The problem is fundamentally due to the level of likelihood you allocate to an event, that is variable $e_i$, should not be measured using a bi-polar scale. Instead, likelihoods should be associated with a percentage, between 0 and 100%. I could easily convert a bipolar scale to a percentage scale by using simple arithmetic. E.g. if the scale is -3, -2, -1, 0, 1, 2, 3, then I would convert these numbers to percentages as 0, 16.6, 33.3, 50, 66.6, 83.3, 100. This avoids the double negative problem, and is a more natural measurement of certainty than the bi-polar scale.

The problem is fundamentally due to the level of likelihood you allocate to an event, that is variable $e_i$, should not be measured using a bi-polar scale. Instead, likelihoods should be associated with a percentage, between 0 and 100%. This is a more natural unit, as a "low likelihood" result usually means that the respondent thinks the probability of the occurance to be low, e.g. 20%. I could easily convert a bipolar scale to a percentage scale by using simple arithmetic. E.g. if the scale is -3, -2, -1, 0, 1, 2, 3, then I would convert these numbers to percentages as 0, 16.6, 33.3, 50, 66.6, 83.3, 100. This avoids the double negative problem, and is a more natural measurement of certainty than the bi-polar scale.