We produced a computer game created using Scratch software (2009, version 1.4; http://scratch.mit.edu/
), in which human participants attempted to find and capture targets against a background. The general design of the study followed a recent experiment [11
], except that we used a touchscreen rather than asking participants to capture targets with a mouse and cursor. This change, with participants directly touching the screen to capture prey, made the task more realistic. In experiments 1 and 3, a single 'prey' item (target) moved at a constant speed against the background: 25 cm/s (approximately 31.13° visual angle per second). In addition to average speed, the distance of target displacement between consecutive refreshes of the screen as it moves may also be important in influencing the outcomes of motion-detection mechanisms, as outlined in previous work with humans [40
]. Our display refreshed at 75 Hz, which would equate to a frame-to-frame displacement of 0.4° visual angle (see below). However, we calculated that the software used refreshed at approximately 40 Hz, which would result in a frame-by-frame displacement of about 0.77 degrees. This relatively large value may make our findings regarding dazzle effects conservative, as past work indicates that human motion detection may work most effectively at short displacements. Therefore, it is possible that smaller displacements (produced by faster screen refresh rates) would result in greater motion-dazzle effects than those reported here.
In the game, the prey changed direction unpredictably between 1° and 3° clockwise during movement, and bounced back from the edges of the background with the addition of a 45° anticlockwise turn (that is, the 45° change was in addition to the normal effect of bouncing back off the screen edge, to make the target trajectory less predictable). After successful capture the prey disappeared, and after a delay of 0.5 seconds, reappeared in a random position on the screen. The patterns on the prey rotated as the main body of the target rotated. For example, the striped patterns were always perpendicular to the direction of movement, irrespective of which direction the target was actually travelling.
In experiments 2 and 4 prey were stationary, and the task was to find and then capture a prey item. Once a target was caught, a uniform gray background appeared for 0.5 seconds, before the experimental background reappeared with a new prey target in a random location. This was essential because if the prey item simply reappeared in a new location after capture, its reappearance would reveal its new location. For all experiments, within a single trial, participants had to capture as many prey targets of the same type as possible within one minute. This was repeated for each type of prey target (treatment). Treatment order was balanced in all experiments so that each treatment appeared an equal number of times in each order. We recorded the number of prey items captured, where a subject managed to touch the point on the screen at which the moving prey item was located at that time. We also recorded the number of missed attempts. Missed attempts in the moving-prey trials were the times when subjects saw the target but misdirected their attacks and touched the background part of the screen instead of the prey item. In the stationary prey trials, the subjects incorrectly touched the screen where they thought the target was located, but it was in fact elsewhere (false detection). Note that we did not expect that misses and hits should be inversely related, because several factors can affect this relationship; for example, a subject may concentrate more on capturing a target that seems to be moving faster, and make fewer capture attempts (and thus also fewer misses).
Prey targets and backgrounds were achromatic (shades of gray) created using Photoshop Elements (version 7.0; Adobe Systems Inc., East Oldsmar, FL, USA) as high-resolution, low-compression JPEG files. In all experiments, targets were 2 cm wide and 0.9 cm tall (3.11° and 1.12° visual angle subtended on the viewer's eye). Targets were presented against patterned artificial backgrounds comprising black, white and gray markings on a uniform gray. Several versions of the backgrounds were used in each experiment (six in experiments 1 and 2, and eight in experiments 3 and 4; for all backgrounds used see Additional file 1
) to remove any potential interactions that could occur between a given treatment and a specific background arrangement. All experiments were conducted on the same 15 inch (38 cm) touchscreen monitor (Elo 1515L; Tyco Electronics, Shanghai, China) with a refresh rate of 75 Hz (higher than the 60 Hz from our previous study [11
]). The flicker of the striped targets was 62.5 Hz (based on calculating the time taken for one complete cycle of white and black stripes), and although relatively high, is still lower than the refresh rate of the display. We calibrated the visual contrast of the different shades of gray displayed on the prey and background in terms of luminance (cd m-2
), using a luminance meter (Minolta LS-110; Osaka, Japan), as described previously [11
]. We determined the background value that would correspond to an intermediate level of gray between black and white, and for experiments 3 and 4, to several intermediate shades of gray. The luminance values were as follows (in cd m-2
): white = 196, black = 8, gray = 40, light gray = 90, and dark gray = 18. The contrast values, based on Michelson contrast, were 0.66 for white/black against intermediate gray, and 0.38 for light gray/dark gray versus intermediate gray. All treatment types, except the white target, had the same average luminance as the background, and so the key difference between them was in their patterning and contrast. Participants were positioned in front of the touchscreen, approximately 46 cm away, under ambient light conditions (standard fluorescent office lights) kept approximately constant.
In all experiments, participants (166 in total: 60 in experiment 1, 18 in experiment 2, 64 in experiment 3, and 24 in experiment 4) were volunteers naïve to the experimental aims, and were predominantly undergraduate students with normal vision or corrected-to-normal vision. We gave participants only the information needed to undertake the trials. No subject participated more often than once across all experiments. All participants carried out a one-minute practice trial before the main experiment, in which they had to capture a uniform black prey item against a white background. Neither this background or prey target was used in any of the main experiments.
