Comments on: Our First Puck Experiment http://groklab.org/handhygiene/2010/02/18/our-first-puck-experiment/ Just another Groklab.org weblog Mon, 28 Nov 2011 02:43:21 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: gthomas http://groklab.org/handhygiene/2010/02/18/our-first-puck-experiment/comment-page-1/#comment-120 gthomas Fri, 19 Feb 2010 15:42:07 +0000 http://groklab.org/handhygiene/?p=598#comment-120 Great start. However, I think that Phil and I were looking for a slightly different analysis. I think the statistics you present measures the interarrival time of the pucks as a function of the interarrival time recoded by the experimenters. My guess is that the R^2 value is largely dependent on the accuracy of the experimenters -- the differences between when the mote was actually pressed and the time the experimenters recorded by hand. What we are most interested in are false alarms and misses. To find this you need to set up a 2x2 grid for each experiment. The two columns should be labeled "Dispenser Pressed" and "Dispenser Not Pressed". The two rows should be labeled "Puck Activated" and "Puck Not Activated". Divide the time sequence into periods during which the protocol indicates a period during which the puck should have been activated (perhaps a 5 second period) and a period when it should not have been activated (perhaps a 15 second period). In experiment 1 we had 40 puck activations and 39 periods during which there was no puck activations. Go through the puck activation record and check for each interval whether or not there was a puck activation. If the puck was completely reliable, it should have 40 activations when we predicted activations and 39 intervals in which there were no activations. There should also be zero periods when we predicated and activation but saw none and no periods when we expected no activation but saw one anyway. We should end up with a 2x2 matrix for each experiment. Great start. However, I think that Phil and I were looking for a slightly different analysis. I think the statistics you present measures the interarrival time of the pucks as a function of the interarrival time recoded by the experimenters. My guess is that the R^2 value is largely dependent on the accuracy of the experimenters — the differences between when the mote was actually pressed and the time the experimenters recorded by hand. What we are most interested in are false alarms and misses. To find this you need to set up a 2×2 grid for each experiment. The two columns should be labeled “Dispenser Pressed” and “Dispenser Not Pressed”. The two rows should be labeled “Puck Activated” and “Puck Not Activated”. Divide the time sequence into periods during which the protocol indicates a period during which the puck should have been activated (perhaps a 5 second period) and a period when it should not have been activated (perhaps a 15 second period). In experiment 1 we had 40 puck activations and 39 periods during which there was no puck activations. Go through the puck activation record and check for each interval whether or not there was a puck activation. If the puck was completely reliable, it should have 40 activations when we predicted activations and 39 intervals in which there were no activations. There should also be zero periods when we predicated and activation but saw none and no periods when we expected no activation but saw one anyway.

We should end up with a 2×2 matrix for each experiment.

]]>