Journal of Experimental Psychology:
Human Perception and Performance
2009, Vol. 35, No. 2, 480 – 498
© 2009 American Psychological Association
0096-1523/09/$12.00 DOI: 10.1037/a0013027
Serial Position Effects in the Identification of Letters, Digits, and Symbols
Ilse Tydgat
Jonathan Grainger
Ghent University
Centre National de la Recherche Scientifique
and Aix-Marseille University
In 6 experiments, the authors investigated the form of serial position functions for identification of letters,
digits, and symbols presented in strings. The results replicated findings obtained with the target search
paradigm, showing an interaction between the effects of serial position and type of stimulus, with
symbols generating a distinct serial position function compared with letters and digits. When the task was
2-alternative forced choice, this interaction was driven almost exclusively by performance at the first
position in the string, with letters and digits showing much higher levels of accuracy than symbols at this
position. A final-position advantage was reinstated in Experiment 6 by placing the two alternative responses
below the target string. The end-position (first and last positions) advantage for letters and digits compared
with symbol stimuli was further confirmed with the bar-probe technique (postcued partial report) in
Experiments 5 and 6. Overall, the results further support the existence of a specialized mechanism
designed to optimize processing of strings of letters and digits by modifying the size and shape of
retinotopic character detectors’ receptive fields.
Keywords: letter perception, serial position, alphanumeric character processing, crowding
Second, from an empirical perspective, one well-established phenomenon in research investigating letter string perception is the
W-shaped function relating identification accuracy to the position
of target letters in the string. This is the so-called serial position
function (for a recent investigation of letter visibility with varying
fixation positions in the string, see Stevens & Grainger, 2003). For
odd numbered strings with fixation on the central letter (e.g., five- or
seven-letter arrays with central fixation), performance is optimal for
the central letter and then drops as a function of eccentricity, except
for the first and last letter, which recover a performance level similar
to that of the central letter (Averbach & Coriell, 1961; Butler, 1975;
Butler & Merikle, 1973; Haber & Standing, 1969; Merikle, Coltheart, & Lowe, 1971; Merikle, Lowe, & Coltheart, 1971; Mewhort
& Campbell, 1978; Nazir, Ben-Boutayab, Decoppet, Deutsch, &
Frost, 2004; Schwantes, 1978; Stevens & Grainger, 2003; Wolford
& Hollingsworth, 1974). For letter detection speed, there is an
M-shaped equivalent (Hammond & Green, 1982; Ktori & Pitchford, 2008; Mason, 1975, 1982; Mason & Katz, 1976; Pitchford,
Ledgeway, & Masterson, 2008).1
However, this W/M-shaped function is unlikely to be due to a
single process. One popular interpretation of the phenomenon is
that it reflects the conjoint influence of two factors: (a) the drop in
visual acuity as a function of distance from fixation and (b) the
It is now a generally accepted fact that visual word recognition
in languages with alphabetical orthographies involves processing
the identities of the word’s component letters. Many researchers
would further agree that a large part of this processing must be
performed in parallel (e.g., Grainger, 2008; Grainger & Jacobs,
1996; McClelland & Rumelhart, 1981). Nevertheless, surprisingly
little research has been aimed at determining precisely what factors
are at play during the computation of letter identities in the earliest
phases of printed word perception and whether this processing
does, indeed, proceed in parallel. The present study builds on some
classic work in this field in an attempt to highlight some of the key
unresolved issues.
First of all, from a methodological perspective, one particularly
useful approach to addressing such issues is to examine the process
of letter perception in random consonant strings. In this way, one
can gather information about letter-level processing while minimizing the influence of higher level phonological and semantic
processes. The assumption is that during visual word recognition,
some form of letter-level processing must be performed before
such higher order codes come into play (Grainger, Granier, Farioli,
Van Assche, & van Heuven, 2006; Grainger & van Heuven, 2003).
Ilse Tydgat, Department of Experimental Psychology, Ghent University,
Ghent, Belgium; Jonathan Grainger, Centre National de la Recherche
Scientifique and Laboratoire de Psychologie Cognitive, Aix-Marseille University, Marseille, France.
