Differential Age Effects for Case and Hue Mixing in Visual Word Recognition

Transcription

1 Psychology and Aging Copyright 2002 by the American Psychological Association, Inc. 2002, Vol. 17, No. 4, /02/$5.00 DOI: // Differential Age Effects for Case and Hue Mixing in Visual Word Recognition Philip A. Allen The University of Akron Karen E. Groth Case Western Reserve University Jeremy W. Grabbe The University of Akron Albert F. Smith Cleveland State University Jody L. Pickle Kent State University David J. Madden Duke University Medical Center The authors compare older adults lexical-decision data with younger adults data reported in P. Allen, A. F. Smith, et al. (2002). On the basis of their work, it was proposed that consistent-case words would be processed by the faster holistic (magnodominated) stream, but that mixed-case words would be processed by the slower analytic (interblob-dominated or blob-dominated) streams. Hue mixing was predicted to have no effect on consistent-case performance, but mixed-hue/mixed-case words were predicted to be recognized faster than monochrome/mixed-case words. Younger adults showed the predicted results, but older adults did not. These results suggest that holistic central processes are maintained, but that older adults exhibited an analytic decrement. In this article, we test predictions about the effect of age on word recognition for a theory of visual word recognition that assumes that words are processed both as single units (holistically) and as component parts (letter by letter; analytically). Our guiding hypothesis is that older adults are more biased than younger adults toward the use of larger processing units. Older adults show evidence of a holistic processing bias (e.g., Allen & Madden, 1989; Allen, Madden, & Crozier, 1991; Spieler & Balota, 2000); it is plausible that age-related slowing (Cerella, 1985; Madden, 2001; Salthouse, 1996) and increases in neural noise (Allen, Kaufman, Smith, & Propper, 1998a, 1998b; Welford, 1958) might be offset by processing larger units. We present evidence from earlier research suggesting that older adults partially compensate for overall age-related slowing and increased neural noise by processing information more efficiently. We propose that older adults exploit the decision complexity advantage better than do younger adults, making a few relatively complex decisions more Philip A. Allen and Jeremy W. Grabbe, Department of Psychology, The University of Akron; Albert F. Smith, Department of Psychology, Cleveland State University; Karen E. Groth, Department of Psychology, Case Western Reserve University; Jody L. Pickle, Department of Psychology, Kent State University; David J. Madden, Center for the Study of Aging and Human Development, Duke University Medical Center. This research was supported by National Institutes of Health/National Institute on Aging Grant AG We acknowledge useful comments from Christopher Chase and Michael Cortese on drafts of this article. Correspondence concerning this article should be addressed to Philip A. Allen, 364 Arts and Science Building, Department of Psychology, The University of Akron, Akron, Ohio paallen@uakron.edu efficiently than a larger number of simpler decisions (Wickens & Hollands, 2000). First, we describe basic findings concerning apparent changes in visual word recognition with age. We then discuss a model of word recognition that proposes that recognition results from a race between information carried by two different channels: a holistic channel and an analytic channel. We argue that two such channels appear necessary to account for behavioral data concerning word recognition and, particularly, changes in word recognition with age. Finally, we discuss the plausible mapping of this two-channel model onto a multichannel model of the primate visual system. Visual Word Recognition Previous research has shown that older adults retain their ability to process words holistically, as whole units, rather than analytically, as component letters (Allen & Madden, 1989; Allen et al., 1991; Allen, Madden, Weber, & Groth, 1993; Spieler & Balota, 2000). However, many of these studies have also shown that older adults exhibit a decrement in processing words analytically (e.g., Allen & Madden, 1989; Allen et al., 1991). In this article, we augment these findings with results from three new experiments and present a preliminary model of cognitive development that accounts for why older adults appear to be more holistically biased than younger adults. Holistically Biased Hybrid Model for Visual Word Recognition We assume that visual word encoding during reading involves a race between whole-word and letter-by-letter processes (Allen & Emerson, 1991; Allen, Madden, & Slane, 1995; Healy, Conboy, & 622

2 DIFFERENTIAL AGE EFFECTS IN VISUAL WORD RECOGNITION 623 Drewnowski, 1987; Johnson, Allen, & Strand, 1989). The unitization model of letter identification (Healy, 1976) proposes that larger units, such as words, conceal smaller units, such as letters. The holistically biased hybrid model (Allen & Emerson, 1991; Allen & Madden, 1990; Johnson et al., 1989) is a revision and extension of the unitization model. According to the hybrid model, word- and letter-level information race to a central processor that bases recognition on the first-arriving data. The speed with which word-level information becomes available to the central processor increases with word frequency, whereas the speed with which letter-level information becomes available is independent of word frequency (although it is assumed that the letter-level channel is adaptive so that it processes more frequently encountered letter sequences faster than it processes uncommon ones). The central processor is capacity limited and responsive to just one source of information at any time (Allen & Madden, 1990). Thus, according to the hybrid model, recognition of very-frequent words is based on information from the word-level channel because this channel outputs a code to the central processor before the letter-level channel does. For verylow-frequency words, information from the letter-level channel is available to the central