This Article Abstract Full Text (PDF) Services Download to citation manager Citing Articles Citing Articles via HighWire PubMed PubMed Citation

Age, Skill Transfer, and Conjunction Search

^a University of Calgary, Alberta, Canada

Geoffrey Ho, Department of Psychology, University of Calgary, 2500 University Drive NW, Calgary, Alberta, Canada T2N 1N4 E-mail: scialfa{at}ucalgary.ca.

Decision Editor: Margie E. Lachman, PhD

	Abstract

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 
Ten younger and ten older adults were provided with 16 sessions of conjunction search under consistent mapping, where target and distractors remain constant in identity. The target and one distractor were reversed after every fourth session. After the first four training sessions, on target-present trials, display size slopes were near zero for both age groups. However, on target-absent trials, older adults continued to show significantly larger display size effects than younger adults. There were no systematic age differences in either the probability of fixating objects that possess the target's features or in the amount of disruption at any reversal. Thus, although older adults exhibited more conservative criteria in visual search, they developed proficient and flexible search skill to the same degree as their younger counterparts. These data have implications for models of visual attention, skill acquisition, and cognitive aging.

ONE of the more important types of skill acquisition is perceptual learning, the ability to associate perceptual stimuli with task-related responses. A prototypical perceptual learning task is visual search, which involves the detection, localization, and identification of a target placed in a group of distractors. Practice has been shown to improve search efficiency in younger adults (Schneider and Shiffrin 1977; Steinman 1987), and there are only small age differences in the development of automaticity in both feature search (Anandam and Scialfa 1999) and two-dimensional conjunction search (Scialfa, Jenkins, Hamaluk, and Skaloud 2000). This investigation addressed two issues. First, how does repeated target–distractor reversal inform us about the mechanisms mediating skill development and plasticity? Second, do older adults differ in the flexible use of skilled conjunction search?

Age differences that are found in a variety of search tasks can persist even after extensive practice (Fisk, Hertzog, Lee, Rogers, and Anderson-Garlach 1994; Fisk and Rogers 1991; Rogers 1992; Rogers and Fisk 1991). Fisk and Rogers 1991 trained younger and older participants in a pure visual search task, a pure memory search task, or a hybrid visual/memory task using letters (Experiment 1) and semantically related words (Experiment 2). Because age differences were found only on tasks requiring visual search that depends on attending preferentially to targets, they concluded that elderly adults have deficits in priority learning. This has been called the priority-learning deficit hypothesis.

Several other investigations have arrived at similar conclusions by using a reversal condition to assess automaticity. For example, Fisk and colleagues 1994 provided consistent mapping (CM) visual search training to older and younger adults and gave a single session of reversal after training. The younger adults showed significant disruption, whereas little disruption was observed in older adults. This led Fisk and colleagues to conclude that only younger adults had learned to effortlessly and involuntarily attend to the target.

Not all studies have found age deficits in the development of proficient visual search. Anandam and Scialfa 1999 gave younger and older participants approximately 3,000 trials of CM training and one session of reversal on an orientation-based feature search task. When targets were presented in central vision, there were no age differences in the display size effect after training or the disruption at reversal. Scialfa and colleagues 2000 trained participants on a CM conjunction search task. Again, there were no age differences in asymptotic display size effects or in the disruption shown in a single session of reversal.

There has been only one recent study of aging and the positive transfer of skilled search. Fisk, Rogers, Cooper, and Gilbert 1997 compared younger and older observers in a semantic category visual search, in which targets and distractors are defined by their membership in a superordinate category (e.g., animals). Younger adults demonstrated positive transfer to new exemplars of the trained target categories, whereas older adults did not. Fisk and colleagues concluded that learning in visual search in older adults is feature specific and does not transfer to semantically related, but physically different, stimuli. This would explain why older adults show no deficits in featured-based search (Anandam and Scialfa 1999; Scialfa et al. 2000) and also suggests that elderly people might be as adept as younger adults at transferring search skill in conjunction search based on object features.

In the present study, older and younger participants were trained for 16 sessions on a two-dimensional conjunction search task. At three points in the training regimen, the target was interchanged with one of the distractors; that is, there was a reversal that then continued for four additional sessions of CM training. Along with global reaction time (RT) and accuracy, we measured the eye movements made as participants performed the search task. Eye movements are highly correlated with RTs and can sometimes discriminate between serial and parallel search (Scialfa et al. 2000; Scialfa and Joffe 1997, Scialfa and Joffe 1998; Williams, Reingold, Moscovitch, and Behrmenn 1997; Zelinsky and Sheinberg 1997). By determining which objects are being fixated (Scialfa et al. 2000; Scialfa and Joffe 1998; Williams and Reingold 2001), researchers also find that eye movements can demonstrate the feature-based selectivity that is assumed to direct attention in several models of visual search (e.g., Schneider and Shiffrin 1977; Wolfe 1994). In the present study, feature-based selectivity was indexed by what we called a selection factor, the ratio of fixations that landed on objects sharing the target's luminance contrast relative to the total number of fixations that landed on any object. Like color (Williams and Reingold 2001), luminance contrast is particularly salient and reliably provides the basis for object selection (Scialfa et al. 2000; Scialfa and Joffe 1998).

According to strength theory, CM training results in a transition from controlled to automatic processing, and this should be reflected in the elimination of display size effects on RT. Moreover, with practice, strength theory predicts that attention–attraction strength increases for objects that share target features, and this should be reflected in an increase in the selection factor. On initial reversal, strength theory predicts significant disruption in performance that is reflected in a return of the display size effect on RTs. Because attention–attraction strength is biased to the former target, there is the expectation of a significant reduction in the selection factor. Similar disruptions should occur with subsequent reversals, with the magnitude depending on the extent to which the search has been automatized.

Rule-based learning (Kramer, Strayer, and Buckley 1990) or guided search (Wolfe 1994) make similar predictions for the first four CM training sessions. Both approaches allow for some disruption at the first reversal, because observers do not know to apply an algorithm or modulate top-down activation until they are exposed to conditions in which such generalization is appropriate. These conditions commence only at the first reversal and may not be immediately apparent. Once learned, however, these mechanisms can be applied at subsequent reversals, leading to a prediction of minimal disruption and perhaps positive transfer.

