Anthony M Harrison

First the obvious numbers:

Pointing latency and error: the only significant factor is position. Specifically, baseline=stay < rotate for latency and pointing error. (error bars are 95% CI)

Interestingly, neither the type of the secondary task nor the execution of it seem to make any difference.

So, doing the 1back (verbal or spatial) doesn’t affect pointing performance at all. I’d initially suspected that the spatial 1back would interfere slightly when subjects don’t move and much, much more when subjects are rotating. Verbal was expected to interfere slightly, but much less than the spatial.

What does the 1back data look like? Here we have some salvation: load (single/dual), load x condition (verbal/spatial), position (stay/rotate), position x condition, and load x position x condition are all significant. The take home story is that rotate < stay (accuracy), dual < single (finally some interference), spatial.rotate < verbal.rotate < spatial.stay=verbal.stay (@ceiling). But the three way interaction, plotted below is the final kicker.

So, while the pointing data didn’t show the predicted interference, the 1back data does.

The distribution of the pointing errors is also interesting in that it points towards an offline updating strategy. Here the proportion of errors are divided into three bins based on when they occurred. Again, we’ve got significances across the board: position, when (early/middle/late), position x when, position x when x cond, load x when x condition, load x pos x when x cond. The full graph is impossible to parse - so let’s ignore the nback only condition and just look at nback&pointing.

A whole slew of crap to parse in this graph - the solid lines (stay) show a constant distribution of 1back errors, across time and 1back type. Even verbal.rotate is equivalent. However, when subjects are engaged in the rotation while doing the spatial 1back, their errors are clustered towards the tail end of the trial (we’re dealing with proportions here, so the fact that spatial.rotate errors are less early and middle is meaningless - the actual errors are equivalent if you look at the scores).

From my current theoretical perspective, this shows that subjects are sacrificing the spatial 1back performance at the end as they try to figure out where the targets are before they have to point to them.

I think that I should probably be using the actual error scores (points lost) for each time block since the proportions hide the absolute decrements.

There are some pointing error distributions to look at as well, but that story hasn’t changed much since the last analysis. Although I still have yet to consider a viable theory for the reduction in pointing biases in the dual task condition.

One question to ask though, is why the story has changed for the 1back pointing. The previous analyses showed the tail end bias, but equivalence between spatial and verbal. The only thing I can come up with right now is that with the simplification of the experiment (3 targets instead of the 4) might have reduced the variability in the spatial 1back numbers - providing a better estimate - but I’m too lazy to go back and check.