processor before word-level information is. For some intermediate level of word frequency, word-level information arrives just before letter-level information and access to letter-level information is delayed by the processing of word-level information. Visual Features or Spatial Frequency Filtering Although most visual word recognition researchers have proposed that perceptual objects (e.g., words) are perceived as collections of features (e.g., Adams, 1979; McClelland & Rumelhart, 1981), it is assumed for the hybrid model that the basic units of analysis are the spatial-frequency patterns of whole words and of component letters. Two classes of evidence support a recognition model based on spatial frequency. First, Gervais, Harvey, and Roberts (1984) showed that letter confusions are better predicted by a spatialfrequency model than by a feature model. (We are unaware of any comparison of a spatial-frequency model to a feature model for whole words.) Second, there is considerable evidence that, at least at the retinal and lateral geniculate nucleus (LGN) levels, visual information processing involves spatial frequency filtering rather than feature analysis (e.g., Van Essen, Anderson, & Felleman, 1992). Age Differences in Visual Word Recognition Previous studies of the relation of the speed of visual word recognition to aging suggest that older adults are slower than younger adults in overall response time (RT), but that older adults are not appreciably slower than younger adults in lexical access. Studies of the relation between age and lexical-decision performance have typically found that the effect of word frequency is comparable across age groups (Allen et al., 1991, 1993; Allen et al., 2002; Allen, Sliwinski, Bowie, & Madden, 2002; Tainturier, Trembley, & Lecours, 1989, 1992). Some naming studies have found a slightly larger effect of word frequency for older adults than for younger adults (Balota & Ferraro, 1993, 1996), but other studies have found no such difference (Allen, Cerella, Madden, Smith, & Lien, 2002). Finally, Balota, Pilotti, and Cortese (2001) found that older adults had slightly larger word frequency effects on a subjective rating task of words. Overall, then, there is mixed evidence for the hypothesis that older adults show a lexical-access decrement, although it does appear that somewhat more studies have found no age differences in word frequency effects (see Allen, Madden, & Slane, 1995). This implies that older adults relatively longer word recognition RTs are due to peripheralprocess decrements rather than to lexical-access decrements (Allen, Lien, et al., 2002; Allen et al., 1993; Allen, Sliwinski, & Bowie, 2002). As people age, they appear to become more biased toward holistic processing of words (Allen & Madden, 1989; Allen et al., 1991; Spieler & Balota, 2000). Spieler and Balota evaluated the predictability of naming RTs by regressing them on word frequency, word length, and orthographic neighborhood size (see also Balota & Spieler, 1998; Spieler & Balota, 1997). Word frequency, a property of the whole word, is a measure of how often the whole word unit occurs in English; word length (the number of separate letter units in a word) and neighborhood size (letter-level overlap across words) are letter-level properties (Spieler & Balota, 2000). Spieler and Balota found, for both older and younger adults, that word frequency was the best predictor of naming RT, and that its advantage over the other predictors was greater for older adults than for younger ones (see Spieler & Balota s Table 2). Other evidence for a holistic processing bias in older adults comes from a letter identification task in which experimental participants indicate whether a target letter shown by itself is the initial letter of a subsequently presented word (Allen & Madden, 1989; Allen et al., 1991). In lexical decision, the effect of word frequency on performance is essentially the same for older and younger adults, with word decision RT decreasing as word frequency increases. However, in letter identification, the effect of word frequency differs for older and younger adults: Younger adults detect the initial letter of medium-high-frequency words more slowly than they detect the initial letter of very-high-, high-, and low-frequency words, whereas older adults exhibit the monotonic pattern characteristic of both age groups in lexical decision (Allen & Emerson, 1991; Allen & Madden, 1990; Allen et al., 1991). Older adults, when forced to access letter-level codes as in the letter identification task, appear to do so by segmenting wordlevel codes into letter-level codes (Allen et al., 1991), whereas younger adults appear to report the output of the letter-level channel. These data suggest that older adults are biased toward holistic processing even if such processing results in poorer performance, as when a task requires an individual to identify component letters of words. Although older adults take longer to process words, certain stages of processing, such as lexical access, show little or no slowing (Allen et al., 1993). This stage-specific sparing is due, at least in part, to older adults greater reliance on holistic processing, which is typically more efficient than analytic processing (Allen & Madden, 1989; Allen et al., 1991; Spieler & Balota, 2000). We discuss next the brain pathways that might yield these behavioral data and how aging might affect these pathways. In the following sections, we first discuss the neurologically plausible model of visual word recognition that was proposed by Allen, Smith, et al. (2002). Then, we discuss how predictions of

