Dear all

I have conducted an experiment in which subjects concurrently perform two tasks. Task A stays always the same, while Task B has two versions, B1 and B2. I furthermore manipulate the difficulty of B1 and B2 the same way (three levels). Dependent variable is performance in Task A.

When I fit a logistic regression model to my data containing the factors task type (B1 coded as 0, B2 as 1), task difficulty (three levels, 1, 2, and 3), and the interaction I find:

- a significant negative effect of task difficulty (the more difficult task B, the lower performance in A)
estimate is -0.25 with confidence interval [-0.43; -0.06]

- a significant negative effect of task type (task B2 affects task A more)
estimate is -1.55 with confidence interval [-2.34; -0.81]

- a significant POSITIVE interaction
estimate is 0.25 with confidence interval [0.12; 0.38]

My interpretation as a lay person is that the task difficulty effect is not there for task B2, so I conduct separate analyses for each task B, I unsurprisingly find a task difficulty coefficient of -0.25 for A/B1 and an almost 0 coefficient for A/B2. However, what happens is that the effect in A/B1 is not significant anymore (confidence interval now overlaps with 0).

How can that be?

Thanks
Scarlatti