Eye Tracking–Pupil Data

In the previous post below I said that the graph tool in the lower left corner of the tool was producing some very interesting results.  Now, I’m going to talk about what those results are.   The graph is there to graph numerical data such as gaze counts in areas of interest and so forth.  What I have it plotting now are the pupil diameters as I have not yet finished my area-of-interest definition routines.

The analysis tool.


These data are from the method and experiment described earlier. For example, the screen above shows the control group’s data during the 10 conditioning trials.  The top left shows where the participants were looking (yellow points) during the second half-second of the CS (they were looking at the CS and the area where the US was expected to arrive).  The bottom left shows the diameter of the participants’ pupils during each .5 second of the CS (the five seconds before the US arrives).    What is very interesting is that initially the pupil diameter doesn’t change (see left most points on trial 1) but then they show an increase over the duration of the CS as conditioning proceeds.   Is that a conditioned response? 

These graphs aren’t beautiful graphs because they are generated by the program, rather than something like Excel where you can go in after the fact and pretty them up.  Nevertheless they suffice for a quick overview of the data.  The tool is setup so that I can double-click a graph (or the top left image) and save it to file directly.

Is that increase a conditioned response?   Lets look at the reaction to the US.

Reaction to the US by Group (latent inhibition vs. control).


These are the pupil responses to the appearance of the US over the 10 conditioning trials.  The data are shown in 1-s bins.   I believe there are two notable patterns here.  First, the US consistently produces a rapid, but transient, increase in arousal (the dilation of the pupil).

Second, this pattern appears to diminish over trials in the control group (turquoise triangles), slightly more so than in the latent inhibition group (gray asterisks).   The behavioral data to the CSs (shown some blogs below) showed that the LI group was retarded in acquiring the behavioral response.   To the extent that a good CS should allow some moderation of the effect of the US, then these data suggest that the response produced by the CS is moderating the response to the US, thus it seems likely that the pupil response to the CS is a conditioned response.  Of course, they could be habituating to the US, but, that it appears faster in the control group than in the LI group suggests it is something more than simple habituation (unless we view it, as we should , from Wagner’s 1981, perspective.)

Reaction to the CS and US in the Control Group on trials 1 and 10 of conditioning.

Control 1 vs 10

The graph above shows the pupil response on the first and last trials of conditioning.  The first 10 points show the response during the first 5-s of the CS (in .5-s bins) before the US arrives.  The remaining 30 points (separated by the gap) show the response during the presentation of the CS and the US. 

The first trial is Orange, last trial is Red.  On the first trial the response to the CS is flat, and the response to the US is huge.  On the last trial the response to the CS increases over time, and the response to the US is reduced.

Reactions during the baseline and CS during pre-exposure in Group LI

LI pre-exposure PRE vs CS

The data above show the response to the CS in the LI group during the six pre-exposure trials (separated by gaps).  The graph also shows the responding during the Pre-CS period.  Take note that these two sets of data are being presented over different bins.   The CS was present for 20-s, while I only measure in the pre-CS for 5 seconds.    To facilitate comparison of the overall level of responding I present the CS data in 2-s bins (shown in Red) and the pre-CS in .5 second bins (Orange).  What is clear is that the CS doe not lead to an increase in responding over pre-exposure.  Rather, responding to the CS seems to habituate to a level somewhat below the pre-CS baseline.

Reaction during the baseline and CS during conditioning in Group LI

LI-Pre x CS conditioning

The data above show the pupil responses in 1-s bins during the first 5-s of the CS (Orange) and during the 5-s pre-CS period (Red) during the conditioning phase (trials separated by gaps).    It appears as though, over trials, the CS comes to elicit an increase in pupil diameter, though it appears slowly, only becoming evident after trial 5.

Reaction during the baseline and the CS during conditioning in Group Control

Cont-Pre x CS conditioning

The data above show the pupil responses in 1-s bins during the first 5-s of the CS (turquoise) and the pre-CS (gray) over the 10 conditioning trials (separated by gaps).   The CS came to elicit an increase in pupil diameter, and appeared to do so much faster than the LI group.  That the two groups differed in their response change over trials suggests that the pupil response is a conditioned response.

Extinction data Control x Latent inhibition groups, pre-CS data.

