Reaction times from incorrect trials were excluded from analyses. Following this, on a subject-by-subject basis, a mean RT was calculated for each of the twelve conditions (3 flanker types by 4 cue types). Data were collapsed across target direction (left/right) and target location (above/below). On the basis of unusually slow and error-prone responses (more than two standard deviations beyond the mean for their age group and gaming status), data from six NVGPs were excluded from analysis completely. For the remaining subjects, if the response time for a trial was greater then two standard deviations from the mean for its condition, then it was excluded as an outlier; neither RT nor accuracy data were analyzed for these outlier trials. Median RTs were then calculated for each condition for each subject and submitted for further analysis.
Gender Differences in RT
Males are more likely than girls to play action video games, and this is reflected in an asymmetric distribution of males and females across the NVGP and VGP categories (). Mezzacappa (2004)
reported a small gender effect on RTs in the ANT in a study of 118 young girls and 131 young boys. An initial analysis of the overall RTs for males and females in our NVGP group revealed no significant difference in RTs (F(1, 68)=1.45, p > .05, eta2p
=.02), suggesting that gender had no measurable impact on RT performance in the age range tested.
Controlling for Baseline Differences in RT
A four-way mixed ANOVA was performed on the median RT data with flanker type (incongruent, congruent) and cue type (absent, center, double, spatial) as within subjects factors, and age group (7–10 yrs, 11–13 yrs, 14–17 yrs, 18–22 yrs) and video game playing (NVGP, VGP) as between subjects factors. This analysis revealed significant main effects of age group (F (3, 117)=50.07, p<.001, eta2p=.56) and video game playing (F (1, 117)=8.68, p=.004, eta2p=.07), suggesting baseline differences in RT as a function of both age and gaming experience.
Before proceeding with any further analyses, these main effects of age group and video game playing were addressed. The median RTs for older subjects were faster than those for younger subjects (M7–10YRS
= 678msec, M11–13YRS
= 554msec, M14–17YRS
= 496msec, M18–22YRS
= 467msec). In addition, VGPs (MVGP
= 525 msec) had faster median RTs than NVGPs (MNVGP
= 597 msec). As outlined previously, these baseline differences are of concern when interpreting interactions (see Faust et al., 1999
; Madden et al., 1992
for more discussion). The next stage of the analysis sought to address these baseline differences before reanalyzing the data.
In order to address baseline differences as a function of age group, the average median RT (collapsed across all target-only conditions, i.e. using only those trials where no flankers accompanied the target) was computed for each NVGP subject, and this was plotted against their age in months. Following Cerella and Hale (1994)
, these data were fitted using an exponential decay function (see Equation 1
Figure 3 (A) Using the function outlined by Cerella and Hale (1994), RT was fitted as a function of NVGP subject age using an exponential decay function. (B) After controlling for age differences in RT, normalized RTs from the flanker absent conditions were used (more ...)
A goodness-of-fit metric for the non-linear function, analogous to R2
, was computed using the method provided by Haessel (1978)
. This revealed a good fit to the data: Cos2
= 0.646. On the basis of this function, a predicted RT score was computed for all NVGP and VGP subjects and used to normalize their median RTs for each condition. For example, if a subject had an age-predicted RT of 450msec and their performance within a condition was 400msec, then their transformed RT would be 400/450 or 0.89. This age-normalized RT
) was used for all further analyses.
To control for RT differences resulting from video game experience – a categorical variable – another procedure was employed; the RTage
for each of the four target-only conditions were computed for the NVGP and VGP groups. These were plotted against each other, and a linear fit obtained (see Equation 2
This linear function was used to transform the RTage for NVGPs in each of the other eight experimental conditions formed by crossing flanker type (incongruent, congruent) with cue type (absent, center, double, spatial). The resulting gamer-transformed age-normalized RTs (henceforth, transformed normalized RTs – RTnormed) represent the extent to which RTs deviate from what is expected given the age and video gaming experience of individual subjects and thus provide a measure of the effects of flanker congruency and cue type that is not biased by baseline differences in speed of response.
Alerting, Orienting and Flanker Compatibility Effects
In line with previous studies using the ANT, we calculated ‘attentional network’ scores to reflect the effects of alerting, orienting and flanker compatibility. These were computed using these RTnormed values, and entered into two-way ANOVAs with age group (7–10 yrs, 11–13 yrs, 14–17 yrs, 18–22 yrs) and video game playing (NVGP, VGP) as between subjects factors.
Alerting effects – measuring the efficiency with which a temporal cue enhances processing of the target – were computed by subtracting RTnormed values for the double cue conditions from those for the no cue conditions for each subject. The main effect of age group was statistically significant (F (3, 117)=2.68, p=.05, eta2p=.06) with younger children exhibiting larger alerting effects than older children and adults (M7–10YRS=0.076, M11–13YRS=0.068, M14–17YRS=0.044, M18–22YRS=0.053). A priori contrasts revealed significant differences between the alerting scores of 7–10 year olds and 11–22 year olds (p=.027). The main effect of video gaming playing (eta2p=.02) and the interaction between age group and video game playing (eta2p=.02) did not approach statistical significance (see ).