Recent studies involving feature and conjunction search (Anandam and Scialfa 1999; Scialfa et al. 2000) suggest that older adults develop an automatic response to the target. As such, compared with their younger counterparts, they exhibit equivalent reductions in display size effects following CM training and equivalent disruption at reversal. Aging hypotheses with regard to repeated reversals are less clear. To the extent that generalization of skill to untrained stimuli is cognitively similar in semantic category and visual search, the data from Fisk and colleagues 1997 would predict less facilitation among elderly people than among younger adults. If, however, elders are not disadvantaged in feature-based learning, then age differences in skill transfer are expected to be minimal.

	Methods

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 
Participants
Ten younger adults (M = 23.20 years, SD = 3.16 years) and 10 older adults (M = 66.20 years, SD = 2.97 years) were recruited from the University of Calgary and the Calgary, Alberta, Canada, community to participate in the study. All the participants received $5.00 Canadian for each session attended and received a bonus of $10 Canadian for completing the entire experiment. Although both age groups were relatively well educated, younger participants had more education (M = 16.1 years, SD = 1.2 years) than did older adults, (M= 13.5 years, SD = 2.72 years), t(18) = 2.77, p = .013.

By self-report, all participants were in good overall and visual health. They had not been hospitalized in the past year nor were they currently under a physician's care for a serious medical condition. Corrected acuity (Burton 100 TLS) was measured at the test distance of 50 cm using postscript generated Landolt Cs with eight targets for each level of minimum angle of resolution, which varied in steps of approximately .05 log units. Older adults (M = .915, SD = .29) did not differ from younger adults (M = .765, SD = .18) on corrected acuity, t(18) = -1.40, p = .178. Intraocular pressure (Reichart NCT II), was within normal limits for all participants.

Apparatus
The conjunction search displays were presented on the Eyegaze Development System (EDS) using software provided by LC Technologies, Inc. (Fairfax, VA). The EDS uses a 486 computer platform to measure eye movements using the pupil center/corneal reflection technique. An LED placed underneath the monitor floods the eye with a low-level infrared light (880 nm). A Sanyo CCD high speed, infrared camera collects the infrared reflections at a rate of 60 Hz. Stimuli were presented on a 15-in Sony Trinitron Multiscan CPD-100 GS monitor. Monitor resolution was set at 640 x 480 pixels and the refresh rate was 60 Hz. Recalibration occurred at the end of each trial block.

Stimuli
The search displays consisted of one target and two types of distractors. The stimuli were white (79.09 cd/m²) and black (6.24 cd/m²) line segments that were oriented 45° to either the right or the left, presented on a gray (43.33 cd/m²) background (see Scialfa and Joffe 1998, for examples of similar displays). Each line was approximately 6.3 mm x 0.84 mm, which at the test distance subtended 0.72° in length and 0.01° in width. The stimuli were restricted to an active, square display area of 131.88 mm (14.78°) and minimum separation for the stimuli was 4.2 mm (0.48°). We used a smaller spatial extent than we had in previous work (e.g., Scialfa and Joffe 1997) because we were not interested in exploring eccentricity effects that arise in conjunction search and may interact with age.

Design
The study consisted of 16 nonconsecutive days of training divided into four 4-day training periods, the last three beginning with reversals. Each session had eight blocks of 30 trials for a total of 240 trials. The first training period involved a white–right target embedded in white–left and black–right distractors. At the beginning of the second training period, all participants underwent a reversal. The target was switched to a black–right line, which was displayed with white–right and black–left distractors. Reversals were provided two additional times such that the same targets and distractors were defined in the first and third training periods and in the second and fourth training periods.

Procedure
Each trial began with a centrally placed black fixation cross on a gray background. Participants were instructed to fixate the cross at the beginning of each trial. To initiate the onset of the search screen, they pressed a key, and the fixation screen disappeared after a randomly determined 50, 100, or 150 ms. Immediately after the offset of the fixation screen, the search displays were presented. The observer was given 5 s to search the display and to respond as to whether the target was present or absent by pressing the corresponding key on the keyboard. Feedback was given after each response; a plus sign was shown if the answer was correct, a minus sign was shown if the answer was incorrect, and a question mark was displayed if the observer had not responded in the 5 s allowed.

Display size varied between 6, 12, and 24 items, and on one half of the trials the target was present. Display size and target presence were randomized within each block. When the target was absent, there were equal numbers of each type of distractor. When the target was present, it replaced one of the two distractor types. This was randomized across the target-present trials so that for any given trial, the number of each distractor type was unequal, but across all trials, both types of distractors were presented equally often.

	Results and Discussion

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 
Performance was analyzed for errors, RT, fixation number, and the selection factor. Because of the high correlation between RT and fixation number (.86–.89 across both age groups at Sessions 1, 4, 5, 8, 9, 12, 13, and 16), we report only the RT data. Only trials ending in a correct response were examined (see A, Note 1). Fixations were calculated if two or more samples of eye position were recorded within an 11-pixel window horizontally and vertically (approximately 0.5°). As noted above, the selection factor was operationalized as the ratio of fixations that landed on objects sharing the target's luminance contrast relative to the total number of fixations that landed on any object. The data from one older participant were omitted from Session 1, and another participant's data were omitted from Sessions 12 and 13 as a result of recording errors. Furthermore, one younger participant consistently made few or no eye movements throughout training and, as a result, the selection factor could not be calculated.

Within specific times in training (e.g., the first session), all dependent measures were submitted to separate 2 (Age) x 2 (Presence) x 3 (Display Size) univariate, split-plot analyses of variance (ANOVA), with age as a between-subjects factor and target presence and display size as within-subjects factors. A Geisser–Greenhouse correction was used to control for violations of sphericity. Because of the large number of analyses, only significant findings with their respective p values are reported and discussed.

Frequently, we summarize the display size effect in terms of the slope relating performance to the number of items in the display. Although this summary measure may obscure nonlinearities in the effect, it is an efficient means of communicating the difficulty of search. Additionally, a 2:1 ratio of target-absent to target-present slopes is typically associated with a serial, self-terminating search, because, on average, the observers must search only one half of the items on a target-present trial but must search exhaustively on a target-absent trial (but see Townsend 1972, for alternative interpretations). In contrast, a 1:1 ratio is often viewed as evidence of parallel search, and ratios greater than 2:1 may result from criterion shifts that have the greater impact on target-absent trials (Chun and Wolfe 1996).

