'Seeing Attention: A Reverse Correlation Paradigm' (AsPredicted #84040)
Author(s) This pre-registration is currently anonymous to enable blind peer-review. It has 2 authors.
Pre-registered on 2021/12/27 - 12:34 PM (PT)
1) Have any data been collected for this study already? No, no data have been collected for this study yet.
2) What's the main question being asked or hypothesis being tested in this study? In the current experiment, we will employ data-driven methods to model internal representations of other people's attention.
3) Describe the key dependent variable(s) specifying how they will be measured. Observers will view a series of images created by superimposing random noise patterns on a constant base face. The base face will be a grayscale average of all male faces in the Karolinska Face Database (Lundqvist et al., 1998), while the noise patterns will be generated according to standard parameters in similar work (Dotsch et al., 2008). On each trial, observers will view two faces side by side (created by adding and subtracting the identical noise pattern), and they will be asked to select which of the two people appears more attentive. We will then average all selected noise patterns and all non-selected noise patterns for each participant; the resulting "attentive" and "non-attentive" classification images (CIs) will constitute the values of our key dependent variables.
4) How many and which conditions will participants be assigned to? Each observer will complete 400 trials, randomly selected among 1200 unique trials and presented in a random order. On each trial, we will also randomly select which of the two faces (i.e. the one with added vs. subtracted noise) appears on the left (vs. the right).
5) Specify exactly which analyses you will conduct to examine the main question/hypothesis. We will conduct 8 paired-sample t-tests comparing attentive and non-attentive CIs for each participant on the following metrics:
1. The absolute deviation of gaze direction in radians, derived from the gaze direction angle as computed via OpenFace, and averaged across the x and y directions. We expect attentive CIs to have a lower absolute deviation in gaze direction.
2. The area of the pupils in pixels, i.e. the area of the elliptical region bounded by the pupil landmarks as computed via OpenFace (landmark indices: 21, 23, 25, 27 for the left pupil; 49, 51, 53, 55 for the right pupil), and averaged across the left and right pupils. We expect attentive CIs to have larger pupils.
3. The vertical position of the labial commissure as a percentage of the height of the mouth, i.e. the height of the region bounded by the mouth landmarks in 2D as computed via OpenFace (landmark indices: 48, 54 for the labial commissure; 50, 52, 57 for the mouth height), and averaged across the left and right commissure. We expect attentive CIs to have labial commissures closer to the top of the mouth.
4. The mean luminance of the nostril region, i.e. the rectangular region bounded by the nose landmarks in 2D as computed via OpenFace (landmark indices: 30, 31, 33, 35). We expect attentive CIs to have darker nostrils.
5. The absolute deviation of head rotation in radians, derived from the head pose rotation as computed via OpenFace, and averaged across the x and y axes. We expect attentive CIs to be no different than non-attentive CIs.
6. The area of the chin in pixels, i.e. the area of the elliptical region with the height determined by the chin landmarks in 2D as computed via OpenFace (landmark indices: 57, 8), and the width determined by the mouth landmarks in 2D as computed via OpenFace (landmark indices: 58, 56). We expect attentive CIs to be no different than non-attentive CIs.
7. The vertical position of the mouth as a percentage of the height of the lower half of the face, i.e. the height of the region bounded by the face landmarks in 2D as computed via OpenFace (landmark indices: 50, 52, 57 for the mouth; 33, 8 for the lower half of the face). We expect attentive CIs to be no different than non-attentive CIs.
8. The mean luminance of a rectangular region on the cheek of width, height, and vertical position equal to the nostril region (see #4), but horizontally centered at the halfway point of the cheek region bounded by the face landmarks in 2D as computed via OpenFace (landmark indices: 1, 31 for the left cheek; 35, 15 for the right cheek), and averaged across the left and right cheek. We expect attentive CIs to be no different than non-attentive CIs.
6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations. We will exclude observers who:
1. Reported having interrupted the survey.
2. Reported past participation in a similar study.
3. Reported not having paid attention (i.e. answered less than 50 on a 1-100 scale, in response to a specific question about how much they were paying attention).
4. Encountered any technical problems.
5. Failed to answer our questions sensibly (e.g. responding 'good' or 'nice' to a question asking what they thought the experiment was about).
6. Have a negative correlation between the averages of selected noise patterns in the first vs. second half of the experiment.
7) How many observations will be collected or what will determine sample size? No need to justify decision, but be precise about exactly how the number will be determined. We will collect data from 200 subjects (post-exclusions), and each will complete 400 trials.
8) Anything else you would like to pre-register? (e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?) As secondary exploratory analyses, we will conduct 4 paired-sample t-tests comparing attentive and non-attentive CIs for each participant on the intensity of the following action units, obtained via OpenFace:
1. The sum of raisers in the brow region (AU1, AU2) minus the brow lowerer (AU4): We expect attentive CIs to have higher brows.
2. The sum of raisers in the eye region (AU5, AU6, AU7): We expect attentive CIs to have higher eyelids.
3. The difference between the lip corner puller (AU12) and depressor (AU15): We expect attentive CIs to have more pulled lip corners.
4. The intensity of blinks (AU45): We expect attentive CIs to have fewer blinks.
(These analyses are reported as secondary given that AUs computed on individual images with one expression per person are less accurate.)
