Behavioral Results
The psychophysical measure used to evaluate the performance of the two groups of gamers was the hit rate, computed as the ratio of the number of correct responses to the triangle targets in the to-be-attended locations to the total number of triangles presented at those locations. The hit rate, both when the regular flicker was attended (e.g., ‘attend to green’) and when it was ignored, i.e., the random broadband flicker was attended (e.g., ‘attend to red’), was calculated for each trial and averaged across all twelve subjects in each experimental group.
We found that when the regular flicker was attended, the hit rate in both groups of gamers decreased as a function of temporal flicker frequency of the stimulus (Fig. 2a). This was expected since the search array was updated at the stimulus flicker rate and therefore, the number of new frames/s to be processed also increased with temporal frequency (Ding et al. 2006). Since the attended and unattended locations were placed on the same peripheral annulus, the attention filtering was expected to be partial, resulting in an effect of the temporal frequency of the unattended search array on the hit rate at the attended locations. As illustrated in Fig. 2b, the hit rate did decrease as a function of the temporal flicker frequency at the unattended locations. A multi-way ANOVA, with temporal frequency (3, 8.6 and 20 Hz), number of attended/unattended locations (1, 2 and 4) and gamer type (FPS and RPG) as fixed factors and individual subject as a factor with random effects, was carried out separately for the two attention conditions (flicker attended and rbbf attended). As evident in Fig. 2, there was a main effect of temporal frequency when the regular flicker was attended (F
2,44 = 86.56, p < 0.0001) and when the random broadband flicker was attended (F
2,44 = 154.85, p < 0.0001).
Figure 2b illustrates the dependence of the hit rate on the number of attended/unattended locations when the regular flicker was attended (F
2,44 = 30.16, p < 0.0001). Figure 2d illustrates the dependence of hit rate when the random broadband flicker was attended (F
2,44 = 9.84, p < 0.0001). In both cases, there was an effect of the number of locations on the performance, with the RPG gamers showing a monotonic decrease in hit rate with the number of locations and the FPS gamers performing similarly when two or four regions had to be monitored.
As expected, the FPS gamers, on an average, always performed better or the same as RPG gamers (Green and Bavelier 2007). This effect of gamer type was not significant when the regular flicker was attended (F
1,22 = 1.11, p > 0.05) nor when the random broadband flicker was attended (F
1,22 = 1.97, p > 0.05). However, from Fig. 2b, d, it is evident that the biggest difference in performance between the two groups of gamers occurred when four regions had to be simultaneously attended (and ignored). A t test between the hit rates for the two groups of gamers at 4 regions demonstrated significantly higher hit rate for FPS players compared to RPG players (p < 0.05 for both ‘Attend to Flicker’ and ‘Attend to RBBF’). This result suggests that the difference in performance, as measured by hit rate, increases with task difficulty. Here, the increase in task difficulty comes both from increased crowding of attended and unattended region and from increased attentional demands, as the number of regions increased. Fast action gamers have been previously reported to have both an increased attentional capacity (Green and Bavelier 2003) as well as higher spatial resolution of vision (Green and Bavelier 2007), which presumably drives the better performance of the FPS players when four regions had to be attended.
EEG Results
Signal to noise ratio of the SSVEP response to both attended and unattended regions of the display was the neurophysiological measure used to compare the neural strategies of the two groups of gamers. Specifically, we used the SNRa (SNR at the stimulus frequency when the flickering stimulus was attended) and the SNRu (SNR at the stimulus frequency when the stimulus was ignored while the broadband flicker was attended).
Dependence of Spatial Distribution of SNR on the Temporal Frequency of the Stimulus
The topographic plots in Fig. 3 illustrate the dependence of the SNR on the stimulus flicker frequency. As in previous studies (Srinivasan 1999; Srinivasan et al. 2006; Ding et al. 2006), the cortical areas that were entrained by a stimulus depended on the flicker frequency of the stimulus. For both categories of gamers, the 8.6 Hz flickering stimulus elicited the largest and most global responses, covering occipital, parietal and frontal cortex (Fig. 3, 8.6 Hz). The 3 Hz flicker also evoked large responses over parietal, occipital, and frontal cortex (Fig. 3, 3 Hz). The spatial distribution of the 3 Hz frontal cortex responses was similar to 8.6 Hz. The responses at 3 Hz over occipital and parietal cortex were distinct from the 8.6 Hz responses with foci away from the midline and somewhat right lateralized (not significant). In FPS players, these right occipital and parietal responses (averaged over stimulus configurations) were larger when the flicker was unattended. This was, however, not significant across subjects. Compared to the 8.6 and 3 Hz flicker, the 20 Hz flicker entrained a more local network in the visual cortex with responses mostly over occipital and parietal regions of the brain (Fig. 3, 20 Hz). The occipital/parietal response was focused on the midline, without the lateral responses seen at 8.6 and 3 Hz. The RPG players generally exhibited larger SNR values at each flicker frequency.
