# Triangulation Guide for CRISP-T

## What is Triangulation?

Triangulation involves validating findings by comparing and contrasting results from different analytical methods or data sources. In CRISP-T, this means linking textual insights with numeric patterns.

## Key Strategies

### Cross-Modal Linking
- Use `temporal_link_by_time` to align text documents with time-stamped numeric data
- Use `embedding_link` to connect documents with similar semantic content to numeric records
- If id columns exist, use them to directly link text and numeric data
- If keywords exist in both text and numeric data (as columns), use them for linking

### Link Topic Keywords to Variables

After topic modeling:
- Topics generate keywords representing themes
- Use `add_relationship` to link keywords to relevant dataframe columns
- Example: If topic discusses "satisfaction", link to satisfaction score column

### Compare Patterns

- Cross-reference sentiment with numeric outcomes
- Compare topic distributions across demographic groups
- Validate clustering results using both text and numbers

### Use Relationships

- `add_relationship("text:keyword", "num:column", "correlates")`
- `get_relationships_for_keyword` to explore connections
- Document theoretical justifications for relationships

### Validate Findings

- Check if text-based themes align with numeric clusters
- Test if sentiment patterns predict outcomes
- Use regression to quantify relationships
- Decision trees reveal which factors matter most

## Example Workflow

1. Topic model reveals "healthcare access" theme
2. Assign documents to topics (creates keyword labels)
3. Link "healthcare access" keyword to "insurance_status" column
4. Run regression with insurance_status as outcome
5. Compare topic prevalence across insurance groups
6. Add relationships to document connections
7. Validate using classification models

## Best Practices

- Document all relationships you create
- Test relationships statistically
- Use multiple analytical approaches
- Save corpus frequently to preserve metadata
- Revisit and refine relationships as analysis progresses
