As with Pilot Experiment 1, we carried out Pilot Experiment 2 in order to control for legibility issues after changing the letter stimuli to have the L distractors appear more T-like. Again, the blurred and clear letters had similar accuracy and reaction times, thus, similarly to Experiment 1, allowing Experiment 2’s results to be interpreted as changes in attention, not difficulty identifying the blurred letters.
Participants
There were 32 participants from Kansas State University’s (KSU) undergraduate research pool (24 females, mean age = 18.5). Participants’ vision was tested and was 20/30 or better. All participants were naïve at the beginning of the pilot. All participants gave their informed consent and participated for class credit.
Apparatus and Stimuli
All apparatus was the same as in the Experiment 1 pilot study. As shown in Figure 10, the stimuli were changed to make the L distractors appear more T-like (Jiang & Chun, 2001). The clear L and T images’ vertical and horizontal lines were 44 pixels (1.65° of visual angle) in length. Both lines have a width of 4 pixels (0.15° of visual angle). The images were blurred using the same methods as explained in the Experiment 1 pilot study. There are two set sizes (4 & 8) for Experiment 2, and the locations of the letters for both set sizes did not overlap, resulting in 12 potential locations along a centered invisible circle set to a radius of 9° degrees of visual angle. When the set size was four, the images appeared from the top of the circle at 22.5°, 112.5°, 202.5°, and 292.5°. When the set size was eight, the images appeared from the top of the circle at 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°.
Figure 10. Images of the L and T blurred/clear letter rotations. Row 1 clear L rotated (0, 90, 180, and 270). Row 2 blurred L with same rotations. Row 3 clear T with same rotations as the Ls. Row 4 displays the blurred Ts with the same rotations as above.
Design
The experimental design was identical to Experiment 1’s pilot study. There was a 2 (Letter: L vs. T) x 2 (Clarity: Clear vs. Blurred) within-subjects design. The counterbalancing of Letter, Clarity, and the 12 potential locations a letter could be presented at, resulted in 48 trials per block, and six blocks, resulting in 288 total trials. All 288 trials were then randomized across each participant’s experimental session. The dependent variables were accuracy for the
identification task and RT.
Procedure
Participants first read and gave their informed consent, then their visual acuity was tested using the Freiburg Visual Acuity and Contrast Test (FrACT) (Bach, 1996; Bach, 2007).
Participants then read instructions to the experiment. This was then followed by another
instruction screen showing the participants examples of what the T targets looked like presented clearly and blurred as well as the L distractors. The trials began by having the participants observe a fixation cross at the center of the screen. Once their eyes were focused on the middle
of the fixation cross they hit the next button on the response pad. The fixation cross was then removed from the screen and a blank neutral gray screen was presented for 506 msec followed by the presentation of a single L or T letter which remained present until a response was made. There was no response screen that appeared. Once the participant had pushed the left response button to indicate they saw an L, or the right response button to indicate they saw a T, there was a blank neutral gray screen presented for 1000 msec, before the next trial. Participants were allowed to quit or take a break at any time during the experiment; also every 72 trials the participants received a built in mandatory break with a total of 288 trials. After finishing the experiment, participants read through a debriefing form and were then thanked for their participation.
Pilot Experiment 2 Results
The analyses were conducted on 32 participants, however prior to the analyses, data from nine other participants were lost due to computer error in recording the participants’ results. Prior to cleaning any of the data, the overall accuracy across all trials was 95%. The data was then cleaned by first removing all reaction times < 150 msec or > 10 seconds. This resulted in 13 trials being removed from 9216 trials (0.1% of all trials).
The analyses of accuracy were conducted with the R statistical software (version x64 3.1.1) to run a multilevel logistic regression for binary data. Prior to analyzing the models, the categorical variables Letter and Clarity were effect coded such that Letter (L = +1, T = -1) and Clarity (Blurred = +1, Clear = -1). Table 9 displays the parameter estimates for the fixed effects and random effects variance from the best model. From the analyses, the best model was
Accuracy ~ Letter + Clarity + Letter x Clarity + Log10(Trial) + (1|Participant) (BIC = 3370.5)
compared to the second best model which included the random effect of (Letter|Participant) (BIC = 3377.6) (See Appendix D for all models with BIC values).
Table 9
Parameter Estimates for Accuracy ~ Letter + Clarity + Letter x Clarity + Log10(Trial) +
(1|Participant) Model
Fixed Effects Estimates Std Error z-value p(z)
Random Effects Variance Intercept 3.521 0.317 11.11 <.001 1.201 Letter(L) -0.048 0.049 -0.98 0.325 Clarity(Blurred) 0.051 0.049 1.05 0.294 Log10(Trial) -0.045 0.118 -0.38 0.705 Letter(L) x Clarity(Blurred) 0.066 0.049 1.36 0.174
Note. Model was performed using effect coding [(Letter: L = +1, T = -1) (Clarity: Blurred = +1, Clear = -1)].
