A crooked picture
Most of us notice right away that a picture on the wall hangs a degree or two crooked. But if the same picture were rotated by 45 degrees, a difference of a degree or two would be much harder to see. Our visual system does not see all directions equally sharply: horizontal and vertical are special for us.
In this experiment a line appears for half a second and vanishes. Then you turn the orange line in the middle to the angle you saw. In four of the eight rounds the line is exactly horizontal or exactly vertical; in four it is oblique. At the end you see your average errors for the two types side by side: the difference is your personal version of the oblique effect that vision scientists have studied for half a century.
The oblique effect
In a review published in Psychological Bulletin in 1972, Stuart Appelle brought together the studies on orientation perception in humans and animals carried out until then. The same pattern appeared in many different tasks: stimuli at horizontal and vertical orientations were perceived and discriminated better than those at oblique orientations. Appelle named these findings the oblique effect.
Later studies examined the details of the effect. In 1984 Orban, Vandenbussche and Vogels measured orientation discrimination with single long lines presented one after the other: sensitivity was highest within a narrow range around the principal meridians, and both sensitivity and the difference between orientations increased as the lines got longer. The lines in this experiment are long too, so we expect to see the effect.
A horizontal and vertical world
Where does this advantage come from? One explanation lies around us: buildings, doors, windows, tree trunks and the horizon are horizontal or vertical. In 1998 Coppola, Purves, McCoy and Purves scanned a large library of photographs of indoor, outdoor and untouched natural scenes with orientation-sensitive filters and found that horizontal and vertical edges predominated in every type of scene.
The brain seems to reflect this imbalance too. In 2000 Christopher Furmanski and Stephen Engel examined human primary visual cortex (V1) with fMRI: V1's response to horizontal and vertical stimuli was larger than its response to oblique stimuli at 45 and 135 degrees, and this difference matched the participants' perceptual performance.
The brain sees with expectations
In 2011 Ahna Girshick, Michael Landy and Eero Simoncelli looked at this through the lens of statistical inference. They asked five observers to compare the average orientation of two arrays of small oriented patterns, some of which were noisy. The observers' orientation judgments were pulled toward horizontal and vertical as uncertainty increased, with relative biases between the two noise levels of up to 12 degrees. When they estimated the observers' prior expectations about orientation, they found that these matched the distribution of orientations in photographs.
But not every finding points the same way. In 2010 Vincent de Gardelle, Sid Kouider and Jérôme Sackur asked participants to reproduce the orientation of briefly shown oriented patterns and found that patterns near the principal orientations were perceived as more oblique than they were, pushed away from the principal orientation. In 2015 Xue-Xin Wei and Alan Stocker reconciled these two observations with a Bayesian observer model based on efficient coding: the same prior knowledge can lead to attraction or repulsion depending on the source of the uncertainty.
Oblique lines in memory
In this experiment you set the line not while you see it but after a half-second display and a brief mask; so short-term memory is involved as well as perception. In 2017 Pratte, Park, Rademaker and Tong showed participants several oriented patterns, asked them to report the orientation of one from memory, and found that errors were clearly larger for oblique orientations.
For oblique lines we look at two things in this experiment: the size of the error and its direction. A positive drift means your answer was pulled toward the nearest horizontal or vertical; a negative drift means it was pushed away. Four oblique lines are too few to show a personal direction conclusively; but which direction dominates in the crowd could be a small contribution to this debate in the literature.
Your score and your screen
Each round's score is calculated from the angle between the true orientation and the one you set: 10 / (1 + (error / 6°)^1.6). A 2-degree error earns about 8.5 points, 4 degrees about 6.6, 6 degrees 5 and 12 degrees about 2.5. The eight rounds add up to at most 80 points. Because both ends of a line are the same, the largest possible error is 90 degrees.
The edges of the screen are a ready reference for horizontal and vertical lines, so we recommend holding your phone upright and level. The dial has no ticks; they appear only after your answer, on the comparison screen. On touchscreens you can turn the line more finely by moving your finger far from the centre.