Part of the research reported in this article was performed while Ilse
Tydgat was on an Erasmus exchange between Ghent University and the
University of Provence. Thanks to Marine Salery for running Experiments
5 and 6 and to Nikki Pitchford for discussion of this work.
Correspondence concerning this article should be addressed to Jonathan
Grainger, Laboratoire de Psychologie Cognitive, Université de Provence, 3
pl. Victor Hugo, 13331, Marseille, France. E-mail: jonathan.grainger@
univ-provence.fr
1
The W/M-shaped function is typically found in strings of five letters or
greater with fixation on the central letter. Experiments testing four-letter
strings and/or using parafoveal stimulus presentation have typically shown
U-shaped functions. That is, for stimulus arrays where fixation does not fall
on the middle letter of the array, a U-shaped serial position function for
letter accuracy has been obtained, with optimal performance for the leftmost and rightmost letters (e.g., Bouma, 1973; Campbell & Mewhort,
1980; Estes, Allmeyer, & Reder, 1976; Jordan & Bevan, 1996; Jordan et
al., 2000, 2003; Mewhort & Campbell, 1978; Tramer, Butler, & Mewhort,
1985).
480
SERIAL POSITION EFFECTS
amount of lateral interference (crowding) determined by the number of flanking letters (Bouma, 1970, 1973; Estes, 1972; Estes,
Allmeyer, & Reder, 1976; Haber & Standing, 1969; Van der
Heijden, 1992). Because of the first factor, letter perception becomes worse for letters that are farther from fixation. Because of
the second factor, perception of letters flanked by a letter on each
side is worse than for letters with only one flanking letter (i.e., the
leftmost and rightmost letter in the string). In summary, the reduced crowding as a result of having only one flanking letter
compensates for the drop in visual acuity with the outermost
letters.
This interpretation of the W/M-shaped serial position function
ignores, however, one fundamental finding that was first reported
by Mason and Katz (1976) and was confirmed by Mason (1982)
and Hammond and Green (1982). This finding is that, although
letters and digits show an M-shaped search function, other keyboard characters (e.g., %, }, and §) do not. In these three studies,
the researchers observed an inverse U-shaped serial position function for search times for symbol stimuli that can be interpreted
uniquely in terms of variations in visual acuity. Clearly, if crowding (as defined by Pelli, Palomares, & Majaj, 2004) is a key
ingredient of the M-shaped search function for letters, then symbol
stimuli do not appear to be affected by crowding in the same
manner as letter stimuli.
Mason (1982) concluded that the absence of an advantage for
outer positions (i.e., the first and last positions in the string) for
symbol stimuli is strong evidence that the outer letter advantage
observed in the processing of letter strings is not due to sensory
mechanisms, such as lateral interference (crowding). She argued,
instead, that these end-letter effects likely reflect top-down mechanisms specific to the processing of stimuli that typically occur in
strings (i.e., letters and digits, but not symbols). Biases in the
deployment of spatial attention could provide one such mechanism. Indeed, greater allocation of attention to the periphery might
compensate for perceptual limitations (lower visual acuity) at these
locations (Krumhansl & Thomas, 1976). A periphery-to-center
scanning mechanism had already been proposed to account for the
end-letter advantage (Bouma, 1973; White, 1976) and has been
repeatedly proposed as the general mechanism behind a so-called
ends-first hypothesis of letter string processing (Butler, 1975;
Butler & Merikle, 1973; Krumhansl & Thomas, 1976; Merikle,
1974; Merikle & Coltheart, 1972; Merikle, Coltheart, & Lowe,
1971; White, 1976). In a simple version of this scanning explanation, outer letters would be processed first and would therefore be
more rapidly and easily identified compared with inner letters.
Combined with the effects of visual acuity, this scanning mechanism would then generate the W/M-shaped serial position function
for letters and digits. Symbols, on the other hand, would show
effects only of visual acuity. A parallel-processing version of this
hypothesis might appeal to biases in attentional deployment that
would improve processing of letters in the outer positions of
strings (e.g., Rumelhart & McClelland, 1982).