3 624 ALLEN ET AL. this model may be evaluated by manipulating the hue and case consistency of letter strings in a lexical-decision task. We then propose hypotheses about how aging affects the utilization of these pathways and their performance. Finally, we test these hypotheses by examining the effect of age in three experiments that extend studies reported by Allen, Smith, et al. Magnocellular and Parvocellular Channels in Visual Word Recognition Early work by Blakemore and Campbell (1969), Breitmeyer and Ganz (1976), and Ginsberg, Carl, Kabrisky, Hall, and Gill (1976) suggested that there are two different information-processing channels in the primate visual system. Building on this work, many vision researchers have assumed that human pattern perception involves parallel channels, each of which uses a series of spatialfrequency filters (e.g., Graham, Sutter, & Venkatesan, 1993; Sagi, 1990; Sutter, Sperling, & Chubb, 1995; Van Essen et al., 1992; Watt, 1991). Although much of this research has concentrated on ganglion-cell channels at the retinal level (Breitmeyer & Ganz, 1976), the LGN and cortical levels of visual information processing (Van Essen et al., 1992) have also been studied. Two visual pathways have been identified at the LGN level: the magnocellular (large-cell) pathway and the parvocellular (smallcell) pathway (Van Essen et al., 1992). These two LGN pathways project to three processing streams in the primary visual cortex: the magnodominated stream (MD), the blob-dominated stream (BD), and the interblob-dominated stream (ID; DeYoe & Van Essen, 1988; Van Essen & Anderson, 1995). ( Blobs are dark patches in Brodmann s Areas 17, 18, and 19, and interblobs are cells between the darker blob cells.) These studies converge on the following conclusions concerning the characteristic sensitivities of these channels: The MD stream is particularly sensitive to lowspatial-frequency stimuli, has a fast conduction rate, and responds to luminance differences but is not sensitive to hue differences. The BD stream is sensitive to high-spatial-frequency stimuli, has a slow conduction rate, is sensitive to hue particularly to color contrasts, that is, to hue variability within a stimulus and processes certain brightness, texture, and shape information (Livingstone & Hubel, 1987; Van Essen & Anderson, 1995). The ID stream is also sensitive to high-spatial-frequency stimuli that can be used to discriminate fine details, has a slow conduction rate, and although responsive to hue differences, the ID stream is less so than the BD stream (Van Essen & Anderson, 1995). This pathway has been hypothesized to process color, texture, and pattern information (Livingstone & Hubel, 1987). The inferotemporal cortex to which the ID and BD streams project has been implicated in object recognition by using high-resolution information; the MD stream has been implicated in the recognition of moving objects and low-contrast objects (Carlson, 1998; Van Essen & Anderson, 1995). Peterzell and Teller (2000) examined correlations between contrast sensitivity thresholds across seven levels of spatial frequency of gratings that were yellow black (i.e., monochromatic or luminance modulated) and seven levels of spatial frequency of gratings that were red green (i.e., chromatic modulated). Using principalcomponents analysis, Peterzell and Teller interpreted three principal components as representing a low-spatial-frequency monochrome channel, a high-spatial-frequency monochrome channel, and a chromatic channel. These channels appear to correspond to the MD, ID, and BD channels, respectively, of the multistream model of Van Essen and Anderson (1995). Application to Word Recognition As we summarized earlier, the hybrid model of visual word recognition (Allen et al., 1995) postulates two informationcarrying channels: a holistic channel and an analytic channel. The information carried by these channels races to a central processor. This model maps fairly straightforwardly onto the three-stream model of cortical pathways (Allen, Smith, et al., 2002; Van Essen & Anderson, 1995). Specifically, the holistic channel maps plausibly onto the MD stream, whereas the analytic channel maps plausibly onto the ID and BD streams (see Allen, Smith, et al., 2002). Each input channel attempts to form a word code by using the spatial frequency pattern of its basic unit of analysis: The achromatic MD stream uses low-spatial-frequency information, and the detail-sensitive ID stream and hue-sensitive BD stream use high-spatial-frequency information. However, the BD and ID streams are slower than the MD stream, so information in the MD stream typically wins the processing race (Allen, Smith, et al., 2002) and visual word recognition is generally based on holistic codes. This is consistent with Kelly s (1979) finding that lowspatial-frequency components were available before high-spatialfrequency components in visual information processing. A variant of this multistream model is that the processing streams are increasingly interactive at higher cortical levels. According to this view, low-spatial-frequency information is processed by the rapid MD stream, and this information is cascaded to the cortical areas that process the high-spatial-frequency information from the ID and BD streams. If the low-spatial-frequency information is sufficient for recognition, then a rapid response is made. Otherwise, low-spatial-frequency information is combined with the high-spatial-frequency information. However, activity in the parvocellular streams is hypothesized to inhibit laterally the magnocellular stream, and this lateral inhibition has predictable consequences for performance. A third possibility is that only the BD and ID streams the parvocellular streams are involved in word recognition, and that the MD stream is not. Livingstone and Hubel (1987) argued that the magnocellular pathway is sensitive to movement and spatial location, but is not differentially sensitive to objects or shapes, so that only the parvocellular pathway is involved in object recognition (although see Allen, Smith, et al., 2002; Van Essen et al., 1992). Empirical Evaluation As we have indicated, there is considerable evidence that older adults are more holistically biased in visual word recognition than are younger adults (Allen et al., 1993; Spieler & Balota, 2000). There is also evidence that analytic processing by older adults is less effective than that by younger adults (Allen & Madden, 1989; Allen et al., 1991). In the experiments reported in this article, we sought to ascertain whether this holistic bias/analytic decrement is manifested as a bias by older adults toward processing in the magnocellular channel.