Table 1 provides cell means for the entire protocol, and Fig. 1 shows the performance on each dependent measure over the 16 sessions as a function of age and target presence. The figure provides a picture of some broad patterns in the data. Errors were relatively low throughout training (Fig. 1), but were higher for the young and increased for both age groups at the first reversal, particularly on target-present trials. Improvement in search efficiency is evident in both younger and older adults. This is clearly seen in the RT data (Fig. 1). Whereas disruption in RT was substantial at the first reversal, it was short lived and reduced considerably on subsequent reversals. Additionally, there is no evidence of disruption in the selection factor (Fig. 1). Feature-based selection was actually somewhat lower for younger adults than we have previously reported (Scialfa et al. 2000) but may be attenuated because of the low number of fixations needed by them before finding a target (see Zelinsky and Sheinberg 1997). On average, older adults perform less efficiently than younger adults. They had longer RTs, especially on target-absent trials (see A, Note 2). However, there are no obvious age differences in the selection factor or in disruption at reversal. What follows is a more fine-grained analysis of these global patterns.

View larger version (23K): [in this window] [in a new window]   Figure 1. Performance as a function of age and target-presence on (A) proportion of errors, (B) reaction time, and (C) the selection factor. Dashed vertical lines indicate reversals. Y = young; O = old; TP = target present; TA = target absent.

Analysis of RTs indicates that older adults were slower at responding than younger adults were (p = .002). RTs were slower on target-absent trials (p < .001) and increased with display size (p < .001). The three-way interaction was significant (p = .046). Post hoc tests indicate that age differences in display size effects were significant only on target-absent trials (p = .025).

On target-present trials, display size slopes for younger adults (15.82 ms/item) and older adults (19.39 ms/item) were similar. However, on target-absent trials, display size slopes among older adults were much larger than among younger adults (49.35 ms/item vs. 27.99 ms/item). Both age groups approximated a 2:1 slope ratio.

The selection factor was quite high for both age groups. On average, the selection factor declined on target-absent trials (p < .001) and with larger displays (p < .001), but no other effects were obtained (ps > .08).

Synthesizing these analyses, we found that performance in the first session of practice was consistent with previous data on conjunction search (e.g., Scialfa and Joffe 1998; Treisman and Sato 1990; Wolfe 1994), yielding display size effects on RT. These effects were larger for target-absent trials. The display size effect suggested that search was effortful and the ratios of target-absent to target-present slopes was consistent with a serial, self-terminating model of search. Accuracy was lower for target-present trials, indicating that targets were sometimes missed, presumably because search terminated prematurely (see Wolfe 1994).

The selection-factor data indicated that participants were quickly able to attend to items that had the same luminance contrast as the target. This is a robust effect. Scialfa and colleagues 2000 and Scialfa and Joffe 1998 both found that participants learned to quickly fixate white objects when they searched for white targets. Similarly, Shen, Reingold, and Pomplun 2000 reported rapid feature-based selectivity for color when participants were presented with items defined by their color and orientation. The selection factor data could be explained as the result of rapid modulation of top-down activation (Treisman and Sato 1990; Wolfe 1994). On the other hand, strength theory (e.g, Schneider and Shiffrin 1977) assumes that priority learning occurs gradually and thus has difficulty accounting for the high level of feature-based selection seen at the initial training session.

There were age differences in display size effects for RT, and these effects were larger for target-absent trials (Plude and Doussard-Roosevelt 1989; Scialfa et al. 2000). At first glance, these results might be anticipated if elderly adults have difficulties in attending selectively to target items, as predicted by the priority learning deficit hypothesis (Fisk and Rogers 1991). There are, however, two findings that run counter to this conclusion. First, there were no age differences in display size effects on target-present trials for RT. Deficits in selective attention should be seen when the target is present, as well as when it is not. Second, the selection factor data indicated that in the first session of training, there were no age deficits in the ability to attend selectively to the target's luminance contrast. To the extent that priority learning mediates feature selection, this result is inconsistent with the view that elderly people demonstrate a priority learning deficit.

The larger age differences on target-absent trials could be explained by arguing that the elderly participants set their activation thresholds to a more conservative level, which has a greater impact on target-absent trials (Chun and Wolfe 1996). Discussion of age differences in cautiousness is deferred until additional data are reported.

Session 4
More errors were made on target-present trials than on target-absent trials (p = .005). The Display Size x Presence interaction was also significant because the presence effect increased with larger displays (p = .042). Again, these were missed signals, a common finding in visual search.

For RTs, age (p < .001), display size (p < .001), and presence effects (p < .001) were significant, as were several higher order interactions. Relative to the young participants, older adults responded more slowly on target-absent trials (p < .001) and with larger display sizes (p < .001). The three-way interaction was also significant (p = .042), and this occurred because age differences in display size effects were found only on target-absent trials (p = .004).

The selection factor was lower on target-absent trials (p < .001) and larger display sizes (p < .001), but still clearly favored white objects (Table 1 ). There was an Age x Display Size interaction (p = .025) that arose because the elderly adults had a larger selection factor for displays containing small numbers of items, whereas there were no age differences in the selection factor at the largest display size.

Learning in Sessions 1 to 4
To assess the degree of learning that took place during the first period of CM training, we examined the change in RT between the first and fourth training sessions in Training Period (2) x Age (2) x Display Size (3) x Presence (2) mixed-model ANOVAs.

There were significant effects of training period (p < .001), indicative of general improvement in performance. There was a Training Period x Display Size interaction (p = .003) because the larger displays showed the greatest improvement and a marginal Training Period x Display Size x Presence interaction (p = .07), with larger changes occurring on target-absent trials. Age did not impact training generally or in interaction with any other factors (ps > .44).

The learning analyses indicate that search proficiency improved with CM training as demonstrated in the reduction in RT, combined with a decrease in the display size effect. Evidence for automatic processing was illustrated in target-present data wherein the display size slopes for both age groups were quite low (younger = 5.31 ms/item; older = 4.84 ms/item). These trends replicate previous research on the development of automaticity in visual search in both younger (Schneider and Shiffrin 1977; Scialfa and Joffe 1998; Steinman 1987; Theeuwes and Kooi 1994) and older adults (Anandam and Scialfa 1999; Scialfa et al. 2000).