Ext controlx LI PRE

Data above show the pre-CS responses in 1-s bins for Group LI (gray) and the Control (turquoise) over the five extinction trials.  Doesn’t appear to be much going on there.

Extinction data Control x Latent inhibition groups, CS data.

Ext control x LI

The data above show the response to the CS during each 20-s presentation in extinction in 1-s bins for group Control (Turquoise) and the latent inhibition group (Yellow).  Not real sure that much is going on here, but it appears as though the Control group might be showing a timed pupil response.  It increases and then decreases as the CS progresses without the appearance of the US.  That effect is overall, perhaps, less evident in the LI group.   The overall response appears to be decreasing somewhat faster in the LI group, at least in places.

Problems / interpretation issues

These data are encouraging.  But, calling my tool an “analysis” tool is a bit misleading.  It simply allows me to rapidly group data and view the results so that I can determine what appears to be the optimum grouping and intervals for export and analysis for statistical significance.   While all the patterns evident are encouraging, at this point I have no idea as to the statistical significance of them.   I am optimistic.  The moment by moment variability is relatively stable—thus patterns of change over multiple intervals are likely to be detectable.

These data were not collected under optimum conditions.  It was a pilot study to sort out procedural details, and those details changed between some subjects in each group.  The participants knew it was a pilot study, and there were two researchers in the room with each subject.  We were not overly cautious about discussing procedural details while a participant was participating.   Control was loose.

The calibration of the eye tracker may not have been as good as it could have been for most subjects.  The calibration routine was run once from within the game, and then the game proceeded.  After a few subjects we built an assessment into the experiment where on the last trial a new sensor was lit and the subject was instructed to look directly at it for its duration.  Thus, the gaze data should be within that sensor on that trial and deviations represent either a subject who was unable to focus, or poor calibration of the machine.   About 25% of the subjects within the data above would fall into those categories.

I have modified the game so that after calibration, a fixation point is presented so that I can assess how good the calibration is, and repeat the calibration if it is bad.  At the end of the session, the same fixation point is presented.  Thus, I’ll have indexes of accuracy at the beginning and end of a session and can linearly adjust the resulting data over the session to account for any systematic drift in the calibration over the session.

In the gaze data presented in the top left of the analysis-tool screen above it appears as though the participants are looking at the CS and where the US is expected to appear.  That possibility is confounded by the weapon that is also in that corner.  That weapon is the one that is used to repel the spaceship.  When the participant begins charging that weapon, the round plasma-like window on the weapon animates.  Thus, they could be looing in the area where the US is expected, or they could simply be looking at the weapon as it charges.

I have modified the game so that the plasma window no longer animates in relation to the weapon charging, and modified the flight-path of the spaceship US so that it comes out of the middle –left of the screen.  Hopefully future participants gaze will be registered in that area.

Despite these technical issues, the data look amazing, and can only get better from here.

A final question regards the pupil data.   It is unclear why the pupil dilates over the CS over trials.   Is it a conditioned response—or an artifact of the behavioral responses.  The increases parallel the behavioral data. 

1: It could be that the pupil dilates because the participants are exerting themselves somewhat by pressing the appropriate key.

2: It could be that the pupil dilates because they expect the US, which elicits strong keypressing on its own.  That is, they could be expecting to respond, which leads to the pupil dilation.

3: They simply could be expecting the US.   

To pursue the pupil reactions I will have to separate the reaction of the pupil from the response.  After successful conditioning, I would need to have some test trials where the CS is presented, and the participant instructed to not respond.  Hopefully that will not affect the pupil response.    On other trials, a new sensor would be presented, and the subject instructed to respond.  Hopefully, that will not affect the pupil response.

Too much to do.. not enough people or machines!

This entry was posted in Eye Tracking, The Learning Game. Bookmark the permalink.

4 Responses to Eye Tracking–Pupil Data

  1. Pingback: Eye Tracker—Areas of interest, scanpaths. | Byron's blog.

  2. Pingback: Eye Tracker–Areas of interest, Hits. | Byron's blog.

  3. Pingback: Eye Tracker–Areas of interest, Hits. | Byron's blog.

  4. Pingback: Eye Tracker–Pupils revisited. | Byron's blog.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s