Diff Terms
Diff Terms (Differentiating Terms) reveals which words and phrases are disproportionately linked to a specific category in your data. Where Top Terms shows what’s most common overall, Diff Terms shows what makes one group’s language distinctive compared to another — surfacing the terms that truly differentiate.
Adding a Diff Terms analysis
Section titled “Adding a Diff Terms analysis”- In the Workbench toolbar, click Diff Terms.
- The Diff Terms configuration panel opens. Work through the Options and Advanced tabs.
- Click Save to add the chart to your output library.
Options tab
Section titled “Options tab”Define two categories
Section titled “Define two categories”Diff Terms compares language between exactly two user-defined categories. You choose a supporting variable, select which values belong to Category 1, and which belong to Category 2. Each category gets a label you can customise.
- In the variable selector, choose the supporting variable you want to use — for example, an NPS score, satisfaction rating, or customer type.
- Select the values that make up Category 1 (for example, scores of 9–10 for Promoters). Give it a meaningful label.
- Select the values that make up Category 2 (for example, scores of 0–6 for Detractors). Give it a meaningful label.
- The chart shows which terms are disproportionately associated with each category.
Chart types
Section titled “Chart types”| Chart type | Best for |
|---|---|
| Bubble | Showing term differentiation and frequency as a scatter-style plot |
| Diagonal | Plotting each term’s position relative to both categories on diagonal axes |
| Horizontal | Side-by-side horizontal bars comparing term association per category |
| Dumbbell | Connecting each term’s position in Category 1 and Category 2 to highlight gaps |
| Lollipop | A cleaner alternative to bars for ranked term differentiation |
Minimum mentions
Section titled “Minimum mentions”Set the minimum number of times a term must appear within a category to be included. This prevents terms that appear only once or twice from skewing the results. Enable Use Dynamic Minimum Mentions to let the system calculate a threshold automatically.
Significance filter
Section titled “Significance filter”Control which terms are highlighted as statistically significant:
- Significance Filter — Toggle on to apply significance filtering to the chart
- Significance Filter Threshold — Set the p-value threshold (default: 0.05). Terms that don’t meet this threshold are de-emphasised
- Significance Filter Mode — Choose whether significance is shown by Point (highlighting the data point) or Label (highlighting the term label)
Category display amount
Section titled “Category display amount”Set how many terms are shown per category in the chart. The default is 6.
Advanced tab
Section titled “Advanced tab”Ignore list
Section titled “Ignore list”Add terms to exclude from the Diff Terms chart — for example, product names or other terms that appear frequently across all categories and therefore don’t differentiate.
- Type a term into the Ignore list field.
- Press Enter or click Add.
- To remove a term, click × next to it.
Frequently asked questions
Section titled “Frequently asked questions”How is “differentiating” calculated?
A term is considered differentiating when it appears significantly more often within one category than would be expected based on its overall frequency in the dataset. The algorithm accounts for the relative size of each category, so it’s not simply about which terms appear most often in a group.
Can I compare more than two categories?
No — Diff Terms always compares exactly two categories. If you need to compare multiple groups, run separate Diff Terms analyses with different category definitions, or use Data Splits to segment your results.
My Diff Terms chart looks very similar to my Top Terms chart — why?
This can happen if your categories don’t differ much in the language respondents use, or if the dataset is small. Try using a variable with more distinct groups — for example, comparing the highest and lowest satisfaction groups rather than adjacent scores.