Age differences in learning rates were not found on RT, and the selection factor data indicated that older adults had no difficulty using feature-based selective attention to facilitate search. However, at Session 4, large age differences in display size effects were obtained on target-absent trials (younger = 13.88 ms/item; older =29.09 ms/item). The target-present data and the selection-factor analyses indicate that this is not the result of a priority learning deficit, which would be manifested even when the target is present and would influence the probability of fixating items that share the target's features. Rather, it appears to be a consequence of a more conservative criterion for terminating search. This explanation has been proposed to account for greater age differences on target-absent trials in both visual search (Madden and Nebes 1980) and memory search (Strayer and Kramer 1994a).

Session 5
First reversal.
Errors were more common on target-present trials (p < .001) and increased with display size (p < .001). A Display Size x Presence interaction was also found (p < .001), reflecting an increase in errors on target-present trials when the display contained 24 items. There were no significant effects involving age (ps > .169).

The RT analyses found effects of age (p < .001), display size (p < .001), and presence (p < .001). Age differences were greater on target-absent trials (p = .009), which were associated with larger display size effects (p < .001). Although the three-way interaction was not significant, the trend is similar to that seen previously, wherein age differences were larger on target-absent trials.

At this first reversal, display size slopes for both groups increased to levels indicative of greater effort. On target-present trials, the display size slopes were 11.47 ms/item and 15.83 ms/item for younger adults and older adults, respectively. On target-absent trials, the display size slopes were 23.80 ms/item and 38.77 ms/item for younger adults and older adults, respectively. The target-present to target-absent slope ratio for both age groups were approximately 2:1.

The selection factor now represents the number of fixations landing on a black object, divided by the number of fixations landing on any object. In contrast with RT and fixation number, the selection factor (Table 1 ) remained relatively stable at reversal. As in previous sessions, there were no age differences in the selection factor (p = .751), which declined with display size (p < .001) and on target-absent trials (p = .004). The Display Size x Presence interaction was also significant (p = .049), because the decline in the selection factor was greater for target-present trials than for target-absent trials.

Disruption at the first reversal.
To calculate the amount of disruption resulting from the reversal of target and distractors, we computed difference scores for each condition that contrasted performance immediately before and after reversals (e.g., sessions 4 and 5). To control for individual differences (see Rogers 1992), we then divided this score by the person's performance immediately prior to reversal and multiplied it by 100 to arrive at a percentage disruption score. Thus, positive values suggest performance suffered after reversal.

Because errors were rare in the sessions immediately prior to reversal, disruption scores could not be calculated for many participants. Therefore, the analysis of disruption scores does not include the error data. This should not be taken to mean that there was no disruption in accuracy. As seen in Fig. 1, both age groups showed an abrupt increase in errors on target-present trials involving large numbers of distractors. As such, the disruption data for RT are probably underestimates. Mean disruption data for RT and the selection factor are shown for the first and all subsequent reversals in Table 2 .

To summarize the Session 5 results, both age groups showed disruption in search efficiency when the target and distractors were reversed. This was seen in the display size effects on RTs. Such a pattern is expected under strength-based models of automaticity (e.g., Schneider and Shiffrin 1977), which assert that attention is being allocated involuntarily to the former targets and away from former distractors. Although not an explicit property of either feature integration theory (Treisman and Sato 1990) or guided search (Wolfe 1994), both of these models could also account for disruption at reversal as a reflection of learned top-down activation that is no longer appropriate to the task.

The selection factor indicated some disruption in feature-based selection for larger, target-absent displays, but the amount of disruption was smaller than previously reported (Scialfa et al. 2000). Perhaps the differences are due to changes in the spatial extent and density of the displays. In Scialfa and colleagues 2000, display elements were presented out to almost 18° from fixation, and display size ranged between 8 and 16 elements. In the present study, maximum eccentricity was reduced to approximately 7° and display size ranged between 6 and 24 elements. It may be that when a larger number of items are placed in a smaller area, feature-based selection can operate more flexibly to eliminate nontarget items even under reversal conditions. Using displays of similar density and extent, Siakaluk, Ho, and Scialfa 1999 also reported little disruption in the selection factor following reversal in a three-dimensional conjunction search task. Importantly, it is clear that disruption following the development of an automatic response need not be manifested in the selection factor data.

Replicating recent work in visual search (Anandam and Scialfa 1999; Scialfa et al. 2000), we found there was little evidence for an age difference in the amount of disruption produced by reversal. This observation cannot be easily explained by the priority learning deficit hypothesis (Fisk and Rogers 1991), which predicts that older adults will show less disruption at transfer because they have not modulated attention–attraction strength of targets and distractors to the same extent as young adults have.

Session 8
As shown in Table 1 , target-present trials again produced more errors than target-absent trials did (p < .001), and this interacted with display size (p < .001). No effects involving age were significant (ps > .093).

Older adults had longer RTs than did younger adults ( p < .001). RTs increased with display size (p < .001) and on target-absent trials (p < .001), and the interaction of display size and presence was significant (p < .001). Older adults required more time on target-absent trials (p = .009), but the three-way interaction was not significant (p > .175).

On target-present trials, both younger (4.94 ms/item) and older adults (8.65 ms/item) had display size slopes that indicate easy search and that were considerably smaller than in Session 5. On target-absent trials, slopes indicated that, relative to the young adults (15.42 ms/item), the older adults (28.10 ms/item) continued to have greater difficulty. The slope ratio, 3.1:1 for the young adults and 3.2:1 for the elderly adults, suggested that criterion setting (Chun and Wolfe 1996) is having an impact on target-absent trials.

The selection factor remained high throughout the second training period (Table 1 ). No age differences were found (p = .783); however, the selection factor decreased with display size (p < .001) and on target-absent trials (p = .013).

Learning in Sessions 5 to 8
We examined the learning that took place between Sessions 5 and 8 as in the first training period. The RT data revealed a significant effect of training period (p < .001) that interacted with display size (p < .001) and age (p = .014). The Training Period x Display Size interaction arose because greater improvements in performance were found for larger displays. The Training Period x Age interaction occurred because older adults showed greater average improvement than did younger adults (276 ms vs. 161 ms). No other effects involving training period and age were significant (ps > .45).