Prediction of Hit Rate from Signal to Noise Ratio (SNR)
In order to examine the difference in neural strategy between the FPS and RPG players, we modeled the relationship between the SSVEP data at each temporal frequency (3, 8.6, and 20 Hz) and performance (hit rate) using a PLS (Bro 1996). We found predictive relationships only at 3 and 8.6 Hz but not at 20 Hz.
The independent variable was the SNR (at each electrode), in response to the regular flicker: (1) when it was attended and (2) when it was ignored. In each case, the corresponding dependent variable was the hit rate at the attended location/s which was updated according to: (1) a regular square wave flicker (regular flicker attended) and (2) random broadband flicker (regular flicker ignored). Figure 4a shows the percentage of variance in hit rate accounted for by the SNR as per the PLS model. The percentage of variance explained in hit rate is equivalent to an r
2 value. Figure 4b. shows the regression coefficients associated with each EEG channel. Positive value of regression coefficients indicated positive correlation between SNR and hit rate, while a negative value of the regression coefficients indicated a negative correlation between SNR and hit rate. The relative strength of the regression coefficients at each channel indicated the relative weighting of each EEG channel in the prediction of hit rate, as well as the extent of correlation with hit rate. Model validation was carried out using a leave one out cross-validation.
3 Hz Response to the Unattended Flicker Predicts the Hit Rate at the Attended Location
We found that the hit rate variation in response to the attended location/s was predicted by the SNR variation of the 3 Hz SSVEP to the unattended location/s (Fig. 4a, cross-validated values) in both groups of gamers. However the direction of correlation between the SNR and the hit rate, as reflected by the regression coefficients, distinguished the two groups. In the RPG players, the SNR at the unattended location/s was negatively correlated with the hit rate (Fig. 4b, 3 Hz). This is expected if signal enhancement is the mechanism being employed, since attention to the unattended location presumably takes away from the attention to the attended location. This model explained 69.01 % of the variance in hit rate. However, if the role of a cortical network were in the monitoring of the unattended locations, possibly to actively suppress the information at those locations, we would expect an increase in the SNR to be positively correlated with hit rate. This is what we observed in the FPS players (Fig. 4b, 3 Hz) with 84.47 % of the variance being explained by the model (Fig. 4a). In both groups of gamers, the responses over the parietal and occipital channels were most predictive of the hit rate, with the right parietal electrodes contributing the most to predicting hit rate in the FPS gamers.
8.6 Hz Response to the Attended Flicker Predicts the Hit Rate at the Attended Location
In the 8.6 Hz case, the variance explained in the hit rate by the leave one out cross-validated model indicated that the SNR of the SSVEP to the flicker at the attended location/s is predictive of the hit rate (Fig. 4a, cross-validated values). In both groups of gamers, SNR was positively correlated with hit rate (Fig. 4b, 8.6 Hz), with the SNR over left parietal electrodes showing the highest correlation. In the RPG players this model explained 70.28 % of the variance in hit rate, while in the FPS players the model explained 66.28 % of the variance. This positive correlation between the responses to the attended stimuli and the hit rate implies that information is selected from the to-be-attended locations by enhancing the inputs from these locations in both groups of gamers.
Dependence of SNR on the Number of Attended/Unattended Locations
We wanted to find how the SSVEP responses of the two groups of gamers differed as a function of the number of regions to be attended. Therefore, for each temporal frequency, we identified the electrodes that showed SNR above a certain threshold (SNR > 2 for 3 and 20 Hz; SNR > 4 for 8.6 Hz) in either gamer group for at least one of the attention conditions. These electrodes are displayed in Fig. 3 as blue circles on the topographical maps. The trends exhibited by subsets of these electrodes (frontal, occipital/parietal), as a function of number of regions, are displayed in Figs. 5, 6 and 7. On each plot, the fraction of electrodes, in each group, that showed significant monotonic trend in the illustrated direction, is displayed in brackets.
