Correlations & Realistic Relationships

Create realistic personas by modeling real-world relationships between attributes like age-income correlation, education-salary relationships, and experience-skill dependencies.

Basic Correlations

Define linear relationships between numeric attributes:

basic-correlations.ts

1import { PersonaGroup, NormalDistribution } from '@jamesaphoenix/persona-sdk';
2
3const group = new PersonaGroup('Professionals');
4
5// Generate personas with age-income correlation
6group.generateWithCorrelations(100, {
7  attributes: {
8    age: new NormalDistribution(35, 8),
9    income: new NormalDistribution(75000, 20000),
10    yearsExperience: new NormalDistribution(10, 5)
11  },
12  correlations: [
13    // Older people tend to earn more
14    { attribute1: 'age', attribute2: 'income', correlation: 0.6 },
15    // Age correlates with experience
16    { attribute1: 'age', attribute2: 'yearsExperience', correlation: 0.8 },
17    // Experience correlates with income
18    { attribute1: 'yearsExperience', attribute2: 'income', correlation: 0.7 }
19  ]
20});
21
22// Results in realistic combinations:
23// 45-year-old, 18 years exp, $95k income
24// 28-year-old, 5 years exp, $65k income

Conditional Dependencies

Add logical constraints and transformations:

conditional-dependencies.ts

1// Add realistic constraints
2group.generateWithCorrelations(50, {
3  attributes: {
4    age: new NormalDistribution(32, 7),
5    yearsExperience: new NormalDistribution(8, 4),
6    education: new CategoricalDistribution([
7      { value: 'High School', probability: 0.2 },
8      { value: 'Bachelor', probability: 0.5 },
9      { value: 'Master', probability: 0.25 },
10      { value: 'PhD', probability: 0.05 }
11    ])
12  },
13  conditionals: [
14    {
15      // Experience can't exceed working years (age - 18)
16      attribute: 'yearsExperience',
17      dependsOn: 'age',
18      transform: (experience, age) => {
19        const maxExperience = Math.max(0, age - 18);
20        return Math.min(experience, maxExperience);
21      }
22    },
23    {
24      // Adjust income based on education level
25      attribute: 'income',
26      dependsOn: 'education',
27      transform: (baseIncome, education) => {
28        const multipliers = {
29          'High School': 0.8,
30          'Bachelor': 1.0,
31          'Master': 1.3,
32          'PhD': 1.6
33        };
34        return baseIncome * (multipliers[education] || 1.0);
35      }
36    }
37  ]
38});

Common Correlation Patterns

Use built-in correlation patterns for realistic relationships:

correlation-patterns.ts

1import { CommonCorrelations, PersonaCorrelationPresets } from '@jamesaphoenix/persona-sdk';
2
3// Use built-in correlation functions
4const correlatedIncome = CommonCorrelations.ageIncome(50000, age);
5const boundedExperience = CommonCorrelations.ageExperience(experience, age);
6const realisticWeight = CommonCorrelations.heightWeight(weight, height);
7const educationBonus = CommonCorrelations.educationIncome(income, educationYears);
8
9// Use preset correlation configurations
10group.generateWithCorrelations(100, {
11  ...PersonaCorrelationPresets.REALISTIC_ADULT,
12  // Adds common correlations:
13  // - Age ↔ Income (0.5)
14  // - Age ↔ Experience (0.8)
15  // - Height ↔ Weight (0.7)
16  // - Education ↔ Income (0.6)
17});
18
19// Professional preset for workplace personas
20group.generateWithCorrelations(50, {
21  ...PersonaCorrelationPresets.PROFESSIONAL,
22  // Adds workplace-specific correlations:
23  // - Experience ↔ Salary (0.8)
24  // - Education ↔ Position Level (0.7)
25  // - Age ↔ Management Role (0.6)
26});

Correlation Types

Linear Correlation

Direct proportional relationship (default):

{
  attribute1: 'age',
  attribute2: 'income',
  correlation: 0.6,
  type: 'linear' // As age increases, income increases proportionally
}

Exponential Correlation

Accelerating relationship for compound effects:

{
  attribute1: 'yearsExperience',
  attribute2: 'skillLevel',
  correlation: 0.7,
  type: 'exponential' // Skill grows faster with more experience
}

Logarithmic Correlation

Diminishing returns relationship:

{
  attribute1: 'hoursStudied',
  attribute2: 'testScore',
  correlation: 0.5,
  type: 'logarithmic' // Returns diminish with more study time
}

💡 Best Practices

• Keep correlations realistic: Use values between -0.8 and 0.8 for most relationships
• Add logical constraints: Use conditionals to prevent impossible combinations
• Layer correlations: Build complex relationships by combining multiple simple ones
• Test your model: Generate samples and verify the relationships make sense
• Use presets: Start with built-in patterns and customize as needed

Distributions

AI Features