Results from the second training period provide several contributions to the understanding of skill development in visual search. Whereas the RT data showed evidence of disruption following the first reversal, this disruption dissipated quickly. By the eighth session of training, observers were searching displays with an efficiency that is associated with effortless processing. Even immediately following reversal, the selection-factor data were inconsistent with the view that attentional allocation to or away from trained objects is reflexive. Rather, because there was no systematic disruption in the selection factor on target-present trials, the data suggest that a more flexible, feature-based allocation of attention mediates efficient search. This may result from rapid modulation of top-down activation (Wolfe 1994) or the application of an algorithm across variations in surface features (Kramer, Strayer, and Buckley 1990).

Contrary to the priority learning deficit hypothesis, older adults had no difficulty learning to attend to the new target features. There were no age differences in search rates in the second training period, and, in fact, older adults showed greater average improvements in RT. There were no age differences in the selection factor, indicating that the elderly participants were able to use object features to facilitate search. Older adults continued to be less efficient at search on target-absent trials, but this may reflect age differences in setting the criteria for termination of a trial that does not contain the target (see Chun and Wolfe 1996).

Session 9
Second reversal.
Younger adults committed more errors than older adults (p = .009), and more errors were committed on target-present trials (p < .001). The Age x Presence interaction was significant (p = .043) because younger adults committed more errors on target-present trials compared with target-absent trials, whereas older adults had comparable performance regardless of target presence (Table 1 ; see also A, Note 3).

The RT data for Session 9 are shown in Table 1 . Older adults were slower than younger adults (p < .001), target-absent RTs were longer than target-present RTs (p < .001), and latency increased with display size (p < .001). Both age groups exhibited larger display size effects on target-absent trials than on target-present trials (p < .001). Age differences were greater on target-absent trials (p = .003) and with larger display sizes (p = .004). The three-way interaction was also significant (p = .002). Post hoc analyses of simple interactions revealed that compared with the younger adults, older adults had more difficulty on target-absent trials (p = .002), but not on target-present trials (p = .167).

For target-present trials, display size slopes for both age groups are indicative of easy search, (younger = 6.04 ms/item; older = 7.75 ms/item), whereas for target-absent trials, slopes were indicative of more effortful search, particularly for older adults (younger = 15.26 ms/item; older = 33.02 ms/item).

The selection factor remained quite high for both age groups. It was higher on target-present than target-absent trials (p < .001) and declined as display size increased (p < .001). No age differences were significant (ps > .171).

Disruption at the second reversal.
Average percentage disruption scores for the second reversal are shown in Table 2 . Latencies decreased (-3.58%) following the second target-distractor reversal (p = .021). This pattern differs markedly from the first reversal, in which disruption was the norm. Age differences were significant (p = .009), with older adults showing greater improvement than younger adults did, but this may suggest floor effects for younger adults. The main effect of display size was significant (p = .001), as was the Age x Display Size interaction (p = .05). However, this interaction was complex. Older adults were less disrupted than younger adults were, particularly when displays contained 6 (p = .021) and 12 objects (p = .002)

There was a nonsignificant improvement (-2.68%) in the selection factor (p = .061) with participants flexibly switching attention back to white objects. No age differences were found (p = .341). There was a significant display size effect (p < .001) that arose because facilitation was found for Display Sizes 6 and 24, whereas disruption occurred with Display Size 12. There was facilitation on target-absent trials but not on target-present trials (p < .001). The Age x Display Size interaction was also significant (p = .019), because younger adults showed less facilitation with larger display sizes, whereas older adults showed less facilitation only on Display Size 12.

In marked contrast to the first reversal, the second reversal did not produce systematic disruption for either younger or older adults. These data pose a challenge for strength theoretic approaches to understanding efficient visual search (e.g., Schneider and Shiffrin 1977). Rather, the data suggest that observers can either rapidly change the top-down activation associated with specific feature values or have learned to apply a rule that transcends surface features. The present design does not allow us to distinguish which of these views provides the best account of skill development in search. What is clear, however, is that older adults show the same flexibility as their younger counterparts.

Session 12
Table 1 shows that younger adults committed more errors than older adults (p = .009) and that more errors were made on target-present trials (p = .004). No other effects were significant (ps > .05).

Relative to younger adults, older adults needed more time to respond (p < .001). Target-absent trials were slower than target-present trials (p < .001), and RTs increased with display size (p < .001). There was a larger display size effect on target-absent trials (p < .001). Age differences were greater on target-absent trials (p = .001) and with larger display sizes (p < .001). The three-way interaction was also significant (p < .001). Post hoc tests showed that on both target-present (p < .001) and target-absent (p = .002) trials, younger adults had smaller display size effects than older adults.

On target-present trials, both younger adults (2.03 ms/item) and older adults (7.83 ms/item) searched displays with little effort. On target-absent trials, this was true for younger adults (8.55 ms/item), but not for older adults (30.08 ms/item). Thus, as at other times in the protocol, older adults were disadvantaged primarily on target-absent trials.

The selection factor decreased on target-absent trials (p < .001) and as display size increased (p < .001). No other effects were significant (ps > .404).

Learning in Sessions 9 to 12
Learning rates were assessed for RT and fixation number with the approach taken in the preceding training periods. For the RT data, there was a significant training effect (p < .001), which was greater with larger displays (p = .017). The Training Period x Display Size x Presence interaction was also significant (p = .003), indicating that the largest gains in performance were on target-absent trials with large display sizes. No effects involving age and training period were significant (ps > .22).

In the third training period, performance improved and feature-based selection remained high. The greatest performance gains were seen with larger displays, and search rates indicated that on target-present trials, both younger and older adults were operating at levels associated with effortless search. The target-absent conditions continued to cause older adults relative difficulty, but the selection factor data indicate that this is not because they were attending to inappropriate features. Again, this probably reflects age differences in cautiousness exercised before termination of a target-absent trial. That is, in the absence of a strong signal, younger adults terminate the trial relatively quickly. This results in more missed signals but also diminishes their RT slopes on target-absent trials.