4 DIFFERENTIAL AGE EFFECTS IN VISUAL WORD RECOGNITION 625 We report data from older adults in three experiments that tested predictions of the multistream model of visual word recognition. These predictions are based on assumptions about the specific sorts of information processed by different visual processing streams and the speeds with which those streams operate. These were lexical-decision experiments: The letter strings used as stimuli were either words or nonwords, and participants were to respond to this stimulus property. Two characteristics of these strings were manipulated: (a) whether the cases of letters in stimulus items were consistent or mixed (with alternate letters in lower- and uppercase) and (b) whether the hues of letters were consistent or mixed (with the hues of letters alternating). In our experiments, we evaluated three basic predictions derived from the multistream model. First, because consistent-case words will be processed by the fast MD stream, but mixed-case strings will be processed by the slower BD and ID streams, responses to consistent-case words will be faster than those to mixed-case words. We hypothesize that mixed-case strings are processed analytically because these stimuli present an unfamiliar spatial frequency pattern to the magnocellular pathway (Allen et al., 1993) and that there is an analytic-processing decrement for older adults. Therefore, we predict that this mixed-case disadvantage will be larger for older adults than for younger adults. Second, among mixed-case stimuli, mixed-hue items will be processed by both the ID and BD streams, whereas items without intraitem hue variation will be processed by only the ID stream (i.e., the BD stream is activated primarily by hue contrasts at an edge). Because the BD and ID streams operate at similar rates, responses to mixed-case/mixed-hue strings should, on average, be faster than responses to mixed-case/monochrome strings; this speed advantage of two channels with similar processing times over a single channel was termed statistical facilitation by Raab (1962). We hypothesize that older adults have less efficient analytic (parvocellular) processing than younger adults. Therefore, we predict that any such facilitation will be smaller for older adults than for younger adults. In contrast, a model that hypothesizes generalized slowing would predict that the magnitude of RT differences across task conditions would be greater for older adults than for younger adults. That is, the generalized slowing model would predict older adults latencies should increase as a linear combination of younger adults latencies as a function of task complexity (Cerella, 1985; Madden, 2001). We test whether the results from the present experiments conform to the predictions of a general slowing model through the use of an RT procedure recommended by Madden, Pierce, and Allen (1992). Third, if the processing streams are increasingly interactive across stages of processing (see DeYoe & Van Essen, 1988; Van Essen & Anderson, 1995), then for mixed-case strings, responses to mixed-hue words should be faster than responses to monochrome words because both ID and BD streams are hypothesized to be activated in the former condition, but only the ID stream is hypothesized to be activated in the latter condition. However, for consistent-case strings, responses to mixed-hue words should be slower than responses to monochrome words. In such a system, the low-spatial-frequency information processed by the MD stream is cascaded to the ID and BD streams, and activation of these parvocellular streams laterally inhibits the MD stream at the cortical level. Such inhibition should be greater when both parvocellular streams are activated, as during processing of a mixed-hue string, than when just the ID stream is activated, as during processing of a string that contains no variation in hue (Allen, Smith, et al., 2002). If older adults have less efficient parvocellular processing than do younger adults, resulting in less cortical-level lateral inhibition for older adults, then for lowercase strings, older adults should exhibit a smaller mixed-hue disadvantage than younger adults. Participants General Method Younger adult participants were undergraduates from Cleveland State University (Experiments 1 and 2) and the University of Akron (Experiment 3) who participated for credit toward the research-exposure requirement of a course. Older adult participants were healthy community-dwelling individuals, each of whom was paid $20 for participation. All participants were native speakers of English or had spoken English as a primary language for at least 5 years. All participants were screened to have near visual acuity of at least 20/40 and for red green color blindness. Participants reported years of education and completed the Wechsler Adult Intelligence Scale Revised (WAIS R; Wechsler, 1981) Vocabulary and Digit Symbol Substitution Task subtests. Stimulus Materials Two lists of 384 items, each consisting of 192 words and 192 nonwords, were used in Experiments 1 3; these full lists are presented in Allen, Smith, et al. (2002, Appendix A). Each set of 192 words (to which we refer to as Sets A and B, respectively) included 24 from each of the eight classes defined by crossing four levels of word frequency with two levels of length. The four word frequency categories, defined by number of occurrences per million in the Kucera and Francis (1967) norms, were very high (255 1,702 occurrences per million), medium high ( occurrences per million), low (40 52 occurrences per million), and very low (1 5 occurrences per million). The two levels of length were five and six letters. From each word, a corresponding nonword was generated by transforming one letter. For example, one five-letter low-frequency word was loose, and its corresponding nonword was loost. For half of the participants in each experiment, the stimuli consisted of the Set A words and the nonwords generated from the Set B words; for the other half of the participants, the stimuli consisted of the Set B words and the nonwords generated from the Set A words. Stimulus Display and Response Collection Personal computers equipped with noninterlaced monitors were used to collect data. Stimulus presentation and timing were controlled by Micro Experimental Laboratory software (Schneider, 1988). Participants sat approximately 50 cm from the display monitor. All stimulus strings were centered on a black background. At a viewing distance of 50 cm, each letter subtended a visual angle of 0.28 horizontally and 0.56 vertically, such that the visual angle subtended by a six-letter string was 1.85.