3 Hz Frontal Responses are Larger When Ignored Only in First Person Shooter (FPS) Players
In the 3 Hz flicker conditions, SNR at the frontal electrodes generally increased with increasing number of locations to be monitored, for both attention conditions and both gamer groups (Fig. 5a, 3 Hz). In both gamer groups, SNRu (response to the unattended flicker) increased when the number of regions to be attended (and ignored) increased from 1 to 2 or 4 locations. In the RPG players, the SNRa (response to the attended flicker) also followed a similar trend, while in the FPS players the SNRa continued to increase even when 4 locations had to be monitored. The fraction of channels that exhibited this significant monotonic trend is displayed in brackets next to the trend lines. The main differences between the two groups of gamers were in the relative strengths of the SNRa and SNRu. In RPG players, the SNRa was always larger than the SNRu. In the FPS players, the SNRu was larger than the SNRa when more than one location had to be attended/ignored. The higher SNRu (averaged over significant monotonic electrodes in FPS players), relative to SNRa approached but did not reach significance (p = 0.0576). This increased response to the unattended locations exhibited by the FPS players again points to an active mechanism of suppression of unattended information, possibly originating in the frontal cortex. That an increased response to the unattended locations was observed only when more than one location had to be attended and ignored further supports our claim of active suppression because subjects are more likely to benefit from suppression in conditions where attended and ignored regions are interleaved.
In RPG players, a subset of left and right occipital-parietal electrodes also showed significant increasing SNRu with number of regions (Fig. 5b, c). In FPS players, on the other hand, the right occipital-parietal electrodes did not show a monotonic trend in SNRu (Fig. 5b), though a small subset of left occipital-parietal electrodes showed a significant monotonic increase in SNRu (Fig. 5c). Also noteworthy, is the fact that in the RPG players, the responses to the attended flicker were always higher than the responses to the unattended flicker even over left and right occipital-parietal cortices. The FPS players, on the other hand, showed lesser/no differences between the responses to the attended and unattended flickers over occipital-parietal cortices compared to those over the frontal cortex.
Direction of the Variation in the 8.6 Hz Frontal Responses to the Attended Flickering Stimulus Depends on the Gamer Type
The modulation of the 8.6 Hz frontal responses with the number of regions to be attended and ignored, distinguished the two groups of gamers. When the flicker was attended, the SSVEP amplitude produced by the flickering stimulus decreased as a function of the number of locations in the FPS gamers, whereas in the RPG gamers these responses increased (Fig. 6a, 8.6 Hz). As described above, because an increase in the strength of the SSVEP responses with task difficulty suggests a compensatory mechanism, it is possible that in RPG players, the 8.6 Hz responses were responsible for enhancing the information at the attended locations. In FPS players, the decreased responses indicate the effect of the increased attentional demands on the neural resources, with no evidence of compensatory gain. The right and left occipital-parietal electrodes in FPS gamers also showed significant decreasing trend in SNRa similar to the frontal electrodes (Fig. 6b, c). On the other hand these electrodes in RPG gamers showed a non significant but opposite trend to the frontal electrodes (Fig. 6b, c).
The Variation of Local 20 Hz Responses with Number of Locations is Similar in both Gamer Groups
In both groups of gamers, SSVEP responses to the attended flickering stimulus decreased with increasing number of regions to be attended and ignored (Fig. 7, 20 Hz). These local responses, over occipital and parietal cortices, reflect the low pass characteristics of attention (Gobell et al. 2004). In the RPG gamers, SNR increased slightly when two regions were attended relative to when only one region was attended and then decreased when four regions were simultaneously attended. In both groups of gamers, there is no difference between the responses to the attended or ignored flickering stimulus when 4 regions had to simultaneously monitored, indicating that there was no effect of attending to the flicker in that condition. This lack of attention modulation is consistent with the poor performance (low hit rates) in this condition (see Fig. 2a, c).
The monotonically modulated frontal responses in the 3 and 8.6 Hz cases are indicative of the compensatory mechanisms that follow the increasing task demands. The occipital/parietal responses that track hit rate (as observed with the PLS modeling) presumably indicate the success of the mechanisms coupled to the processes in the frontal cortex.