Session 13
Third reversal.
Table 1 shows that target-present trials produced more errors than did target-absent trials (p = .001). On target-present trials, errors increased when display size was 24, but this increase in errors did not occur on target-absent trials (p = .028). No other effects were significant (ps > .07).

The RTs demonstrate patterns similar to previous sessions. Older adults were slower than younger adults (p < .001), everyone was slower on target-absent trials (p < .001), and RT increased with display size (p < .001). The display size effect was greater on target-absent trials (p < .001). Age differences increased with display size (p = .005) and on target-absent trials (p = .003). The Age x Display Size x Presence interaction was also significant (p < .001). Post hoc analysis of the three-way interaction revealed a pattern similar to that reported in prior sessions. On target-present trials, both age groups performed comparably (p = .631), but on target-absent trials, there was an Age x Display Size interaction (p < .001), suggesting that older adults had more difficulty responding in the absence of a target.

On target-present trials, slopes were near zero for both younger adults (5.97 ms/item) and older adults (7.47 ms/item). On target-absent trials, whereas search slopes for younger adults were shallow (10.70 ms/item), the older adults' slopes were much steeper (31.11 ms/item).

As in previous sessions, no age differences were found in the selection factor (see Table 1 ), but there was a significant effect of display size (p < .001) and target-presence (p = .002). That is, the selection factor decreased with display size and was lower on target-absent trials.

Disruption at the third reversal.
As shown in Table 2 , although there was overall disruption (10.51%) in RT (p < .001), no effects involving age, display size, or presence were significant (ps > .078). Little average disruption is evident (2.01%) in the selection factor data (p = .353). Less disruption was found for Display Size 12 (p = .001) than for Display Sizes 6 or 24. No age effects were significant (ps > .468).

There are three important observations to be gleaned from the third reversal. First, there is disruption in RT, as was seen at the first reversal. Second, as with the second reversal, there was no general effect on the selection factor. Thus, observers were able to quickly adapt to changing target features. Finally, counter to the priority learning deficit hypothesis, there were no age differences in the disruption shown at reversal.

Session 16
Seen in Table 1 , target-present trials produced more errors than target-absent trials did (p = .012). The Display Size x Target Presence interaction was significant, with errors increasing with display size only on target-present trials (p = .028).

The pattern in the RT data from previous sessions persisted to the last training session. Older adults had slower overall RTs (p < .001), RTs were slower on target-absent trials (p < .001), and significant display size effects remained (p < .001). Moreover, display size effects were greater for both groups on target-absent trials (p < .001). Relative to the young, older adults showed greater display size (p < .001) and presence effects (p = .001). Again, the three-way interaction was significant (p < .001). Post hoc analyses showed a significant Age x Display Size interaction on target-present trials (p = .047) and on target-absent trials (p < .001).

Both younger adults (3.57 ms/item) and older adults (6.58 ms/item) found that target-present trials required little effort. However, only younger adults demonstrated shallow slopes on target-absent trials (6.46 ms/item). In contrast, the display size slopes for older adults (21.56 ms/item) indicated that they had much greater difficulty terminating a trial, particularly when the display contained a large number of items.

The selection factor continued to be high through the last session of training. Although feature-based selection declined with display size (p < .001) and was higher on target-present trials (p = .001), no age differences emerged.

Learning in Sessions 12 to 16
For the RT data, there was a significant effect of training period (p < .001) that was greater for larger display sizes (p = .002). Older adults demonstrated larger training benefits (p = .043), but this may be due to floor effects among the younger observers. There were no other interactions involving age and training period (ps > .25).

Thus, the RT data showed improvement during the last period of CM training. Performance gains were greater for larger display sizes and target-absent trials, indicating that search efficiency was enhanced with additional CM training. For younger adults, at both the start and end of this final training period, performance was indicative of effortless search. Among older adults, performance on target-present trials indicated that they were acquiring targets automatically. The data for target-absent trials might be used to argue that they had not reduced attention–attraction strength to distractors, but the selection factor shows that they had no difficulty fixating objects that possessed the relevant target features. It appears, therefore, that age differences in well-practiced CM conjunction search are due to some other factor such as more conservative criteria for terminating a target-absent response.

	General Discussion

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 
Younger and older adults improved with CM training as demonstrated by the reduction in RT, combined with a decrease in the display size effect. Evidence for automatic processing was illustrated in target-present data wherein the display size slopes for both age groups were near zero. These trends replicate previous research on the development of efficient visual search in both younger (Schneider and Shiffrin 1977; Scialfa and Joffe 1998; Steinman 1987; Theeuwes and Kooi 1994) and older adults (Anandam and Scialfa 1999; Scialfa et al. 2000). The initial reversal produced disruption in RT, as well as a return of the display size effect. What has not been previously reported is that subsequent reversals yielded less disruption (and, at the second reversal, facilitation) in RT. In addition, the selection-factor data indicate that not all measures of attentional demand have the same time-course or disruption effects.

The reduction in display size effects and the disruption in RT at first reversal can be explained as the result of increased attention–attraction strength (Schneider and Shiffrin 1977), improvement in the efficiency of top-down relative activation (Treisman and Sato 1990; Wolfe 1994), or rule-based learning (Kramer et al. 1990). None of these models would have difficulty accounting for fixations being biased toward objects sharing the target's features, although strength theory does not predict the speed with which feature-based selection is evident.

Although disruption was seen in the RT data, after the first reversal, participants were able to quickly adapt to search efficiently for the new target. At the second reversal, both younger and older adults responded with little disruption and, on the final reversal, the amount of disruption remained relatively low. Additionally, there was no disruption in the selection-factor data at any point in the protocol.

The guided search model (Wolfe 1994) may be capable of explaining reversal data through very efficient channel switching. To search for a white line oriented 45° to the right, the channels for high luminance contrast, right, and an intermediate slope would receive the greatest activation. At reversal, the channel for high luminance contrast would have to switch to low luminance contrast very quickly to allow the selection factor to remain high.