However, before speculating further on the origins of the endletter advantage and why it is not observed with strings of symbol
stimuli, we decided to first provide further empirical support for
the critical findings of Mason (1982) and Hammond and Green
(1982). Perhaps these results are, for some unknown reason, specific to the target search task used in these experiments. Searching
for a symbol in a string of symbols might invoke a different search
481
strategy compared with looking for letters or digits in strings of
letters and digits, especially because stimulus category was
blocked in these studies. The basic question guiding the present
research is therefore whether distinct serial position functions
would be found for symbol stimuli compared with letters and
digits in an identification paradigm.
To address this issue, the present study used data-limited identification accuracy as the dependent measure. Experiments 1– 4
used two-alternative forced choice (2AFC), whereas Experiment 5
used a postcued partial report bar-probe procedure (a probe indicated the position where item identity was to be reported). The
2AFC procedure was popularized by Reicher (1969) and Wheeler
(1970) in their adaptation of the paradigm to study the word
superiority effect while controlling for possible effects of differential memory loss during response read out and pure guessing
strategies (McClelland & Rumelhart, 1981; Rumelhart & McClelland, 1982). For the present study, the 2AFC task allowed us to
control for a possible influence of the number of elements in each
stimulus category (not the actual number that are tested) that
differed for letters, digits, and symbols. Category size is more
likely to influence performance in an unconstrained partial report
procedure than in 2AFC. Furthermore, 2AFC controls for effects
of positional uncertainty that arise when a participant has identified the target character but incorrectly reports a character from the
wrong position. The possible influence of positional uncertainty
was examined in Experiment 5 with the partial report bar-probe
procedure, and Experiment 6 provided a direct comparison of
performance with the 2AFC and bar-probe techniques.
Anticipating the results, in Experiments 1– 4, we found distinct
serial position functions for letters and digits compared with symbol stimuli, and these functions were principally driven by a strong
initial-position advantage for letters and digits. We did not observe
the standard final-position advantage for letter stimuli when the
2AFC procedure involved presentation of the two alternatives
above and below the target immediately following stimulus offset.
A final-position advantage was found for letters and digits, but not
for symbols, when a postcue stimulus replaced the two alternatives
in Experiments 5 and 6. We explain how hypothesized differences
in the size and shape of receptive fields of retinotopic character
detectors can provide a unified account of the complete pattern of
results.
Experiment 1
Prior research examining serial position effects with the 2AFC
procedure has exclusively used four-letter strings. These studies
typically reported U-shaped functions with optimal performance
on outer letters (e.g., Jordan & Bevan, 1996; Jordan, Patching, &
Milner, 2000; Jordan, Patching, & Thomas, 2003). The aim of
Experiment 1 was, therefore, to examine serial position functions
for letter, digit, and symbol stimuli in a data-limited identification
task with 2AFC. Figure 1 illustrates the 2AFC and bar-probe
procedures used in the present study.
Method
Participants. Twenty-four students (3 men and 21 women) at
the University of Provence, Marseille, France, were paid €5 for
their participation in the experiment. Their mean age was 20 years,
and they all reported having normal or corrected-to-normal vision.
482
TYDGAT AND GRAINGER
Figure 1. Illustration of the paradigms used in the present study. A pattern mask plus fixation bars were always
followed by a stimulus presented briefly (for 100 ms), which was followed, in turn, by a pattern mask
accompanied either by two alternatives (Experiments 1– 4) or by two hyphen marks (Experiments 5– 6) located
above and below a specific location in the string. The two-alternative forced-choice paradigm of Experiments
1– 4 is illustrated in the left three schemas (Experiments 1 and 2 in the first schema, Experiment 3 in the second,
and Experiment 4 in the third). The bar-probe procedure of Experiments 5 and 6 is shown in the rightmost
schema.
