The primary measures of interest were response time (RT) and error rate. The RTs for each observer were log-transformed, and the resulting averages were used as the basis for the search slope (the average time needed per item in the display) and baseline (time needed for the first item). The search slope was calculated by regressing the average RTs for each set size (3, 6, 12) against the set size itself. The baseline was calculated by taking the intercept of the regression line with the y-axis (corresponding to the setup time for processing) and adding the search slope (corresponding to the time needed for a set size of 1). We omitted data from participants whose search slopes deviated more than three standard units from their group’s mean. Applying this standard, we excluded the data from two participants (one each from the R-h and R-e groups).
The results for the various groups are shown in Fig. 3. In all cases, the analysis involved target-present trials only; target-absent slopes were highly variable (likely indicating large differences in high-level search strategies), and so were omitted from the analysis. Since we found no significant differences in the error rates for any group [F(4, 159) = 1.6, p > .19], these were omitted from the subsequent analysis.
Search asymmetry was assessed in two ways. The first was via a within-subjects t test of the slopes between the short- and long-target conditions for each group. Here, “asymmetry” was regarded as categorical: It existed if and only if the slopes differed reliably. The second, more quantitative measure of asymmetry was the slope ratio, the ratio of the short- to the long-target slopes (cf. Treisman & Souther, 1985). These ratios were log-transformed to obtain a reasonably normal distribution; analysis of the slope ratios was done on these transformed values.
The results are shown in Table 1. Replicating the results of Ueda et al. (2016), Group R-e showed no significant asymmetry (short = 56 ms/item, long = 50 ms/item; p > .2, ratio = 1.29), and Group R-h also showed no significant asymmetry (short = 47 ms/item, long = 42 ms/item; p > .4, ratio = 1.33). Although both groups tended to have slightly higher baselines for short targets, these tendencies were not significant, either for Group R-e (short = 947 ms, long = 904 ms) or Group R-h (short = 1,025 ms, long = 943 ms; both ps > .35). More generally, no significant differences between the groups were apparent in either the slopes (both ps > .17) or baselines (both ps > .4), suggesting no effect of language of instruction for East Asians with only a limited exposure to a Western environment. For most purposes, therefore, we combined these into a single group (R). This combined group (n = 44) still showed no significant asymmetry in slopes (short = 52 ms/item, long = 46 ms/item; p > .12, ratio = 1.3) or baselines (short = 979 ms, long = 922 ms; p > .1).
Meanwhile, replicating previous studies of Westerners (e.g., Treisman & Gormican, 1988), Group W—the Western counterpart of Group R—showed an asymmetry that was quite strong: Differences were found both in both slopes (short = 53 ms/item, long = 39 ms/item) [t(29) = 3.5, p < .002, d = 0.66, ratio = 1.81) and baselines (short = 864 ms, long = 785 ms) [t(29) = 2.4, p < .03, d = 0.46].
Interestingly, group L-e (long-term immigrants given the instructions in English) also showed a strong asymmetry, with search being much faster for long targets than for short (short = 58 ms/item, long = 37 ms/item) [t(59) = 6.4, p < .0001, d = 0.81, ratio = 2.05]. Baselines also tended to differ (short = 894 ms, long = 847 ms) [t(59) = 2.03, p < .05], though this effect was relatively small (d = 0.26). Post-hoc comparisons using Tukey’s HSD showed that the L-e slope ratio did not differ from that of Group W (2.05 vs. 1.81, p > .4). Indeed, Group L-e did not differ significantly from Group W in either slopes [short: t(59, 29) = 0.52, p > .41; long: t(59, 29) = 0.28, p > .67] or baselines [short: t(59, 29) = 0.48, p > .56; long: t(59, 29) = 0.94, p > .16]. In other words, performance in long-term immigrants was much the same as that of native-born Westerners.
In contrast, post-hoc comparisons using Tukey’s HSD confirmed that the slope ratios of Group L-e differed from those of Group R (p < .02). To analyze this in greater detail, we conducted a two-way mixed analysis of variance (ANOVA) with Group (L-e, R) and Target Length as factors. This revealed a main effect of target length [F(1, 102) = 25.2, p < .0001], no main effect of time in Canada (p > . 55), but an interaction between target length and time in Canada [F(1, 102) = 7.9, p < .007]: Namely, only Group L-e was faster at searching for long than for short targets (p < .05), whereas Group R was equally slow at searching for long and short targets (p > .05). Thus, although length of time in Canada did not affect overall search speeds, it did affect the differences: participants who had lived in Canada longer showed a greater difference in speed—and in particular, were faster for long targets—than recent arrivals.