The reversal data may also be explained by rule-based learning (Strayer and Kramer 1994b, Strayer and Kramer 1994c). Kramer and colleagues 1990(p. 519) argued that disruption need not occur "as long as subjects can capitalize on the higher-order consistencies in a task." In the present study, perhaps participants learned the rule to attend only to objects that shared the target's luminance contrast and then performed a subset search for orientation on the remaining items. Other evidence for subset search has been demonstrated in studies focusing on RTs (Egeth, Virzi, and Garbart 1984; Friedman-Hill and Wolfe 1995; Plude and Doussard-Roosevelt 1989) and, more recently, in eye movement data ( Scialfa et al. 2000; Scialfa and Joffe 1998; Siakaluk et al. 1999; Williams and Reingold 2001; Williams et al. 1997).

If top-down modulation were perfect, then one might expect no disruption on any measure, even at the first reversal. How can we explain the observation that featured-based selection responded rapidly to the first reversal, but disruption is still manifested in RT and fixation number (the latter of which is not shown)? It is likely that this first reversal reduced the visual span over which attention can operate in parallel (Pomplun, Reingold, Shen, and Williams 2000; Scialfa and Joffe 1997). More fixations would be required and RT would increase proportionately. Thus, it is the case that there is some initial behavioral cost associated CM training when it is no longer appropriate, but it is not at the level of feature-based selection and diminishes over subsequent reversals.

Several patterns in the age comparisons were seen consistently throughout training. Although errors were low for both groups, when there were age differences in error rates, it was the elderly adults who were more accurate. Older adults were generally slower prior to executing a response. Like their younger counterparts, among the elderly participants, RT decreased with practice. There was often an Age x Display Size x Presence interaction. Older adults often showed larger display size effects than the younger participants did on target-absent trials, an effect that could arise because of age-related differences in criterion-setting (Madden and Nebes 1980; Salthouse 1982; Strayer and Kramer 1994a). Generally, there were smaller age differences on target-present trials. Additionally, there were no age differences on the selection factor or disruption at any point in the protocol.

These observations cannot be easily explained by the priority learning deficit hypothesis (Fisk and Rogers 1991), which predicts age differences in asymptotic display size effects on both target-present and target-absent trials, along with age differences in disruption and the selection factor. It may be that search for visual primitives such as luminance contrast and orientation does not make the same age-sensitive demands on semantic categorization (Kutas and Iragui 1998) or speed of semantic retrieval (see MacKay and Abrams 1996) as in semantic category search. It may also be that the task-switching used in studies of semantic category search put the elderly adults at a disadvantage because they have difficulty with executive control (Kray and Lindenberger 1998). Answers to these questions will be the topic of future research, but meanwhile, it seems clear that older adults do not have difficulty developing proficient, flexible skill at conjunction search.

	Acknowledgments

 
This article was written as part of Geoffrey Ho's master's thesis and was supported by a grant from the Natural Sciences and Engineering Research Council of Canada. Thanks go to Dr. J. K. Caird, Dr. J. H. Mueller, and Dr. J. R. Parker for providing valuable comments regarding this study and to David Stewart for his computer programming assistance.

Received for publication July 24, 2000. Accepted for publication September 10, 2001.

	Appendix

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 
Notes

In fact, there were other dependent measures analyzed, including first, average, and last fixation duration. Like the selection factor and unlike RT, these measures evidenced relative stability throughout training.

We carried out Brinley analyses (Brinley 1965) of the RT data across the entire protocol. In each case, the data were collapsed across display size but not across target presence, as in Fig. 1. Predicting older adults' RTs using only younger adults' RTs yielded a slowing factor of 1.19 that accounted for 45% of the variance. After adding target presence as a second predictor, the slowing factor remained relatively unchanged at 1.13. However, the explained variance increased to 90.1% and the parameter associated with target presence was significant (p < .001). Its value was 202 ms, indicating that, in addition to generalized slowing, older adults exhibit a constant delay of 202 ms on target-absent trials.

Because younger adults were more error-prone in Session 9, particularly on target-absent trials, and in Session 12 on both target-present and target-absent trials, we calculated nonparametric measures of sensitivity (A') and bias (B'') for Sessions 1, 4, 5, 8, 9, 12, 13, and 16. Sensitivity ranged from .93. to .99 and was not related to age. Bias measures indicated that older adults were more conservative, particularly with larger display sizes.

Additionally, as seen in Fig. 1, error rates tend to increase for younger adults only. To assess these trends, we determined linear fits to each of the functions shown. The slopes were negative for older adults and positive for younger adults. These findings are consistent with our view that age differences in display size effects for target-absent RTs are probably the result of differences in criteria for terminating the trial, with younger adults terminating search too quickly.

	References

TOP Abstract Methods Results and Discussion General Discussion Appendix References

 