Stimuli and design. All stimuli consisted of horizontal arrays
of five characters. Three different types of stimuli were used:
consonant letters presented in uppercase (B, D, F, G, K, N, L, S,
and T), digits (1 to 9), and symbols (%, /, ?, @, }, ⬍, £, §, and ␮).
These three categories were assigned to three blocks, so that one
block consisted of 90 letter-array stimuli, one of 90 digit-array
stimuli, and one of 90 symbol-array stimuli. Although the presentation order of the blocks was counterbalanced among participants,
all other factors were manipulated as within-subjects variables.
These were (a) target type (letter, digit, or symbol) and (b) target
position in the array (Positions 1–5). Each array consisted of a
quasi-random sequence of characters, with each of the target
characters being presented 2 times at each of the five target
positions and 40 times at a nontarget position. Thus, accuracy for
each type of target at each target position was based on 18
observations per participant. For the purposes of the forced-choice
task, each target character was paired with an alternative character
so that a target character was always presented with the same
alternative at each of the five positions. Each target character
served as the alternative for one other character. The incorrect
alternative presented for forced choice was never present in the
stimulus array. The correct alternative appeared once above the
backward mask (with the incorrect alternative below) and once
below the backward mask (with the incorrect alternative above) for
each target position.
Procedure. The experiment was run inside a dimly lit room
and was controlled with E-Prime software (Psychology Software
Tools, www.pstnet.com/eprime). Participants were seated in front
of a computer screen at a viewing distance of approximately 60
cm. At that distance, target stimuli in the five different positions
were located at ⫺1.2°, ⫺0.6°, 0°, 0.6°, and 1.2° of visual angle
from the central fixation point, and each letter, digit, or symbol
character subtended on average 0.44° of visual angle. Stimuli were
displayed in white on a black background and were presented in
18-point Courier New font. Participants first saw instructions and
practiced the task for 81 trials to become familiarized with the
procedure and the stimuli. Each trial began with two vertical
fixation bars, placed above and below the center of a forward
mask. The forward mask consisted of five hash marks and stayed
on the screen for 515 ms. Then the fixation bars and the mask
disappeared, and the array of five characters immediately appeared
for a duration of 100 ms. This was followed by a backward mask,
which was identical to the forward mask and was accompanied by
two characters, one above the mask and one below at one of the
five possible positions. Participants had to decide which one of
these two characters was present in the corresponding position of
the preceding array. They were asked to respond as accurately as
possible by pressing one of two keys on the keyboard. They had to
choose either the upward arrow key (for the alternative above) or
the downward arrow key (for the alternative below). After the
response, the screen was cleared, and the two fixation bars appeared on the screen. There was a delay of 515 ms before the next
trial started. The participants could take a short break between the
three blocks of 90 trials each. The experiment lasted approximately
30 min.
Results
Mean accuracies for all target types and target positions are
presented in Figure 2. An analysis of variance (ANOVA) was
performed on the data with target type (letter, symbol, or digit) and
target position (1–5) as within-subjects variables. In the item
analysis, target type was a between-items variable, and target
position was a within-items variable. In this and each of the
following experiments, the Greenhouse and Geisser (1959) correction was applied to determine the significance values for variables with more than two levels. The main statistical results are
presented in Table 1. Symbol targets were identified at slightly
lower levels of accuracy than letter and digit targets (respectively,
57.6%, 59.9%, and 60.1%). Accuracy varied significantly as a
function of serial position, and the serial position function for
symbol stimuli differed from that found for letters and digits
(Target Type ⫻ Target Position interaction). As can be seen from
Figure 2, this interaction was mostly driven by differences in
performance at the first position in strings. In line with this
observation, the critical Target Type ⫻ Target Position interaction
was significant when considering only the first and second positions in strings and was not significant when considering only the
fourth and fifth positions in strings.