Figure 4 A. Alerting effects were computed by contrasting the no cue and double cue conditions. A significant effect of age was observed, with 7–10 year olds having larger alerting effects than older children and adults. These larger alerting effects possibly (more ...)
Orienting scores – measuring the efficiency with which a valid spatial cue enhances processing of the target – were computed by subtracting RTnormed values for the spatial cue conditions from those for the center cue conditions for each subject. The ANOVA revealed no significant main effect of age group (F (3, 177)=0.17, p=.914, eta2p<.01) nor a significant interaction between age group and video game playing (eta2p=.03). However, the analysis revealed a significant main effect of video game playing on orienting effects (F (1, 117)=10.20, p=.002, eta2p=.08), with VGPs (MVGP=0.060) exhibiting larger orienting effects than NVGPs (MNVGP=0.038; see ). This effect will be returned to in the ANOVA analysis reported below.
Finally, flanker compatibility effects – measuring the extent to which flankers interfere with processing of the target – were computed by subtracting RTnormed values for the congruent flanker conditions from those for the incongruent flanker conditions for each subject. The ANOVA revealed no significant age group effect (F (3, 117)=1.08, p=.361, eta2p=.03) and no interaction between age group and video game playing (eta2p=.02). There was, however, a significant main effect of video game playing (F(1, 117)=19.71, p<.001, eta2p=.14), with VGPs (MVGP=0.103) having larger flanker compatibility effects, or in other words experiencing more interference from flankers, than NVGPs (MNVGP=0.070).
The data failed to reveal a significant interaction between age group and the size of flanker compatibility effects. Although the data reported in suggest that such an interaction may be present – with 7–13 year old gamers having disproportionately larger flanker compatibility effects than their non-gaming peers – the effect appears to be driven by large flanker compatibility effects for 7–10 year old gamers and small flanker compatibility effects for 11–13 year old non-gamers. Therefore the data are inconclusive with respect to action video gaming having greater effects for younger gamers.
Figure 5 Incongruent flankers slowed down the responses of video game players more than it did those of non-video game players, given the RTs expected as a function of both age and video game experience. Higher flanker compatibility effects provide an index of (more ...)
The omnibus ANOVA below further addresses how changes in orienting and flanker compatibility effects may be best understood in terms of attentional allocation, by looking at changes to performance in which of the experimental conditions lead to the observed differences.
Omnibus RTnormed ANOVA
The omnibus ANOVA was repeated using the RTnormed that were used to calculate the attention effects. Importantly, the main effects of age group (F (3, 117) = 1.27, p = .289, eta2p = .03) and videogame playing (F (1, 117) = 1.19, p = .277, eta2p = .01) did not approach significance, nor did they interact significantly (F (3, 117) = 0.34, p = .771, eta2p = .01). With the applied corrections achieving their aims – there were no statistically significant baseline differences in RT between groups – the analysis also revealed, as expected, significant main effects of flanker type, due to slower RTnormed values in the presence of incongruent flankers (F (1, 117) = 484.88, p < .001, eta2p = .81) and of cue type (F (3, 351) = 164.71, p < .001, eta2p = .59). Two-way interactions between flanker type and cue type (F (3, 351) = 12.69, p < .001, eta2p = .10; ), flanker type and video game playing (F (1, 117) = 19.71, p < .001, eta2p = .14; ) and cue type and video game playing (F (3, 351) = 4.96, p = .002, eta2p = .04; ) were statistically significant. There was also a statistically significant three-way interaction between flanker type, cue type and video game playing (F (3, 351)=2.76, p=.042, eta2p=.02).
The effects of incongruent flankers were more pronounced – relative to those of congruent flankers – following center or double cues that provided only temporal information about target onset.
Figure 7 (A) Incongruent flankers slowed down the responses of VGPs more than it did those of NVGPs, given the RTs expected as a function of both age and video game experience; (B) there were differential effects of cue as a function of video game experience, (more ...)
Omnibus Error Analysis
A four-way mixed ANOVA was performed on the error data with flanker type (incongruent, congruent) and cue type (absent, center, double, spatial) as within subjects factors, and age group (7–10 yrs, 11–13 yrs, 14–17 yrs, 18–22 yrs) and video game playing (NVGP, VGP) as between subjects factors. This analysis revealed significant main effects of flanker type (F (1, 117)=103.67, p<.001, eta2p=.47) and age group (F (3, 117)=8.46, p<.001, eta2p=.178). These were qualified by a significant two-way interaction between flanker type and age group (F (3, 117)=5.06, p=.002, eta2p=.115). For incongruent flanker trials, younger subjects made more errors than older subjects (M7–10YRS=7.7%, M11–13YRS=6.0%, M14–17YRS=4.6%, M18–22YRS=2.4%). Error rates were equivalent for congruent flanker trials across the ages tested (M7–10YRS=1.8%, M11–13YRS=1.1%, M14–17YRS=1.3%, M18–22YRS=0.8%), reflecting the small impact of congruent flankers observed in RT measures. Importantly, the main effect of video game playing on error rate was not statistically significant (F (1, 117)=2.74, p=.102, eta2p=.02; MNVGP=2.84%, MVGP=2.91%), nor did it interact with any other factor (all Fs < 1).