Concept: Question Rotation Algorithm

Setting a question frequency dictates how often each employee should receive a question from each question set. When using an automated survey cadence, the survey asks all active questions in each survey round, but not to all employees at the same time. The algorithm checks the last time that each employee received each question, to determine whether they’re due again, for the specific employee.

The exception is manual survey frequency, which triggers all active questions of the question sets you enable. When using manual frequency, the survey runs once and it's not possible to set question frequency. The order of questions is the same regardless of manual or automatic question sampling.

For every employee, the algorithm prioritises questions and evaluates their importance based on:

The last time the question was in the employee’s survey.
The organization’s question frequency settings.
Number of questions per driver and subdriver.

Example: Subdriver questions are on a monthly frequency. The last time the employee received the standard question from the Equipment subdriver was 10 weeks ago. The algorithm will assign the question higher importance as the question has been long due, according to the frequency settings.

The algorithm will assign lower importance to questions of a driver or subdriver and push its questions further down the queue if at least 1 of these conditions is true:

After assigning importance to the questions, the algorithm probabilistically selects due questions from the top of the queue.

The main driver question takes precedence over subdriver questions if the employee hasn’t received it at all.

If an employee submits a score that's 2 standard deviations away from the benchmark for that question, the algorithm follows up with subdriver questions to find out more information.

When a survey launches, the visibility window (Example: 1 year) shifts forward, enabling new scores to enter the mix and old scores to drop out of the aggregated dashboard. Note that even when certain scores fall outside of the visibility window, you can still access dashboards that include them when viewing historical data.

If you deactivate a driver question, existing scores fall out of the current score with each new survey round, without new scores coming in.

The question rotation algorithm helps gather feedback on every aspect of your company’s culture, and present trends to you, as quickly as possible. Not asking all questions to all employees at the same time enables you to keep surveys short, and provides high-quality feedback by avoiding survey fatigue.

Peakon distributes driver questions by random sampling to ensure that the dashboard results are unbiased. We can then use the statistical properties of the scores it observes, to assess when segments are significantly different from the benchmark. We only assign focus areas and strengths once they pass a statistical test that the results are significant.

Only a relatively small number of results are required to obtain high quality estimates, and the accuracy and stability of segment scores quickly stabilizes as employees complete further rounds.

In extremely small segments we would advise waiting until a majority of employees have answered before conducting any comparisons. In larger segments this restriction can be relaxed, and in very large segments, even a small proportion of responses can give a highly accurate result.

There is no issue of statistical validity with comments or topics, which can give instant insight into ongoing issues.