Meanwhile, Group L-h (long-term immigrants given the instructions in their heritage language) exhibited a pattern largely similar to that for Group R. For observers in Group L-h, search slopes did not differ significantly (short = 43 ms/item, long = 40 ms/item; p > .5, ratio = 1.2), nor was there a significant difference in baselines (short = 899 ms, long = 836 ms; p = .08). A post-hoc two-factor ANOVA of Groups L-h and L-e showed no main effect of instruction group [F(1, 86) = 2.6, p > .1], but did show a strong interaction between instruction group and target length [F(1, 86) = 8.07, p < .006]: In particular, Group L-h had shallower short-target slopes than did Group L-e [t(86) = 1.9, p < .05, d = 0.62], whereas long-target slopes did not significantly differ (p > .05). In summary, then, instructions in their heritage language appeared to cause these observers to use a perceptual mode that resulted in relatively little search asymmetry, with search being faster overall than for Group L-e.
Although the proportions of male to female participants were roughly similar overall, the proportions did differ between groups (p < .02). To rule out the possibility that these differences were responsible for the effects found, we tested whether gender had any effect on the slope ratios. No significant effect was found (p > .39), making it unlikely that these imbalances affected our results.
Correlations with questionnaire responses
To determine whether search performance depends on nonvisual environmental factors (in particular, ones involving culture), we examined the extent to which the degree of asymmetry—or more precisely, the slope ratio—correlated with group and individual responses on the questionnaires. The first of these measures (VAI) assessed two independent dimensions (affinities to the heritage and host cultures), each based on several questions involving a Likert scale ranging from 1 (strongly disagree) to 9 (strongly agree). Differences in participants’ raw VAI responses can be dampened by a bias to default to the “neutral” response. To minimize dampening effects from participants who did not respond strongly to some questions, we calculated for each question the “median-split score”—that is, the percentage of responses from each participant that were greater than 5. Although the median-split score is a coarser measure than the raw responses, neutral responses do not dampen it. As such, it was a more sensitive measure than the means for group and individual polarizations toward the heritage versus host cultures. (See Appendix C). The second source of information was the demographic questionnaire, which described the life history of the participant, especially with regard to exposure to different cultures. (For details, see Appendix D). Westerners were exempted from these analyses. In addition, several participants (three from Group L-h, one from L-e, one from R-h, and one from R-e) turned in questionnaires that were blank or otherwise incorrectly completed. The data from these participants were excluded.
For the VAI, we calculated the average median-split percentages on the heritage and Western subscales for each group. Scores are shown in Table 2.
The performance of East Asians on the framed-line task appears to depend on their level of self-identification with their heritage culture (Hedden et al., 2008). Such self-identification could capture a host of factors that might influence visual search, such as susceptibility to visual or linguistic (i.e., the language of instruction) primes. In accord with the findings of Hedden et al., VAI responses at the group level were consistent with some of the patterns in visual search. Recent arrivals had relatively symmetric search and also indicated greater self-identification with their heritage than with Western culture. On the other hand, although the L-h and L-e groups differed in search, the L-h group did not show a significantly greater identification with their heritage than with Western culture, suggesting that these factors may not have played a large role in visual search.
We also calculated for each East Asian group the Pearson correlations between the slope ratios and the various quantities on the two questionnaires. This was also done for all East Asians pooled together. (We refer to this pooled group as Group A.) Almost none of our predictors had a significant effect. The only exceptions were the North American subscale and the number of months in Canada. Controlling for months in Canada, individuals with greater North American subscale scores showed a weak correlation with faster search for long lines (r = –.18, p < .04). Meanwhile, when controlling for scores on the North American subscale, a positive relationship was found between slope ratio and months in Canada (r = .25, p < .004); this essentially restates our main finding that the slope ratio depended on length of time in Canada.
The absence of a relationship between search performance and responses to these questionnaires (other than length of time in Canada), coupled with the observation that group-level differences in cultural affinity failed to differentiate the L-h and L-e groups, suggests that socio-cultural identity plays at most a minor role in visual search asymmetry.
Finally, to get an approximate estimate of how search asymmetry is affected by the amount of time spent in a new environment, a simple post-hoc analysis was carried out on the East Asians given instructions in English. This group was split into four quartiles according to time spent in Canada, with the average of each quartile taken as its value. The results are shown in Fig. 4. As is clear from this figure, search asymmetry is low during the first year or so in Canada, after which it climbs to a level similar to that of native-born Westerners and stays at that level.
