Reading a result

A finished study hands you a number. The question that follows is always the same —«is that good?»— and it has no answer in the abstract. This page is about getting to one that is actually useful.

What to look at, in what order

1. The overall index, but only to orient yourself. It is the least actionable number on the screen: a whole organization averaged has no «where» to act. It tells you whether you are in comfortable territory or not, and nothing more.

2. The categories. This is where it gets useful. An organization is rarely «bad»: it is bad in two categories and fine in the rest. Those two are the year's work.

3. The branches. The same category can be healthy in half the company and broken in the other half. If the problem is concentrated, the plan is local and cheap; if it is spread evenly, it is a policy and it is expensive. Two very different diagnoses, and the difference shows up here.

4. The distribution, for the ones that worry you. The average hides the shape. A 60 can be everyone lukewarm, or half the company delighted and half furious. The second is urgent and the first is not, and in the average they look identical.

5. Segments, last. Crossing branch with sex, age and tenure is the finest tool and the one that most easily leaves you with nothing publishable. Use it to confirm a suspicion, not to explore blind.

Question by question

A study's Statistics tab shows the indices and the categories. To look at a single question there is the By question tab: the distribution of answers to that question in each branch, one question at a time.

• You go through one section at a time. The «Section» selector, next to Filters, chooses between Satisfaction, Work climate, Recommendation (eNPS) and Additional, and only offers the ones the study has. The arrows take you to the next question, and «Search or jump» takes you straight to one by its text.
• The chart is set by the question type, not chosen. Below it, a table gives the exact distribution per branch. See Question types.
• A question that depends on an earlier answer appears inside the card of the question it depends on, and only counts the people who reached it.
• Open-ended answers are read exactly as written, grouped by branch and under the same confidentiality threshold as the rest of the screen. That threshold protects the numbers, not the content: a comment can give away its author through what it says, so read them without trying to guess who wrote them.
• The filter is the same as in Statistics: switching tabs does not lose it. From Reports you can download «The question you're viewing» with that filter applied. Open-ended answers are not exported.

If a question carries the «Approximate wording» label, its text changed and it cannot be guaranteed to be the one people read when they answered: what you see is the current wording.

What «acceptable» means

In tables and distributions you will see responses grouped into acceptable and not acceptable, plus an acceptance percentage per question.

«Acceptable» groups the values on the favourable side of the scale. The criterion is the same across every scale in the product and does not change between studies, so the acceptance percentage is comparable across questions, across branches and over time. It is the most useful number on the screen when you want to follow something.

Where exactly the cut falls on each scale is visible in the product: the distribution shows the full spread, value by value, alongside the grouping.

The traffic light

A category's color is a reading aid, not a verdict. It marks where to look first when twenty categories are on screen; it does not mean red is a breach or green is the end of the work.

A red in a category your organization does not care about this year is less urgent than an amber in one it does. The traffic light does not know that.

eNPS has its own traffic light

eNPS is not a 0 to 100 percentage but a subtraction running from −100 to +100, so it does not use the same cuts as the rest of the screen:

Red means there are more detractors than promoters. An eNPS of +40 is not comparable with a 40 % acceptance rate or with an index of 40: they are three different numbers that only happen to share an axis. See Question types.

Four conclusions you cannot draw

«Climate is six points below Satisfaction.» No. They are computed by different rules, so subtracting them describes the difference between two methods, not something about your company. See Satisfaction and Climate.

«This branch is three points worse than that one.» With small samples, three points is noise. Before reading a difference as a finding, look at how many responses sit on each side.

«It dropped because we changed the area manager.» The study measures state, not cause. Two things happening in the same period does not relate them. What you can do is use the result to know who to ask why.

«In the ranking, this option took 40 %.» Ranking questions are multiple-select, so the counts add up to more than the responses. A percentage there is not a share — which is why the product does not normalize them to 100 % and shows counts instead.

When something cannot be seen

If a result comes out hidden, it is neither a bug nor a permission: it is the confidentiality threshold, and it sits above every role. It is almost always solved by segmenting more coarsely.

See Anonymity and minimum N.

What to do with all of it

The useful output of a study is not the report: it is two or three concrete actions with an owner, chosen by the lowest category in the branch with the most people.

Everything else —the remaining categories, the mid-sized branches, the interesting segments— is material for next year. A plan that attacks ten things at once changes none of them, and the next study will show it.