The statistical approach has been described previously [11
]. Where possible, we analyzed the results with general linear models (GLMs), with the factors of prey type and order of presentation, with the subject as a random factor. When the data violated the assumptions of a GLM and could not be successfully transformed, we analyzed the results with a non-parametric Friedman test. For most experiments we used planned post hoc
] by rerunning the main test with the factor prey type replaced with each comparison in turn, using no more comparisons than spare degrees of freedom. This is a much more powerful approach than undertaking multiple unplanned comparisons, and best reflects our specific hypotheses [11
]. In experiment 2, for the data on number of misses, planned comparisons were not intuitive, and so we reran the main test with all pairwise comparisons, and used a sequential Bonferroni procedure to adjust critical P
-value thresholds to control for multiple testing. We did not always use the same comparisons for the data on hits and misses because these data can reveal different aspects of the subject's strategy and success, and need not be inversely related (see above and Discussion).
Experiment 1: The role of pattern arrangement in motion dazzle
We used six prey types: a uniform gray (G) target matching the average background luminance; a conspicuous uniform white (W) target; a target with perpendicular black (3 mm wide) and white (2 mm wide) alternating stripes (S); a target with pairs of 3 mm black and 1.2 mm white stripes separated by interval gray stripes of 4 mm (IS); a camouflaged background matching target (B), consisting of a random sample of the background with the stipulation that no markings touched the target edge; and a camouflaged disruptive prey type (D), comprising samples of the background with at least some patterns located on the target edge (Figure ).
The results were analyzed with a GLM. We included the uniform white prey item in the overall treatment test because there was no prior reason to believe that this should be treated differently from the other conspicuously-marked treatments. In any case, we focused specifically on planned comparisons between individual prey types rather than on omnibus comparisons across all treatments in our interpretation of results. For the data on misses, we had to transform the data to the power of 0.06, calculated using a Box-Cox procedure (we found a Box-Cox transformation produced the most effective transformation in this case). For our planned comparisons, we compared (i) white versus the aggregate of all other prey types, (ii) gray versus the aggregate of all patterned prey types, (iii) dazzle prey (stripes and interval stripes) versus the camouflage prey (background matching and disruptive coloration), (iv) striped versus interval stripes, and (v) background matching versus disruptive coloration.
Experiment 2: The role of pattern arrangement in preventing detection and capture when stationary
Experiment 2 comprised the same treatments as in experiment 1, but the targets were stationary. Our aim was to test whether, when stationary, the camouflaged-prey targets would be harder to detect (and therefore capture) than either the dazzle or uniform targets.
The main results and post hoc comparisons for both captures and misses were analyzed using a Friedman test. For the capture data, we predicted that the conspicuous white target would be the easiest to capture, and the background matching and disruptive targets most difficult. In addition, although their average luminance matched the background, the arrangement of the striped patterns might render the striped and interval-striped prey more conspicuous than the uniform gray prey. Alternatively, the stripes may function in disruptive camouflage. Based on these predictions, we conducted a series of planned stepwise comparisons, comparing treatments with successively lower capture levels in turn. This resulted in comparisons of (i) W versus S, (ii) S versus IS, (iii) IS versus G, (iv) G versus B, and (v) B versus D. For the misses it was difficult to make clear a priori predictions. Therefore, we compared all treatments with each other, and then ranked contrasts in terms of P value (lowest to highest). We then selected the comparisons with the five smallest P values, and used critical P-value adjustment based on a sequential Bonferroni correction.
Experiment 3: The role of pattern contrast in motion dazzle
In experiment 3, there were eight treatments: a uniform white (W) target; a uniform gray (G) target; background matching targets with either low-contrast (LB) light and dark gray spots, or high-contrast (HB) with black and white spots; disruptive targets of either high (HD) or low (LD) contrast; and striped prey with markings of either high (HS) or low (LS) contrast. In ensuring both types of patterned prey had the same average luminance, we slightly modified the width of the stripes in the low-contrast striped prey (2 mm light-gray and 2 mm dark gray stripes). The backgrounds comprised an average intermediate gray, plus white, black, light-gray, and dark-gray spots in approximately equal proportion.
Results for both captures and misses were analyzed with GLMs. Planned comparisons were: (i) white versus the aggregate of all prey types, (ii) gray versus the aggregate of all patterned prey types, (iii) the aggregate of the low-contrast patterned prey versus the aggregate of the high-contrast prey, (iv) low-contrast stripes versus high-contrast stripes, (v) low-contrast camouflaged disruptive and background matching versus high-contrast camouflaged prey, (vi) striped versus camouflage prey, and (vii) background matching versus disruptive.
Experiment 4: The role of contrast in concealment
The design of experiment 4 and treatments followed that of experiment 3, but with stationary targets. Results were analyzed with GLMs, with data for the number of misses being square root transformed. There was a single prominent outlier in this data even after transformation. However, rerunning the analysis with this outlier excluded did not change the results (the effect of prey type became more significant without the outlier included). We conducted the following planned comparisons: (i) white versus the striped prey, (ii) stripes versus the gray target, (iii) gray versus the camouflage targets, (iv) high versus low-contrast prey, (v) disruptive versus background matching, (vi) high versus low-contrast stripes, and (vii) high versus low-contrast disruptive prey.