5 626 ALLEN ET AL. In various experiments, the letters of stimulus strings were bluish-white, red, and green. The luminance of stimulus strings varied considerably with hue. (The implications of this variation, and whether a variance necessitates concern, are discussed at appropriate points.) As measured by a Minolta LS-100 luminance meter, for white six-letter displays, the luminances of lowercase, mixed-case, and uppercase strings were 9.0 cd/m 2, 11.1 cd/m 2, and 11.8 cd/m 2, respectively; for green six-letter displays, the luminances of lowercase and mixed-case strings were 6.3 cd/m 2 and 7.9 cd/m 2, respectively; and for red six-letter displays, the luminances of lowercase and mixed-case strings were 5.5 cd/m 2 and 7.4 cd/m 2, respectively. For mixed-case/mixed-hue displays, lowercase/mixed-hue displays, and uppercase/mixed-hue displays with six letters, the luminances were 7.2 cd/m 2, 9.1 cd/m 2, and 9.6 cd/m 2, respectively. The luminance for the black background was 1.6 cd/m 2. The color palette settings used for red were r 63, g 48, and b 63; for green they were r 33, g 63, and b 40. The Michelson contrast values for the aforementioned conditions were as follows: uppercase/mixed hue.71, uppercase/monochrome (blue).76, mixed case/mixed hue.70, mixed case/monochrome (green).66, mixed case/monochrome (red).64, mixed case/monochrome (blue).75, lowercase/ mixed hue.64, lowercase/monochrome (green).59, lowercase/monochrome (red).55, and lowercase monochrome (blue).70. Participants responded by using the index and middle fingers of their right hand to operate the left and right arrow keys located in the lower right-hand corner of the keyboard. Procedure In each experiment, for each presented item, participants were to respond word if the letter string was an English word and to respond nonword if otherwise. (Allen, Smith, Lien, Weber, and Madden, 1997, showed that the typical Cleveland State University student knew the preponderance of even the very-low-frequency words, as indicated by their performance on an untimed lexicaldecision task after which participants selected the correct definition of the letter string it they decided that it was a word.) Responses were to be made as rapidly as possible while maintaining accuracy. Stimuli were presented until a response was made. RTs were measured from the stimulus onset of a participant s press to a response key. The assignment of responses to response keys was balanced across participants. Data Analysis In the Results sections of the various experiments, we first present analyses of responses to words and then analyses of responses to all stimuli. The predictions of the multistream model are about the relative speeds of word responses for words presented in various case-consistency and hue-consistency formats. However, to evaluate the general slowing model requires analyses of responses to all items. Experiment 1 Experiment 1 was designed to test both the basic predictions of the multistream model and, particularly, the predictions concerning the relationship of performance to age. As we have discussed, we make three predictions for the multistream model and three additional predictions about how performance depends on age. The multistream model predicts that color mixing will affect responding when the MD stream is prevented from outputting a code, as when the cases of letters are mixed: lowercase strings will be processed by the MD stream regardless of their hues, mixedcase/monochrome strings will be processed by the ID stream, and mixed-case/mixed-hue strings will be processed by either or both of the BD and ID streams (Allen, Smith, et al., 2002). Although the BD and ID streams have similar processing rates, RTs for mixedcase/mixed-hue stimuli should depend on the faster of these streams, so responses to mixed-case/mixed-hue strings should be faster, on average, than responses to mixed-case/monochrome stimuli (see Allen, Smith, et al., 2002; Raab, 1962). Crossing case consistency with hue consistency in a lexical-decision experiment permits evaluation of these predictions. This design also permits us to evaluate predictions of the multistream model variant, which proposes that processing streams are progressively interactive as information progresses from subcortical areas (e.g., the retinas and LGN) to cortical areas. This variant, too, is a holistically biased/multiple-channel model: It predicts faster responding to mixed-case/mixed-hue stimuli than to mixedcase/monochrome stimuli. However, the variant also predicts slower responding to lowercase/mixed-hue stimuli than to lowercase/monochrome stimuli: According to this model, information from lowercase stimuli is processed in the MD stream and is then cascaded to the ID and BD streams. If the parvocellular pathways inhibit the MD stream, there will be more inhibition when both the ID and BD streams are active (during processing of lowercase/ mixed-hue stimuli) than when only the ID stream is active (during processing of lowercase/monochrome stimuli). Several additional predictions pertain specifically to performance differences among age groups. Because older adults are hypothesized to be more holistically biased than younger adults, older adults should exhibit a larger mixed-case disadvantage than younger adults. In addition, if the interactive variant model is correct, the performance decrement for lowercase/mixed-hue stimuli compared with lowercase/monochrome stimuli should be smaller for older adults than for younger adults. Perhaps the most important prediction of the hybrid model is that for mixed-case words, for which recognition depends on processing by one or more of the parvocellular streams, older adults will exhibit a smaller benefit from hue mixing than younger adults. Such a finding would provide precise evidence that older adults process letter-level information less efficiently than younger adults. We report data collected from older adults along with data collected from younger adults and reported by Allen, Smith, et al. (2002, Experiment 1). Method Participants. Sixty-seven individuals participated: thirty-three were younger adults who ranged in age from 18 to 38 years, with a mean of 23.5 years, and 34 were older adults who ranged in age from 60 to 87 years, with a mean of 70.6 years. Relative to older adults, younger adults showed higher Digit Symbol scores (older 47.6, younger 69.0), t(65) 9.23, p.001, lower Vocabulary test scores (older 44.4, younger 38.9), t(65) 2.06,

