Key driver analysis is a statistical analysis that examines how strongly the categories or items in an engagement survey are related to the engagement index. It answers the question "What drives engagement most?" and shows where an organization should spend its limited resources. The core idea is simple: each area's score and its impact on engagement are read together, and areas with a low score and a high impact are the priority action areas. In this article we explain how the analysis is done, how to interpret it and the most common mistakes.
- Key driver analysis shows which areas of the survey have the strongest relationship with the engagement index; it is also called impact analysis.
- Priorities are set by reading score and impact together: areas with a low score and a high impact on engagement are the priority action areas.
- Asking employees "what matters most to you?" usually puts pay at the top; statistical analysis shows what actually makes the difference in engagement.
- The analysis needs enough responses to be reliable; it shows relationships, not causation, and should be read together with open-ended comments.
Why Looking at the Lowest Score Is Not Enough
Most managers who open an engagement survey report look first at the lowest-scoring category. That category is usually pay and benefits. But the lowest-scoring area is not necessarily the one with the greatest impact on engagement.
For example, pay may score low, yet employees who are satisfied with their pay and those who are not may show similar levels of engagement. By contrast, employees who view career development positively may be markedly more engaged, and those who view it negatively markedly less so. In that case the most effective lever for raising engagement is career development, not pay.
Driver analysis brings exactly this difference to light. The score shows "where we are"; the impact shows "what makes the difference."
Stated Importance vs. Statistical Importance
Asking employees directly "what matters most to you?" is also a method. This is called stated importance. Pay and job security almost always come out on top, because they are basic expectations everyone cares about.
Statistical (derived) importance does not ask employees what they want; it looks at which areas' responses move together with the engagement score. The two approaches complement each other:
| Stated importance | Statistical importance | |
|---|---|---|
| How is it measured? | Employees are asked to rank what matters | Relationships between survey responses are analyzed |
| What does it show? | What employees consciously value | The areas that actually differentiate engagement |
| Typical result | Pay, job security and benefits come first | Areas such as the manager, career, recognition and trust in senior leadership come first |
| Risk | Highlights areas everyone values but that do not differentiate engagement | Cannot see the impact of an area the survey does not ask about |
When basic expectations such as pay are not met, they create a serious problem; once they are met above a certain level, however, they may not be as effective as other areas at raising engagement. So the finding "employees care about pay" does not contradict the finding "career development differentiates engagement most."
How to Run a Driver Analysis
1. Define the Outcome Variable
The outcome variable is usually the engagement index: the average of a few items that measure engagement (for example "I am proud to work for this organization" and "I would recommend this organization as a place to work"). We describe how the index is structured in our article on what an employee engagement survey is. Some organizations also use eNPS or an "intention to stay" item as the outcome variable.
One important rule: the items that make up the index must not be included in the analysis as drivers. Otherwise the index explains itself and the result becomes meaningless.
2. Define the Drivers
The drivers are the other categories or items in the survey: the manager, career and development, recognition, trust in senior leadership, workload, pay, the work environment and so on. Analysis at category level gives the overall picture; analysis at item level gives more concrete findings. Good practice is to identify the priority areas at category level first and then drill down into the items within those categories.
3. Choose the Method
| Method | How it works | Strength | Limitation |
|---|---|---|---|
| Correlation | Measures the strength of each area's relationship with engagement separately (from -1 to +1) | Simple, easy to understand and can be calculated even for smaller groups | Cannot separate the shared effect of interrelated areas |
| Multiple regression | Puts all areas into the model at once and estimates each one's effect "holding the others constant" | Shows each area's independent contribution | When areas are highly correlated, results become unstable and some areas' effects can wrongly come out close to zero |
| Relative importance (relative weights, dominance) | Divides the total explained effect among the areas | Gives fairer and more stable results for interrelated areas; results can be read as percentages | More complex to calculate and requires expert support |
High correlations between survey responses are normal: an employee who is satisfied with their manager usually also views recognition and development more positively. That is why classic regression on its own can be misleading. In practice, correlation is a good starting point for seeing the direction; for important decisions, relative importance analyses are more reliable.
4. Read Correlations Correctly
A rough benchmark widely used in the social sciences for interpreting the correlation coefficient (r) is:
- Around 0.10: weak relationship
- Around 0.30: moderate relationship
- 0.50 and above: strong relationship
In engagement surveys, most categories are positively related to engagement. What matters is less the absolute value than how the areas rank relative to each other.
5. Read Score and Impact Together
For each area you get two values: the score (the percentage of favorable responses) and the impact (the strength of its relationship with engagement). These two values place each area in a priority matrix that divides the areas into four groups:
| Low score | High score | |
|---|---|---|
| High impact on engagement | Priority action area: The strongest lever for raising engagement | Strength to protect: A source of engagement for the organization; it must not be allowed to weaken |
| Low impact on engagement | Area to monitor: May matter, but does not differentiate engagement | Area to maintain: The current state is sufficient |
The lines dividing the matrix are usually the areas' average (or median) score and average impact. This way areas are classified relative to each other rather than against an absolute threshold.
Example: A Driver Analysis Table
| Category | Favorable response rate | Correlation with engagement (r) | Class |
|---|---|---|---|
| Career and development | 52% | 0.62 | Priority action area |
| Trust in senior leadership | 55% | 0.60 | Priority action area |
| Recognition | 48% | 0.55 | Priority action area |
| Relationship with manager | 74% | 0.58 | Strength to protect |
| Workload and balance | 61% | 0.34 | Area to maintain |
| Pay and benefits | 38% | 0.29 | Area to monitor |
| Work environment and tools | 70% | 0.22 | Area to maintain |
The values in the table are for illustration only. The lines dividing the matrix are the areas' average score (about 57%) and average correlation (about 0.46).
