Key driver analysis: see which questions move your score

Key driver analysis shows which of your questions move together with the outcome you are tracking. The outcome can be an NPS, rating or slider question. Instead of comparing one pair of questions at a time in a crosstab, you get every eligible question ranked in one view.

It is available on Growth plans and above.

What it ranks

You choose one outcome question. SurveyRock then ranks the other eligible questions, called drivers, by their share of the variation in that outcome that the answers explain. It does this with a method called Johnson’s relative weights.

• Outcomes you can choose: a rating, NPS or slider question. Each statement of a slider counts as its own question.
• Questions that can be drivers: rating, NPS and slider questions, yes/no questions, rows of a single-answer grid, and cells of a dropdown grid.

Other question types are not used as drivers. Open-text, email, date and upload questions are left out without comment. Questions whose answers have no natural order are listed under Questions that can’t be used, each with the reason. These include multiple-choice, ranking and constant-sum questions, grids that allow several answers per row, and questions asked once per loop item.

The ranking is descriptive. It tells you which answers move together with the outcome. It does not prove that one thing causes another, so read it as a place to start asking questions, not as a verdict.

Run the analysis

1. Open the survey and click Responses in the top navigation.
2. In the sidebar on the left, select Key Drivers.
3. Under What do you want to explain?, choose the outcome you want to understand, such as your NPS question or an overall satisfaction rating.
4. Optionally, open Questions to compare and clear any questions you want to leave out. Every eligible question is selected to start with.
5. Click Find key drivers.
6. Read the ranked list. The question that moves most together with your outcome comes first.
7. Read any caveats shown on the result before you act on it. They appear on the result itself, so you can see how far the data supports the ranking.

A respondent counts only if they answered the outcome and every question you are comparing. Leaving out questions that few people answered keeps more respondents in the analysis.

Read the quadrant chart

Below the ranked list, the result includes an Importance vs. performance chart. It places each driver by how much it matters to the outcome and by how well it scores. Performance is the question’s average answer on a 0 to 100 scale, where 0 is the worst possible answer and 100 is the best.

A driver that matters a lot and scores low is the one to look at first. Those drivers are listed under Fix this first, below the chart. A question with no defined scale stays in the ranked list but is left off the chart, and the chart says so.

Narrow, weight and export the result

• Filter. The Key Drivers page has its own filter bar above the form, so you can look at one group of respondents at a time. Set the filter, then click Find key drivers again.
• Weighted view. If the survey has an active weighting scheme, use the Unweighted / Weighted switch in the Responses toolbar to run the analysis on weighted answers. Response weighting is on Pro and above.
• Export. After a result appears, click Export and choose CSV or Excel (.xlsx). The file reflects the outcome, questions, filter and weighting you used. Exporting needs a role that allows exporting responses.

Turn a driver into an action

When a driver stands out, you can create an action from it so someone owns the follow-up. Use the plus icon, labelled “Create an action for this driver”, or click the driver in the chart. See Turn a theme, a driver, or an answer into an action.

Example

A team runs a customer satisfaction survey. It asks an overall satisfaction rating and three rating questions: how easy setup was, how quickly support replied, and how fair the price felt.

They choose the overall satisfaction rating as the outcome. The analysis ranks the three questions by how much each moves together with it. Support speed comes out on top. On the chart it sits in the high-importance, low-score corner.

That does not prove faster support would raise satisfaction. It does tell the team where to look first. They create an action from that driver and assign it to the support lead.

When the analysis won’t rank

We would rather show you nothing than a ranking that misleads. The analysis declines to rank in these cases, and the message on screen says which one applies:

• Too few complete respondents. It needs at least 30 respondents who answered the outcome and every compared question, and at least 5 for each question you compare. Comparing 10 questions needs 50. Collect more responses, compare fewer questions, or leave out the ones few people answered.
• Fewer than two questions to compare. It needs at least two drivers besides the outcome, and each needs a spread of answers. A question everyone answered the same way is left out, and the result says so.
• Duplicate or inseparable drivers. Two or more questions that the data cannot tell apart. For example, the survey asks nearly the same thing twice. Remove one from the comparison and run it again.
• An outcome nobody varied on, or nobody answered. If every respondent gave the same answer to your outcome question, there is nothing to explain. Try a different outcome, widen the filter, or wait until answers differ.
• No relationship at all. None of the compared questions move together with the outcome in this selection.

Troubleshooting

The question I want isn’t on the list. Only the question types listed above can be drivers. Open Questions that can’t be used at the bottom of the page to see each excluded question and the reason.

The page says “Nothing to analyse yet.” The survey has no rating, NPS or slider question to use as the outcome. Add one, or run the analysis on a survey that has one.

Key Drivers shows a padlock. It needs a Growth plan or above. On a lower plan, the page shows an upgrade notice instead of the analysis. See Reading your results for the analysis tools your plan includes.

A filtered view is refused on an anonymous survey. On a survey that collects responses anonymously, a filtered view is refused unless enough completed responses match the filter. The default minimum is 5. It can be raised with Minimum group size in Survey Settings, never set below 5, and can’t be lowered once responses exist. Widen the filter to include more respondents.

The ranking looks wrong to me. Check the caveats shown with the result first. Then check whether one of the refusal cases above is close to applying, for example two questions that are very similar. If it still looks wrong, contact support and tell us which survey and outcome you used.