6 DIFFERENTIAL AGE EFFECTS IN VISUAL WORD RECOGNITION 627 p.05, and had completed fewer years of education (older 15.0 years, younger 13.9 years), t(65) 2.01, p.05. Design and procedure. Four stimulus conditions were defined by whether stimulus strings were lowercase or mixed case and whether they were monochrome or mixed hue. These two stimulus characteristics were combined orthogonally. In mixed-case strings, the first letter was always a lowercase letter, and the cases of letters alternated systematically. In monochrome strings, all letters were bluish-white. In mixed-hue strings, the hues of the letters alternated between red and green: For half of the participants, the first letter was always red; for the other half, the first letter was always green. The two 384-item stimulus sets described in the General Method section, each consisting of 192 words and 192 nonwords, were used. Ninety-six different strings (48 words and 48 nonwords) were presented in each of the four stimulus conditions defined by case consistency and hue consistency. For each stimulus set, we used two different assignments of items to the four stimulus conditions. Trial presentation was blocked by stimulus condition; the order of the four 96-trial blocks was approximately counterbalanced across the participants in each age group. Results Table 1 shows, for each type of lexical item, mean RTs and error rates for younger and older adults in each of the four stimulus conditions. Words. RTs for correct responses to word trials were analyzed by using a 2 (age group: younger vs. older) 2 (case consistency: lowercase vs. mixed case) 2 (hue consistency: monochrome vs. mixed hue) mixed analysis of variance (ANOVA). On average, older adults responded more slowly than younger adults (mean Table 1 Mean Response Times (RTs; in Milliseconds) and Error Rates (With Standard Deviations) in Experiment 1; by Age Group, Case Consistency, Hue Consistency, and Lexical Type Condition Lowercase Mixed-case M SD M SD Younger adults (n 33) Mixed-hue Word RT Word error Nonword RT Nonword error Monochrome Word RT Word error Nonword RT Nonword error Older adults (n 34) Mixed hue Word RT Word error Nonword RT , Nonword error Monochrome Word RT Word error Nonword RT , Nonword error RT: older adults 805 ms, younger adults 685 ms), F(1, 65) 16.14, p.001, and responses to mixed-case words were slower than responses to lowercase words (mixed case 809 ms, lowercase 683 ms), F(1, 65) , p.001. The Age Hue Consistency Case Consistency interaction was statistically significant, F(1, 65) 4.37, p.05: For mixedcase stimuli, the mixed-hue advantage (the amount by which responses to mixed-hue stimuli were faster than those to monochrome stimuli) was smaller for older adults (12 ms) than for younger adults (78 ms), whereas for lowercase stimuli, there was a mixed-hue disadvantage that was relatively consistent in size across age groups (younger adults 26 ms, older adults 11 ms). Separate Age Hue Consistency analyses at each level of case consistency were carried out: For mixed-case words, the Age Hue Consistency interaction was statistically significant, F(1, 65) 4.27, p.05, but for lowercase words, it was not ( p.40). An analysis of error rates showed main effects for case consistency, F(1, 65) , p 001 (mean percentage error: lowercase 6.8%, mixed case 18.4%), and hue consistency, F(1, 65) 18.69, p.001 (mixed hue 10.9%, monochrome 14.3%); an Age Case Consistency interaction, F(1, 65) 5.18, p.05 (younger lowercase 7.4%, younger mixed case 16.7%, older lowercase 6.1%, and older mixed case 20.2%); and a Case Consistency Hue Consistency interaction, F(1, 65) 28.39, p.001 (mixed case/mixed hue 14.6%, mixed case/monochrome 22.2%, lowercase/ mixed hue 5.7%, and lowercase/monochrome 5.2%). All stimuli. An analysis of RTs to all stimuli (see Table 1 for means) showed main effects for age, F(1, 65) 16.20, p.001 (younger 738 ms, older 887 ms), stimulus type, F(1, 65) , p.001 (words 738 ms, nonwords 879 ms), and case consistency, F(1, 65) , p.001 (lowercase 721 ms, mixed case 903 ms). The were also significant interactions involving Age Stimulus Type, F(1, 65) 4.98, p.05 (younger: words 685, nonwords 792; older: words 805 ms, nonwords 963 ms), Stimulus Type Case Consistency, F(1, 65) 51.22, p.001 (words: lowercase 683 ms, mixed case 809 ms; nonwords: lowercase 760 ms, mixed case 998 ms), Age Stimulus Type Case Consistency, F(1, 65) 10.78, p.01 (older adults showed a relatively larger mixed-case disadvantage for nonwords compared with words; see Table 1), Case Consistency Hue Consistency, F(1, 65) 8.58, p.01 (mixed case: mixed hue 878 ms, monochrome 929 ms; lowercase: mixed hue 734 ms, monochrome 709 ms), and Age Case Consistency Hue Consistency, F(1, 65) 4.22, p.05 (older adults showed a relatively smaller mixed-case/mixed-hue advantage than did younger adults; see Table 1). No other effects were significant ( p.05). For the analogous analysis of error rates, a similar pattern of results was observed (see Table 1). There were main effects for stimulus type, F(1, 65) 9.44, p.01, case consistency, F(1, 65) , p.001, and hue consistency, F(1, 65) 5.73, p.05. There were also significant interactions involving Stimulus Type Case Consistency, F(1, 65) 4.50, p.05, Stimulus Type Hue Consistency, F(1, 65) 9.36, p.01, and Stimulus Type Case Consistency Hue Consistency, F(1, 65) 27.05, p.001. The three-way interaction occurred because nonwords did not show a mixed-case/mixed-hue advantage, but words did (see Table 1).