Following Mason (1982) and Hammond and Green (1982), we
also performed trends analyses to examine the best fitting serial
position functions for the different types of stimuli. Both significant quartic and linear effects were found for letter stimuli, F1(1,
18) ⫽ 21.58, p ⬍ .001; F2(1, 8) ⫽ 13.34, p ⬍ .01, and F1(1, 18) ⫽
8.71, p ⬍ .01; F2(1, 8) ⫽ 15.47, p ⬍ .01, respectively. These
effects explained 95% of the observed variance, with 54% due to
a quartic function alone (W shape) and 41% due to a linear
function. For the symbol characters, there was a significant qua-
SERIAL POSITION EFFECTS
483
Figure 2. Serial position functions for letters, symbols, and digits in Experiment 1 in which the different
stimulus categories were presented in different blocks. Data are percentage correct for the two-alternative
forced-choice task, and vertical bars represent standard errors.
dratic effect (inverted U shaped), F1(1, 18) ⫽ 10.29, p ⬍ .01; F2(1,
8) ⫽ 18.68, p ⬍ .01, which explained 86% of the observed
variance. For the digit characters, the effects were comparable to
those of letters. There was a significant quartic effect, F1(1, 18) ⫽
40.71, p ⬍ .001; F2(1, 8) ⫽ 25.93, p ⬍ .001, and a smaller linear
effect, F1(1, 18) ⫽ 8.31, p ⬍ .01; F2(1, 8) ⫽ 13.40, p ⬍ .01. These
two effects explained 85% of the variance, with 64% for the
quartic function and 21% for the linear function.
Discussion
The results of Experiment 1 clearly demonstrate that the different serial position functions for letter, digit, and symbol stimuli
reported by Mason (1982) and Hammond and Green (1982) are not
specific to the target search task. Experiment 1 tested letter, digit,
and symbol stimuli in a data-limited identification paradigm with
2AFC and found serial position functions that differed as a function of stimulus type. Letters and digits showed a distinct advantage for the first and central positions, with variance across position being best captured by a quartic function. Symbols, on the
other hand, showed an inverted U-shaped function, with variance
across position being best captured by a quadratic function.
Experiment 2
In Experiment 1, the bulk of the difference between symbol
stimuli, on the one hand, and letters and digits, on the other, was
carried by performance for items at the beginning of strings. One
possible account of this initial-position advantage is that the pres-
ence of strings of letters or digits generates an endogenous bias
toward the beginning of these strings. This would provide a mechanism for initiating a left-to-right attentional scanning of such
stimuli (e.g., Mewhort & Popham, 1991). This attentional bias
could be driven by the expectation that the stimuli to be processed
are strings of letters or digits, and it would not occur when the
expected stimuli are symbols. This type of strategy would be
encouraged by the blocked presentation of stimulus type in prior
studies that have used the target search paradigm and in Experiment 1 of the present study. Participants knew in advance what
kind of stimulus they would receive and could therefore adjust
endogenous attention as a function of the expected stimulus type.
In Experiment 2, we tested this hypothesis by randomly intermixing the three types of stimuli during the experiment. This manipulation should cancel the initial-position advantage found for letters and digits and should produce similar serial position functions
for all three types of stimuli.
Method
Participants. Twenty-four students (7 men and 17 women) at
the University of Provence, Marseille, volunteered to participate in
the experiment. Their mean age was 21 years, and all reported
having normal or corrected-to-normal vision.
Stimuli and design. Stimuli and design were exactly the same
as in Experiment 1.
Procedure. The procedure was also the same as in Experiment
1, except that the three blocks of 90 trials now contained a random
Table 1
Main Statistical Results for Experiment 1
Effect
F1
␩2
F2
␩2
Target type
Target position
Target Type ⫻ Target Position
Only first and second positions:
Target Type ⫻ Target Position
Only fourth and fifth positions:
Target Type ⫻ Target Position
F1(2, 36) ⫽ 3.54ⴱ
F1(4, 72) ⫽ 7.85ⴱⴱⴱ
F1(8, 144) ⫽ 3.86ⴱⴱⴱ
0.01
0.12
0.05
F2 ⬍ 1
F2(4, 96) ⫽ 17.46ⴱⴱⴱ
F2(8, 96) ⫽ 3.47ⴱⴱ
0.07
0.03
F1(2, 36) ⫽ 6.75ⴱⴱ
0.07
F2(2, 24) ⫽ 5.88ⴱⴱ
0.04
F1(2, 36) ⫽ 2.13
0.02
F2(2, 24) ⫽ 1.24
0.01
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
ⴱⴱⴱ
p ⬍ .001.