Figure 11 displays the mean transformed accuracy values back to proportion from logit space for the clear and blurred Ls and Ts at the end of the experiment. There were no significant main effects or interactions for Letter or Clarity. Most importantly, this suggests that level of Clarity had no influence on accuracy in identifying the letters.
Figure 11. The mean accuracy for each letter presented clearly or blurred. Error bars = 95% CI after being transformed back from logit, therefore the error bars are asymmetrical, having slightly longer negative error bars.
50 55 60 65 70 75 80 85 90 95 100 L T Ac cura cy (Propor ti on C orr ec t)
Letter x Clarity
Blurred ClearPrior to analyzing the data for reaction times there was further cleaning of the data such that all of the incorrect responses were removed. From the 9203 total trials, 461 trials were removed due to being incorrect responses (5% of trials).
The analyses were conducted using a linear multilevel model with effect coding in JMP Pro 12. Table 10 displays the parameter estimates for the model. The model analyzed was the same as Experiment 1 pilot study, Log10(RT) ~ Letter + Clarity + Letter x Clarity + Log10(Trial)
+ (Letter + Clarity + Letter x Clarity + Log10(Trial)|Participant) with R2 = .23, adjusted R2 = .22,
RMSE = 0.12. Table 10
Parameter Estimates from Log10(RT) ~ Letter + Clarity + Letter x Clarity + Log10(Trial) +
(Letter + Clarity + Letter x Clarity + Log10(Trial)|Participant) Model
Fixed Effects Estimates Std Error DFDen t Ratio p-value
Intercept 2.991 0.020 46.31 147.53 <.001
Letter(L) 0.003 0.003 31.08 1.05 0.301
Clarity(Blurred) -0.001 0.002 29.04 -0.87 0.391
Letter(L) x Clarity(Blurred) -0.005 0.002 31.25 -3.24 0.003
Log10(Trial) -0.062 0.009 30.66 -7.04 <.001
Note. Model was performed using effect coding [(Letter: L = +1, T = -1)(Clarity: Blurred = +1, Clear = -1)]. DFDen = degrees of freedom used in the denominator.
Table 11 presents the Log10(RT) M with within-subject SD, and the RT* GM with within-
subject GSD for the clear and blurred Ls and Ts in msec. Figure 12 displays the Log10(RT) M
and SEM for Clarity x Letter in msec with RT* GM on secondary y-axis. As in the Experiment 1 pilot study, the linear multilevel model, Log10(RT) ~ Letter + Clarity + Letter x Clarity +
Log10(Trial) + (Letter + Clarity + Letter x Clarity + Log10(Trial)|Participant) was analyzed for
significant main effects and an interaction. There was a significant main effect for Log10(Trial),
F(1, 30.7) = 49.61, p <.001, indicating that participants responded faster as they progressed through the experiment. There was no significant main effect for Letter, F(1, 31.1) = 1.11, p = .301, with Ls identified at similar rates to Ts. There was no significant main effect for Clarity, F(1, 29.0) = 0.76, p = .391, indicating blurred and clear letters are identified at similar rates. There was a significant interaction effect between Letter x Clarity, F(1, 31.25) = 10.47, p =.003, therefore one letter was identified faster or slower than at least one of the letters at a specific Clarity level.
Table 11
Letter x Clarity: Log10(RT) M with SD and RT* GM with GSDs
Letter Clarity Log10(RT) Log10(RT) RT* RT* RT*
M SD GM -1 GSD +1 GSD
L Blurred 2.863 0.116 729 558 952
L Clear 2.875 0.113 749 577 972
T Blurred 2.866 0.124 734 551 977
T Clear 2.857 0.119 720 548 947
Note. RT* = Untransformed Reaction Time. M = Marginal Means. GM = Geometric Mean. SD = Within-subject Standard Deviation. GSD = Geometric within-subject Standard Deviation. RT* has -1 and +1 GSD because of the positive skew of the RT data when untransformed.
Two planned contrasts compared the RTs for the levels of Clarity for each Letter. The first contrast found a significant difference between the RTs for clear and blurred Ls, t(31.25) = 2.92, p = .005, with the blurred Ls slightly faster. The second contrast found no significant difference in RT between the clear and blurred Ts, t(31.25) = 1.70, p = .094. These results showed there was no significant accuracy difference between the Ls and Ts. Clear and blurred Ls did significantly differ on Log10(RT) with the blurred Ls being responded to faster than the
clear Ls. Importantly, the blurred and clear Ts did not significantly differ (14 msec). This is important because in Experiment 2, only the Ts were used as the target for each search. Also, while the Ls did significantly differ, they differed by 20 msec. This difference should have a very minimal effect on the results for Experiment 2 with Ls being distractors. Therefore, legibility was controlled for between the clear and blurred letters. Results from Experiment 2 should not be based on whether or not a blurred target was harder to identify than a clear target.
Figure 12. The Log10(RT) marginal means for each letter presented clearly or blurred with +/- 1
Chapter 5 - Experiment 2