7 628 ALLEN ET AL. Slowing analyses. Madden (2001) has suggested that researchers routinely test generalized slowing as a null hypothesis in experiments on aging. We do so by using the procedure outlined by Madden et al. (1992). First, a general slowing function is obtained by using a linear combination of younger adults conditional means to predict older adults conditional means. Next, younger adults conditional means are transformed by using this slowing function. If older adults latencies increase relative to younger adults latencies simply as a function of increased complexity, then all interactions with age and task complexity should be eliminated after younger adults conditional means are transformed by using the best fitting slowing function (Madden et al., 1992). After carrying out this transformation, we found Age Task Complexity interactions. Specifically, the Age Stimulus Type Case Consistency interaction, F(1, 65) 6.36, p.05 (older adults showed a relatively larger mixed-case disadvantage for nonwords compared with words), and the Age Case Consistency Hue Consistency interaction, F(1, 65) 5.25, p.05 (older adults showed a relatively smaller mixed-case/mixed-hue advantage than did younger adults), remained significant. On the basis of these results, the hypothesis of generalized slowing is rejected for these data. Discussion Older adults were slower than younger adults, and older adults showed larger case-mixing effects (in error analyses) than younger adults did. This finding is consistent with the idea that older adults have relatively more difficulty than younger adults in processing analytic letter-level codes (e.g., Allen et al., 1991, 1993; Spieler & Balota, 2000). The performance results for mixed-case stimuli are most critical. For mixed-case stimuli, older adults showed a significantly smaller benefit from hue mixing than did younger adults: For younger adults, but not for older adults, responses to mixed-case words were significantly faster with the mixed-hue than with the monochrome presentation. This pattern is consistent with the hypothesis that for mixed-case stimuli, the MD stream did not output a code and that the ID and BD streams were involved in a processing race for mixed-hue but not for monochrome stimuli. Facilitation for mixed-case/mixed-hue stimuli (relative to mixed-case/monochrome stimuli) was evident for younger adults, but not appreciably so for older adults. In addition, the error data showed a larger mixed-case disadvantage for older adults than for younger adults, indicating that relative to younger adults, older adults have an analytical processing decrement. For lowercase stimuli, the effect of hue consistency was similar for younger and older adults: Both age groups showed a small, but statistically significant, mixed-hue disadvantage. In contrast to the results for mixed-case stimuli, participants took longer to respond to mixed-hue/lowercase words than to monochrome/lowercase words. Lowercase words are likely processed holistically by the MD stream (Allen, Smith, et al., 2002). These results suggest that the age groups do not differ in holistic processing. The observed hue-mixing effect on RTs for lowercase stimuli is consistent with a model in which the processing streams are progressively interactive. According to this model, the magnocellular and parvocellular pathways do not interact appreciably at the retinal and LGN levels, but interact through lateral inhibition at the cortical level (Van Essen et al., 1992; Van Essen & Anderson, 1995). The finding, for lowercase words, that the effect of hue consistency was the same for older and younger adults suggests that both age groups show similar patterns of cortical lateral inhibition. This inference, if correct, is somewhat paradoxical because it would mean that low-level parvocellular processing is compromised by increased adult age, but that high-level cortical parvocellular processing is not. The replicability of this effect is evaluated in Experiment 2. We used the procedure proposed by Madden et al. (1992) to determine the viability of a general slowing model as an account of the observed age differences across task complexity. After carrying out the appropriate transformations, we observed a statistically significant Age Task Complexity interaction. Thus, a singlefactor generalized slowing model does not account well for these data (although the relatively high error rates for mixed-case stimuli suggest that inferences about generalized slowing from these RT means should be made with caution). Experiment 2 We conducted Experiment 2 to replicate the critical results of Experiment 1 but with different luminance arrangements. For lowercase words in Experiment 1, responses to mixed-hue items were slower than responses to monochrome items. In contrast, for mixed-case words, responses to mixed-hue items were faster than those to monochrome items (significantly so for younger adults, although not for older adults). However, the lowercase stimuli used in Experiment 1 were less luminant than mixed-case stimuli, and there may be some luminance threshold that must be exceeded for the BD stream to be consistently activated. Thus, one might account for the results of Experiment 1 by arguing that lowercase/mixed-hue items were insufficiently luminant to consistently activate the BD stream, but that mixedcase/mixed-hue items were, at least, in younger adults. We evaluated this luminance account in Experiment 2, the design of which was identical to that of Experiment 1 except that the consistent-case stimuli were uppercase rather than lowercase so that the luminance of consistent-case stimuli exceeded that of mixed-case stimuli. If the processing of lowercase/mixed-hue stimuli suffered in Experiment 1 because lowercase stimuli are less luminant than mixed-case stimuli, then in Experiment 2, the mixed-hue advantage should be larger for uppercase stimuli than for mixed-case stimuli. Alternatively, replicating the pattern of results of Experiment 1 with reversed luminance arrangements would discredit the luminance account of the results of Experiment 1. The same predictions concerning the effect of age that were made for Experiment 1 apply to Experiment 2 as well. There should be an Age Case Consistency interaction, with older adults showing a larger mixed-case disadvantage. There should also be an Age Case Consistency Hue Consistency interaction, for mixed-case words, with older adults showing a smaller benefit of hue mixing (relative to mixed-case/monochrome presentation) than younger adults. A larger overall mixed-case disadvantage for older adults than for younger adults would suggest that older adults have particular difficulty processing mixed-case words. The predicted three-way interaction would provide further support for the hypothesis that older adults do not show as much

