Variable Charts Overview

What Are Variable Charts?

Variable Charts in MixModeler allow you to visualize your data with interactive charts to explore patterns, trends, and relationships between variables before and during model building.

Purpose: Explore data visually to understand trends, relationships, and patterns that inform model specification and validate assumptions.

Why Use Variable Charts?

Data Exploration:

  • Understand time trends before modeling

  • Identify relationships between variables

  • Detect seasonality and patterns

  • Spot outliers and data quality issues

Model Validation:

  • Verify expected relationships exist

  • Check for multicollinearity visually

  • Compare raw vs. transformed variables

  • Validate model assumptions

Communication:

  • Create publication-ready visualizations

  • Share insights with stakeholders

  • Support data-driven discussions

  • Illustrate key findings

Accessing Variable Charts

Navigation:

  1. Click "Variable Charts" in the left sidebar (under Data Management)

  2. Select your model from the dropdown

  3. Choose variables to visualize

  4. Select chart type

  5. Generate chart

Prerequisites:

  • Data must be uploaded

  • At least one model created (to access variables)

  • Variables available for charting

Available Chart Types

MixModeler provides five chart types for different analytical needs:

Chart Type
Best For
Variables Needed

Line Chart

Time trends, temporal patterns

1+ variables

Scatter Plot

Relationships between two variables

Exactly 2 variables

Bar Chart

Period comparisons, discrete data

1+ variables

Stacked Chart

Part-to-whole relationships, mix analysis

2+ variables

Correlation Matrix

Multi-variable correlations, multicollinearity

2+ variables

Typical Workflow

Step-by-Step Process:

  1. Start with Line Charts

    • Understand individual variable trends over time

    • Identify seasonality and patterns

    • Check for outliers

  2. Use Scatter Plots

    • Explore relationships (e.g., Spend vs. KPI)

    • Validate expected correlations

    • Detect non-linear patterns

  3. Check Correlation Matrix

    • Identify multicollinearity before modeling

    • Find redundant variables

    • Understand variable relationships

  4. Compare Transformations

    • View raw vs. transformed variables

    • Validate transformation effects

    • Ensure transformations make sense

  5. Use Stacked Charts

    • Analyze media mix over time

    • Understand budget allocation

    • See proportional changes

Chart Interface Layout

Left Panel - Variable Selection:

  • Model dropdown

  • Variable search box

  • Variable checkboxes with info

  • Chart type buttons

  • Options (dual axis, KPI inclusion)

  • Generate Chart button

Right Panel - Chart Display:

  • Title editor (customizable)

  • Interactive chart

  • Zoom/pan controls

  • Legend

  • Tooltip on hover

Variable Selection

Search Functionality:

  • Type to filter variables by name

  • Makes finding variables easier

  • Works across all variable types

Select All:

  • Quickly select all filtered variables

  • Uncheck to deselect all

  • Saves time with many variables

Individual Selection:

  • Check/uncheck specific variables

  • See variable details:

    • Type (KPI, Feature, etc.)

    • Transformation applied

    • Group assignment (if set)

Variable Limit:

  • Line charts: 2-8 variables recommended

  • Scatter plots: Exactly 2 required

  • Correlation: 2+ variables

  • Stacked charts: 2+ variables

Key Features

Dual Y-Axis Support

When to Use:

  • Comparing variables with very different scales

  • Example: Spend ($1000s) vs. Percentage (0-100)

  • Prevents one variable from being flattened

How It Works:

  • First half of variables use left Y-axis

  • Second half use right Y-axis

  • Each axis scales independently

Include KPI Option

Purpose:

  • Add KPI variable to any chart

  • See relationships with target variable

  • Validate expected correlations

Use Case:

  • Check if marketing spend correlates with KPI

  • Verify timing of impacts

  • Identify leading/lagging relationships

Transformed vs. Raw Variables

Toggle Option:

  • View transformed variables (adstock, curves, etc.)

  • Compare transformation effects

  • Validate transformations make sense

Note: Transformed variables are created in Variable Workshop and Model Builder

Customizable Chart Title

Edit Inline:

  • Click in title box to edit

  • Descriptive titles for exports

  • Professional presentation

Interactive Features

All charts include:

Zoom:

  • Mouse wheel zoom

  • Selection box zoom (click and drag)

  • Zoom in/out buttons in toolbar

Pan:

  • Drag to move around when zoomed

  • Navigate across time or data range

Tooltip:

  • Hover over data points for exact values

  • Shows date, variable name, value

  • Multiple series shown together

Legend:

  • Click to show/hide individual series

  • Isolate specific variables

  • Compare subsets

Toolbar:

  • Top right corner

  • Zoom controls

  • Pan toggle

  • Reset to original view

Common Use Cases

Pre-Model Exploration

Before building a model:

  1. Plot all potential variables as line charts

  2. Identify which show clear trends

  3. Check scatter plots for expected relationships

  4. Use correlation matrix to find redundant variables

  5. Remove or combine highly correlated variables

Model Validation

After fitting a model:

  1. Compare predicted vs. actual (line chart)

  2. Check residuals over time

  3. Validate relationships still hold

  4. Confirm no obvious patterns missed

Transformation Verification

When applying transformations:

  1. Plot raw variable

  2. Toggle to show transformed

  3. Verify transformation achieves desired effect

  4. Check for unintended consequences

Budget Mix Analysis

Understanding spend allocation:

  1. Use stacked chart with all spend variables

  2. See proportions over time

  3. Identify shifts in mix

  4. Inform budget allocation decisions

Multicollinearity Check

Before finalizing model:

  1. Run correlation matrix on all planned variables

  2. Identify correlations > 0.8

  3. Decide which variables to keep/remove

  4. Reduce VIF issues before modeling

Best Practices

Start Simple:

  • Begin with one or two variables

  • Understand individual patterns

  • Then add complexity

Use Appropriate Chart Types:

  • Time data → Line charts

  • Relationships → Scatter plots

  • Comparisons → Bar charts

  • Mix/composition → Stacked charts

  • Correlations → Matrix

Limit Variables Per Chart:

  • Too many variables = cluttered

  • 2-5 variables ideal for most charts

  • Use multiple charts if needed

Customize Titles:

  • Descriptive titles help later

  • Include date ranges if relevant

  • Professional for exports

Leverage Interactivity:

  • Zoom into specific periods

  • Hide/show variables via legend

  • Use tooltips for exact values

Save Insights:

  • Screenshot important charts

  • Export for presentations

  • Document findings

Tips for Success

Efficient Selection:

  • Use search to filter variables

  • Select All then uncheck unwanted

  • Saves time with many variables

Compare Scales:

  • Use dual axis when scales differ greatly

  • Prevents visual distortion

  • Makes both variables visible

Check Correlations:

  • Always run correlation matrix before modeling

  • Identifies multicollinearity early

  • Saves time in model building

Include KPI:

  • Add KPI to see target relationships

  • Validates variable usefulness

  • Identifies timing of effects

Iterate:

  • Charts inform next questions

  • Explore, discover, repeat

  • Build understanding gradually

Next Steps

After exploring with Variable Charts:

  • Proceed to Chart Types to learn about each chart in detail

  • Review Interactive Features for advanced usage

  • Learn about Export & Sharing to use charts in presentations

  • Apply insights in Variable Engineering and Model Building

PreviousExporting Decomposition DataNextChart Types

Last updated