Performance on this task was assessed by comparing two measures: (i) the accuracy of recognition when a focal object reappears in the background in which it was originally seen, and (ii) the accuracy when it appears in a new background (referred to as “old” and “new,” respectively). The degree of holism is reflected in the extent to which these measures differ: Worse performance for new backgrounds indicates a more holistic mode (Chua et al., 2005). Similar to the case for search asymmetry, holism was assessed in two different ways. First, it was treated as a categorical factor, with error rates in the two conditions being compared via within-subjects t tests. Next, the degree of holism was assessed, using the recognition ratio (error rate for new backgrounds/error rate for old ones). Owing to random experimenter error, 11 participants (three in Group R-h, two in R-e, two in L-e, one in L-h, and three in W) left the study room before completing the object recognition task. Because these errors were distributed uniformly over all five groups, they reduced the degrees of freedom but not the internal validity of our analyses of the groups’ object recognition performance.
The results are shown in Fig. 5, and detailed comparisons are in Table 3. Consistent with Chua et al. (2005), all East Asian observers exhibited a clear holistic mode, performing measurably worse for new than for old backgrounds. Interestingly, participants instructed in their heritage language (Groups R-h and L-h) performed no better on the object recognition task than those given instructions in English. Indeed, the recognition ratios for all groups—including Group W—did not differ significantly from each other (one-way ANOVA, p > .57). In other words, Group W (Westerners who had lived in Vancouver most of their lives) had about the same degree of holism (old = 22 %, new = 35 %; t(39) = 3.3, p < .003, d = 0.83, ratio = 2.1) as Group A (old = 28 %, new = 43 %; t(125) = 7.2, p < .0001, d = 0.69, ratio = 2.7).
Westerners appeared to have lower error rates on the task than did East Asians. If our Westerners had an analytical processing style, East Asians should have shown a greater difference in performance between the old and new backgrounds than Westerners. To assess this possibility, we conducted a two-way mixed ANOVA comparing the error rates for old and new backgrounds between the Westerners and the (pooled) Asian group. The test confirmed the main effect of group, with Westerners performing better than East Asians overall [F(1, 174.7)Footnote 5 = 8.34, p < .005], and the main effect of background, with new backgrounds yielding higher error rates than old backgrounds [F(1, 174.7) = 28.5, p < .0001]. Crucially, the test failed to find an interaction between background and group [F(1, 174.7) = 0.23, p > .63]: Both groups appeared equally affected by the switch from a new to an old background, consistent with a common, holistic processing style. The RTs of Groups W and A also did not differ significantly (W = 1,850.6, A = 1,845.3; p > .65). In any event, given the lack of difference in relative error rates between the two groups, it is unlikely that speed–accuracy trade-offs can account for this pattern of results.
It is not clear why the native Vancouverites tested here (Group W) performed differently from the Midwesterners tested by Chua et al. (2005). Assuming that this was not due to sampling error or to the slightly smaller images used in our experiments, it may be that a more holistic mode is induced by living in the high visual density of the Vancouver urban area. It may also be that Vancouver has enough of an East Asian influence in its visual environment or social culture to induce all of its inhabitants to adopt a more holistic visual mode. Settling this issue must await future studies. In any event, it is worth pointing out that since the Westerners tested here used a holistic mode in visual recognition similar to that of East Asians, the analytic/holistic distinction (at least, as applied to visual memory) is unlikely to account for the differences in search asymmetry encountered in Experiment 1.
Correlations with search
To further investigate possible connections between visual search and visual recognition, we calculated Pearson correlations between the recognition ratios and slope ratios (as well as between the slopes for short and long targets). This was done for each of the East Asian groups, as well as for all observers pooled together. No correlations were significant (all ps > .07). This result accords with the conclusion above that analytic/holistic differences in object recognition are unlikely to explain the differences in search found here.
Correlations with questionnaire responses
Finally, for each of the four non-Western groups, we calculated Pearson correlations between the recognition ratios and various quantities from the questionnaire data, such as months in Canada and the median-split scores on the two VAI dimensions. This was also done for these four groups pooled together.
As in the case of search asymmetry, almost none of our predictors had a significant effect. The only exception was a modest (r = –.19, p < .04) inverse relationship between holism and the participant’s median-split score on the North American VAI dimension. Pooling across all East Asians, observers who were more strongly oriented to North America tended to be somewhat less holistic, regardless of how they stood on the Heritage subscale. In general, however, cultural factors appeared to play a relatively weak role in visual recognition, consistent with performance on this task being largely the same for all cultural groups tested.