In this example pay and benefits has the lowest score, but its relationship with engagement is weaker than the other areas'. Career and development, trust in senior leadership and recognition all score below average and have the strongest relationships with engagement. This year's priorities should be chosen from among these three areas. The relationship with the manager is one of the most important sources of engagement; keeping this positive picture should be considered, for example, when managers change and when developing new managers.
Pay coming out as an "area to monitor" does not mean pay is unimportant. If pay falls clearly below the market, it can increase intention to leave; that is why pay results should be tracked alongside intention to stay and exit survey data.
Drilling Down to Items: "Career" Is a Broad Concept
The finding "career and development is a priority area" is not enough on its own to act on. Looking at the items within the category makes the source of the problem clear:
- If the item "The criteria for career progression here are clear" scores low and has a high impact, the problem lies in the transparency of the promotion process.
- If the item "My manager makes time for my development" scores low, the problem lies in managers' development conversations.
- If the item "I have opportunities to learn new things" scores low, the problem lies in training and job enrichment opportunities.
The action for each of these three situations is completely different. Support item-level findings with open-ended comments: if the topics that come up most often in career-related comments match the item the analysis points to, the priority is confirmed. We explain how to analyze comments in our article on AI analysis of open-ended comments.
Running Driver Analysis by Segment
The areas that drive engagement are not the same in every employee group. Career development may stand out for office employees, while the relationship with the manager, shift patterns or occupational safety may matter more for frontline employees. Onboarding and fitting into the team may stand out for new joiners, while recognition and new responsibilities may matter more for long-tenured employees.
That is why running a separate driver analysis for sufficiently large groups helps identify actions suited to each group. However, driver analysis requires far more responses than calculating an average. As a rough rule:
- For correlation: Around 100 responses or more in the group is recommended. In smaller groups, correlations can swing considerably by chance.
- For regression and relative importance: At least 10–20 responses are needed per driver. For example, a model analyzing 10 categories at once should have at least 100–200 responses.
Running driver analysis for every team that passes the 10-person reporting threshold is not the right approach. For small teams it is sounder to use the drivers found at organization or unit level and to look at the team's own scores to see which of those drivers it lags behind on. We explain how to build groups in our article on segmentation in employee surveys.
Presenting the Results to Leadership and Managers
The clearest way to present driver analysis is to show the priority matrix as a scatter chart: the score on the horizontal axis, the impact on the vertical axis and each point a category. The top-right corner shows strengths to protect, the top-left corner shows priority action areas. This single chart answers the leadership question "Why are we focusing on career rather than pay?"
Points to keep in mind when presenting:
- Simplify the technical language: Instead of "the correlation coefficient is 0.62," say "employees who view career development positively are markedly more engaged."
- Limit the number of priorities: Choose 2–3 priority areas across the organization; prioritizing everything means prioritizing nothing.
- Give managers a ready framework: Team reports that show the organization's priority areas side by side with the team's scores in those areas make it clear where managers should start.
We explain the whole process from analysis to the leadership presentation step by step in our guide on how to analyze employee survey results.
Common Mistakes
- Confusing correlation with causation: Driver analysis shows that an area and engagement move together; it does not prove that improving that area will definitely raise engagement. Validate findings with open-ended comments, focus groups and the change in the next measurement.
- Running the analysis on small groups: Correlations calculated for a team of 30 can come out completely different in the next survey.
- Using index items as drivers: The items that make up the engagement index must not enter the analysis as drivers that "explain" the index.
- Relying only on classic regression: With interrelated areas, some areas' effects can wrongly come out close to zero or even negative. Check such results with relative importance analysis.
- Overlooking the ceiling effect: For an item that almost everyone answers favorably, the correlation comes out low because there is not enough variation between responses. This does not mean the area is unimportant; it may mean it is already in good shape.
- Not seeing what is not in the survey: Driver analysis only ranks the areas the survey asks about. If a topic that is critical for the organization (such as the hybrid work arrangement) is not in the survey, it will not appear in the analysis either. When choosing items, you can draw on the categories in our article on employee engagement survey questions.
- Changing priorities every year: Lasting improvement in an area usually takes more than a year. Instead of switching the priority area every year based on new findings, track progress in the same area over two or three measurements.
Driver Analysis and Engagement Models
Engagement models such as Gallup Q12 offer a general framework for the conditions under which engagement develops. Driver analysis shows how that framework works in your own organization: even in two organizations using the same model, the areas that drive engagement most can differ. We explain the differences between the models in our article on Gallup Q12 and employee engagement models.
Tracking driver analysis results from year to year also matters. If the steps taken in a priority area worked, that area's score rises in the next measurement and, over time, is reflected in the engagement index. We explain how to tell whether such a change is real in our article on comparing engagement survey results year over year.
Frequently asked questions
What is key driver analysis?
It is a statistical analysis that examines how strongly the categories or items in an engagement survey are related to the engagement index. It shows which areas make the biggest difference to engagement and is used to set action priorities. It is also called impact analysis.
Which methods are used for driver analysis?
The most common methods are correlation analysis, multiple regression and relative importance analyses (relative weights, dominance analysis). Correlation is simple and easy to understand; relative importance analyses separate the effects of interrelated areas more fairly.
How many responses does driver analysis need?
There is no strict threshold, but as a rough rule at least 100–200 responses across the organization are recommended, and at least 10–20 responses per area when regression is used. That is why driver analysis should be run for sufficiently large groups, not for every small team that passes the reporting threshold.
Does driver analysis show causation?
No. The analysis shows how strong the relationship between an area and engagement is. A strong relationship suggests that improving that area is likely to raise engagement, but it is not definitive proof of cause and effect. Open-ended comments and focus groups should be used to validate the finding.