8 DIFFERENTIAL AGE EFFECTS IN VISUAL WORD RECOGNITION 629 statistical facilitation as younger adults when both parvocellular pathways are presumed to be active. Method Participants. Eighty-four individuals participated: forty-two were younger adults who ranged in age from 17 to 43 years, with a mean of 22.3 years, and 42 were older adults who ranged in age from 61 to 82 years, with a mean of 71.0 years. Relative to older adults, younger adults showed higher Digit Symbol scores (older 50.6, younger 69.3), t(82) 8.79, p.001, lower Vocabulary scores (older 47.6, younger 38.9), t(82) 4.04, p.001, and had completed fewer years of education (older 15.6 years, younger 14.4 years), t(82) 2.37, p.05. Design and procedure. Four stimulus conditions were defined by whether stimulus strings were uppercase or mixed case and by whether they were monochrome or mixed hue. These two stimulus characteristics were combined orthogonally. The construction of mixed-case and mixedhue strings was as described for Experiment 1. The two 384-item stimulus sets described in the General Method section, each consisting of 192 words and 192 nonwords, were used. Ninety-six different items (48 words and 48 nonwords) were presented in each of the four stimulus conditions defined by case consistency and hue consistency. For each stimulus set, we used two different assignments of items to the four stimulus conditions. Trial presentation was blocked by stimulus condition; the order of the four 96-trial blocks was approximately counterbalanced across the participants in each age group. Results Table 2 shows, for each type of lexical item, mean RTs and error rates for younger and older adults in each of the four stimulus conditions. Words. For mixed-case stimuli, responses were faster for mixed-hue strings than for monochrome strings, whereas for uppercase stimuli, responses were slower for mixed-hue strings than for monochrome strings. The interaction of case consistency and hue consistency was statistically significant, F(1, 82) 10.57, p.01 (mixed case: mixed hue 1,035 ms, monochrome 1,068 ms; uppercase: mixed hue 813 ms, monochrome 788 ms). Older adults were slower at responding to words than were younger adults, F(1, 82) 12.67, p.001 (mean RT: older 1,014 ms, younger 838 ms). The mixed-case disadvantage was larger for older adults than for younger adults; this was reflected in the significant Age Case Consistency interaction, F(1, 82) 6.23, p.05; for older adults, the mean disadvantage was 298 ms, whereas for younger adults, the mean disadvantage was 203 ms. However, the Age Case Consistency Hue Consistency interaction was not significant, F(1, 82) 2.24, p.14. Because we hypothesized that for mixed-case words, older adults would show a smaller mixed-hue advantage than younger adults, we analyzed mixed-case stimuli separately. The main effect for hue consistency, F(1, 82) 5.39, p.05, and the Age Hue Consistency interaction, F(1, 82) 4.37, p.05, were significant. In contrast, in a similar separate analysis of uppercase words, there was a significant main effect of age, F(1, 82) 6.53, p.05, but no Age Hue Consistency interaction, F(1, 82) 0.11, p.74. An analysis of error rates showed main effects of age, F(1, 82) 16.62, p.001 (mean percentage error: older 6.7%, younger 10.3%), case consistency, F(1, 82) 80.10, p.001 Table 2 Mean Response Times (RTs; in Milliseconds) and Error Rates (With Standard Deviations) in Experiment 2; by Age Group, Case Consistency, Hue Consistency, and Lexical Type Condition Uppercase Mixed-case M SD M SD Younger adults (n 42) Mixed-hue Word RT Word error Nonword RT , Nonword error Monochrome Word RT Word error Nonword RT , Nonword error Older adults (n 42) Mixed-hue Word RT , Word error Nonword RT 1, , Nonword error Monochrome Word RT , Word error Nonword RT , Nonword error (mixed case 11.0%, uppercase 6.0%), and hue consistency, F(1, 82) 48.70, p.001 (mixed hue 7.2%, monochrome 9.9%). No other effects were statistically significant ( ps.40). Separate analyses at each level of case consistency showed fewer errors for mixed-hue words than for monochrome words. For uppercase words, this result contrasts with the finding for RTs. For responses to uppercase words, the small, but significant, RT disadvantage for mixed-hue relative to monochrome items may have been offset by a small, but significant, accuracy advantage for mixed-hue relative to monochrome items. All stimuli. A combined analysis of RTs to all stimuli (see Table 2 for means) showed main effects for age, F(1, 82) 8.75, p.01 (younger 953 ms, older 1,124 ms), stimulus type, F(1, 82) , p.001 (words 926 ms, nonwords 1,152 ms), and case consistency, F(1, 82) , p.001 (uppercase 873 ms, mixed case 1,205 ms). These interactions were also significant: Age Case Consistency, F(1, 82) 6.23, p.05, Stimulus Type Case Consistency, F(1, 82) 83.69, p.001 (mixed-case disadvantage: words 251 ms, nonwords 412 ms), Age Stimulus Type Hue Consistency, F(1, 82) 4.36, p.05 (older adults showed a relatively larger mixed-hue disadvantage for nonwords than for words, but younger adults did not; see Table 2), and Age Case Consistency Hue Consistency, F(1, 82) 4.35, p.05 (younger adults showed a larger mixedcase/mixed-hue advantage than older adults, but both age groups showed comparable uppercase/mixed-hue disadvantages). The analogous error analysis for all stimuli showed main effects for age, F(1, 82) 5.72, p.05 (younger 10.6%,