Roslin Methodology

Questions

Core Principle

When calculating the score for a question for a given month, Roslin will first average each user's scores for that question for the month in question (i.e. in case they were asked the same question more than once for the month) and then each user's score will then be averaged.

This ensures that each user is equally weighted when contributing to a question's score.

The month in which the question was answered is the month which will be impacted and not the month in which the survey was sent out.

Carry Over

In order to provide larger samples for greater statistical significance and thus more reliable and smoother data, Roslin allows for user scores to be carried over for up to 3 months. This means that if a user answers a question and scores it a 7, then if that user doesn't answer that question again for three months, the 7 will be carried forward when calculating the score for the question.

The calculation for a question score (with carryover) is best illustrated below:

	January	February	March	April	May
User 1	7	8	8	8	9
User 2	7	7	7	7
User 3		6	6	7	7
User 4			4	6	7
User 5				7	8
User 6			7	7	8
Score	N/A	7.0	6.4	7.0	7.8

Notes

1) Underlined numbers represent actual data created when users answer surveys. The other numbers represent data that was carried over because that user didn't answer the question during that month. E.g. User two answered this question in Jan but not in the following months and thus their data was carried forward for Feb, Mar and Apr but finally dropped in May due to being stale.

2) For January, you can see that no score was calculated because only 2 data points were available which is less than the minimum of 3. These minimums are in place to protect staff anonymity.

Drivers

An Engagement Driver acts as a theme in the workplace and is a simple grouping of questions. It allows leaders to more quickly grasp their data. You can read more about Engagement Drivers here.

Drivers are calculated as the simple average of their underlying question scores. This is done with an equal weighting so questions aren't over or under-represented despite their volume of answers as long as they meet the minimum threshold for the relevant month.