random transaction which is classified as a “positive” actually involves fraud?

Suppose that the computed fraud probability is used to support a practical decision (e.g. block the transaction immediately or send it for manual review). How large should the probability of fraud be before we decide to act?

The simple formulation of Bayes’ Theorem Equation 7.3 states that

\[ \begin{equation} \text{P}(H_1 | D) = \frac{\text{P}(D | H_1) \text{P}(H_1)}{\text{P}(D)} \end{equation} \]

which becomes1

\[ \begin{equation} \text{P}(H_1 | D+) = \frac{\text{P}(D+ | H_1) \text{P}(H_1)}{\text{P}(D+ | H_1) \text{P}(H_1) + \text{P}(D+ | H_2) \text{P}(H_2) } \end{equation} \]

or

\[ \begin{equation} \text{P}(H_1 | D+) = \frac{0.99 \times 0.002}{0.99 \times 0.002 + (1 - 0.99) (1 - 0.002) } \simeq 16.6\% \end{equation} \]

The same result can be obtained through the odds formula Equation 7.4

\[ \begin{equation} \frac{\text{P}(H_1 | D+)}{\text{P}(H_2 | D+)} = \frac{\text{P}(D+ | H_1)}{\text{P}(D+ | H_2)} \frac{\text{P}(H_1)}{\text{P}(H_2)} = \frac{0.99}{(1 - 0.99)} \frac{0.002}{(1 - 0.002)} = \frac{0.00198}{0.00998} \end{equation} \]

which leads to a probability of \(0.00198 / (0.00198 + 0.00998) \simeq 16.6%\).

Whether 16.6% is high enough to justify action depends on a decision threshold and the relative costs of false positives (blocking legitimate transactions) and false negatives (allowing fraud to proceed).

Interactive Shiny app (click to load).
Open in new tab
NoteEarly Threshold Insight (used again later)

The fraud example above introduces a core idea that will return throughout the handbook:

there is no universally correct decision threshold.
The threshold must be chosen by the purpose of the decision.

Once we know the posterior probability of fraud (here about 16.6%), we still need a rule for action: should the system block the transaction, allow it, or send it for manual review? A stricter threshold may reduce false positives (fewer legitimate transactions blocked) but increase false negatives (more fraud missed). A more permissive threshold may do the opposite. The appropriate choice depends on the relative costs of these errors and the operational context.

This is the same logic used later in hypothesis testing:

  • in classification, we choose a classification threshold;
  • in hypothesis testing, we choose a significance threshold \(\alpha\) (or equivalently a confidence level).

In both cases, changing the threshold changes the balance between error types. This is why the handbook later distinguishes between thresholds for confirmatory tests and thresholds used for diagnostic screening or assumption checks.

For the classification-threshold version of this idea, see Chapter 60.
For the general framework (including \(\alpha\) and confidence levels), see Chapter 112.


  1. The symbol D is replaced by D+ because we predict that the transaction is fraudulent.↩︎

7  Bayes’ Theorem
9  Naive Bayes Classifier

© 2026 Patrick Wessa. Provided as-is, without warranty.

Feedback: e-mail | Anonymous contributions: click to copy (Sats) | click to copy (XMR)

Cookie Preferences