What is it?
The standard deviation value represents how confident the inference engine is in its prediction of cognitive load. When confidence is very high, the standard deviation value will be low (i.e. the bar around the cognitive load value will be thin). Alternatively, when confidence is low, the standard deviation value will be high (i.e., the bar around the cognitive load value will be thick). The standard deviation represents a summary value of the uncertainty of multiple models.
What to do with it?
A confidence value greater than .1* is an opportunity to investigate what happened in the task or simulation at this point in time. A confidence value will be added and subtracted from the recorded cognitive load at that time which creates a range of what the user's actual cognitive value is. A confidence value that creates too wide a range means you cannot make a good determination of where the cognitive load truth lies.
Reported cognitive load: .45
Confidence value: .15
Actual cognitive load : [.3, .6]
Depending on your implementation’s ranges for your task, this could be low cognitive load to medium high, and not revealing any clarity about the user's cognitive load in the experience.
*This >=.1 flag is determined by you, you may want to be more sensitive to flag=.04, etc.
What to do about it?
It is useful to go back to the experience that caused this confidence spike. Is it reproducible? Is there another unintended task or experience happening that is causing this deviation?
See more on confidence value in our Explainer Series video coming in March.