• Anandam B. T., Scialfa C. T., 1999. Aging and the development of automaticity in feature search. Aging, Neuropsychology, and Cognition 6:117-140.
• Brinley J. F., 1965. Cognitive sets, speed and accuracy of performance in the elderly. Welford A. T., Birren J. E., , ed.Behavior, aging and the nervous system 114-149. Charles C Thomas, Springfield, IL.
• Chun M. M., Wolfe J. M., 1996. Just say no: How are visual searches terminated when there is no target present?. Cognitive Psychology 30:39-78. [Medline]
• Egeth H., Virzi R., Garbart H., 1984. Searching for conjunctively defined targets. Journal of Experimental Psychology: Human Perception and Performance 10:32-39.
• Fisk A. D., Hertzog C., Lee M. D., Rogers W. A., Anderson-Garlach M., 1994. Long-term retention of skilled visual search: Do young adults retain more than old adults?. Psychology and Aging 9:206-215. [Medline]
• Fisk A. D., Rogers W. A., 1991. Toward an understanding of age-related memory and visual search effects. Journal of Experimental Psychology: General 120:131-149.
• Fisk A. D., Rogers W. A., Cooper B. P., Gilbert D. K., 1997. Automatic category search and its transfer: Aging, type of search, and level of learning. Journals of Gerontology: Psychological Sciences and Social Sciences 52B:91-102.
• Friedman-Hill S., Wolfe J. M., 1995. Second-order parallel processing: Visual search for the odd item in a subset. Journal of Experimental Psychology: Human Perception and Performance 21:531-551.
• Kramer A. F., Strayer D. L., Buckley J., 1990. Development of transfer and automatic processing. Journal of Experimental Psychology: Human Perception & Performance 16:505-522.
• Kray, J., & Lindenberger, U. (1998, April). Age differences in internal cognitive control: The role of coordination and reconfiguration during task switching. Paper presented at the Cognitive Aging Conference, Atlanta, GA.
• Kutas M., Iragui V., 1998. The N400 in a semantic categorization task across 6 decades. Electroencephalography & Clinical Neurophysiology: Evoked Potentials 108:456-471.
• MacKay D. G., Abrams L., 1996. Language, memory, and aging: Distributed deficits and the structure of new-versus-old connections. Birren J. E., Schaie K. W., , ed.Handbook of the psychology of aging 4th ed. 25-265. Academic Press, San Diego, CA.
• Madden D. J., Nebes R. D., 1980. Aging and the development of automaticity in visual search. Developmental Psychology 16:377-384.
• Plude D., Doussard-Roosevelt J. A., 1989. Aging, selective attention, and feature integration. Psychology and Aging 1:4-10.
• Pomplun, M., Reingold, W. M., Shen, J., & Williams, D. E. (2000). The area activation model of saccadic selectivity in visual search. In Proceedings of the 22nd Annual Conference of the Cognitive Science Society (pp. 375–380). Mahwah, NJ: Erlbaum.
• Rogers W. A., 1992. Age differences in visual search: Target and distractor learning. Psychology and Aging 7:526-535. [Medline]
• Rogers W. A., Fisk A. D., 1991. Are age differences in consistent-mapping visual search due to feature learning or attention training?. Psychology and Aging 6:542-550. [Medline]
• Salthouse T. A., 1982. Adult cognition: An experimental psychology of human aging Springer-Verlag, New York.
• Schneider W., Shiffrin R., 1977Psychological Review, 84, 1–66.and attention. Controlled and automatic human information processing: I. Detection search:Psychological Review, 84, 1–66.and attention.
• Scialfa C. T., Jenkins L., Hamaluk E., Skaloud P., 2000. Aging and the development of automaticity in conjunction search. Journal of Gerontology: Psychological Sciences 55B:P27-P46. [Abstract]
• Scialfa C. T., Joffe K. M., 1997. Age differences in feature and conjunction search: Implications for theories of visual search and generalized slowing. Aging, Neuropsychology, and Cognition 4:1-21.
• Scialfa C. T., Joffe K. M., 1998. Response times and eye movements in feature and conjunction search as a function of target eccentricity. Perception and Psychophysics 60:1067-1082. [Medline]
• Shen J., Reingold E. M., Pomplun M., 2000. Distractor ratio influences patterns of eye movements during visual search. Perception 29:241-250. [Medline]
• Siakaluk, P. D., Ho, G., & Scialfa, C. T. (1999, July). Feature-based selection in triple-conjunction search. Paper presented at the 9th Annual Conference of the Brain, Behaviour, and Cognitive Science Society, Edmonton, Canada.
• Steinman S. B., 1987. Serial and parallel search in pattern vision. Perception 16:389-398. [Medline]
• Strayer D. L., Kramer A. F., 1994a. Aging and skill acquisition: Learning–performance distinctions. Psychology and Aging 9:589-605. a. [Medline]
• Strayer D. L., Kramer A. F., 1994b. Strategies and automaticity: I. Basic findings and conceptual framework. Journal of Experimental Psychology: Learning, Memory, and Cognition 20:318-341. b.
• Strayer D. L., Kramer A. F., 1994c. Strategies and automaticity: II. Dynamic aspects of strategy adjustment. Journal of Experimental Psychology: Learning, Memory, and Cognition 20:342-365. c.
• Theeuwes J., Kooi F. L., 1994. Parallel search for a conjunction of contrast polarity and shape. Vision Research 34:3013-3016. [Medline]
• Townsend J. T., 1972. Some results concerning the identifiability of parallel and serial processes. British Journal of Mathematical and Statistical Psychology 25:168-199.
• Treisman A., Sato S., 1990. Conjunction search revisited. Journal of Experimental Psychology: Human Perception and Performance 16:459-478.
• Williams D. E., Reingold E. M., 2001. Preattentive guidance of eye movements during triple conjunction search tasks: The effect of feature discriminability and saccadic amplitude. Psychonomic Bulletin and Review 8:476-488.
• Williams D. E., Reingold E. M., Moscovitch M., Behrmenn M., 1997. Patterns of eye movements during parallel and serial visual search tasks. Canadian Journal of Experimental Psychology 51:151-164.
• Wolfe J. M., 1994. Guided search 2.0: A revised model of visual search. Psychological Bulletin and Review 1:202-238.
• Zelinsky G. J., Sheinberg D. L., 1997. Eye movements during parallel–serial visual search. Journal of Experimental Psychology: Human Perception and Performance 23:244-262.

This article has been cited by other articles:

				K. Ball, J. D. Edwards, and L. A. Ross The Impact of Speed of Processing Training on Cognitive and Everyday Functions J. Gerontol. B. Psychol. Sci. Soc. Sci., June 1, 2007; 62(suppl_Special_Issue_1): 19 - 31. [Abstract] [Full Text] [PDF]

				P. E. Scalf, S. J. Colcombe, J. S. McCarley, K. I. Erickson, M. Alvarado, J. S. Kim, R. P. Wadhwa, and A. F. Kramer The Neural Correlates of an Expanded Functional Field of View J. Gerontol. B. Psychol. Sci. Soc. Sci., June 1, 2007; 62(suppl_Special_Issue_1): 32 - 44. [Abstract] [Full Text] [PDF]

				M. Grahame, J. Laberge, and C. T. Scialfa Age Differences in Search of Web Pages: The Effects of Link Size, Link Number, and Clutter Human Factors: The Journal of the Human Factors and Ergonomics Society, January 1, 2004; 46(3): 385 - 398. [Abstract] [PDF]

				L. C. McPhee, C. T. Scialfa, W. M. Dennis, G. Ho, and J. K. Caird Age Differences in Visual Search for Traffic Signs During a Simulated Conversation Human Factors: The Journal of the Human Factors and Ergonomics Society, January 1, 2004; 46(4): 674 - 685. [Abstract] [PDF]

This Article Abstract Full Text (PDF) Services Download to citation manager Citing Articles Citing Articles via HighWire PubMed PubMed Citation