484
TYDGAT AND GRAINGER
mixture of target types (letters, symbols, and digits). This was also
the case for the practice trials.
attentional capture). Strings of letters and digits might in some way
automatically draw attention to their beginnings.
Results
Experiment 3
Mean accuracies for each target type and target position are
presented in Figure 3. The results of a 3 (target type) ⫻ 5 (target
position) ANOVA performed on the accuracy data are shown in
Table 2. Performance for symbol targets was slightly lower than
for letters and digits, with 57.3%, 60.0%, and 61.7% mean accuracy, respectively. Accuracy varied significantly as a function of
target position, and the resulting serial position functions differed
as a function of target type. This Target Position ⫻ Target Type
interaction reflects the superior performance on letter and digit
stimuli that was mostly evident at the first position.
For letters, there was a significant linear trend in the serial
position function, F1(1, 23) ⫽ 17.95, p ⬍ .001; F2(1, 8) ⫽ 12.69,
p ⬍ .01, which explained 83% of the observed variance, and there
was a smaller quartic effect, F1(1, 23) ⫽ 4.21, p ⬍ .1; F2(1, 8) ⫽
6.16, p ⬍ .05, which explained about 15% of additional variance.
For the symbols, both a quadratic, F1(1, 23) ⫽ 6.95, p ⬍ .05; F2(1,
8) ⫽ 10.83, p ⬍ .05, and a quartic effect, F1(1, 23) ⫽ 6.36, p ⬍
.05; F2(1, 8) ⫽ 6.39, p ⬍ .05, were found, and they explained,
respectively, 53% and 44% of the variance. For digits, there was a
linear effect, F1(1, 23) ⫽ 17.38, p ⬍ .001; F2(1, 8) ⫽ 32.38, p ⬍
.001, and a quartic effect, F1(1, 23) ⫽ 8.47, p ⬍ .01, F2(1, 8) ⫽
32.89, p ⬍ .001, which explained 60% and 30% of the variance,
respectively.
Discussion
The results of Experiment 2 provide a replication of Experiment
1 and demonstrate that differences in the serial position function
for symbols compared with letters and digits are not affected by
randomly intermixing the three types of stimuli. On the contrary,
the linear trend for letters and digits found in Experiment 1 was
even more exaggerated in Experiment 2. This result allowed us to
rule out an endogenous top-down attentional bias as the source of
the initial-position advantage in the perception of letters and digits
in strings. If some form of attentional mechanism is the basis of the
first-position advantage found for letters and digits, then this
mechanism must be automatically triggered on presentation of a
string of letters or digits on each trial (i.e., bottom-up exogenous
Experiment 3 provided a further test of the attentional bias
hypothesis. If this bias is automatically triggered on presentation of
a string of letters or digits, even when these trials are intermixed
with symbol strings, then one ought to be able to remove the
initial-position advantage by placing letter targets in a string of
symbols. The assumption here is that it is the dominant stimulus
type that should determine the nature of bottom-up processing of
the whole string (i.e., four symbols with one embedded letter
should not trigger an attentional bias). Therefore, single letters
embedded in a string of symbols should not show an initialposition advantage, whereas symbol targets embedded in a string
of letters should show the advantage.
Experiment 3 also allowed us to test an alternative explanation
of the first-position advantage found with letters and digits expressed in terms of a chunking mechanism that forms higher order
combinations of items of the same category, with the higher order
combinations being more resistant to masking than individual
items. This chunking mechanism would be a basic component of
the general process of word and number recognition and would
therefore operate only on letter and digit strings and not on symbol
strings (if we assume that there are no higher order symbol
combinations stored in long-term memory). The first-position advantage found for letters and digits could then be due to the precise
nature of this chunking process, with initial letters and digits
playing a dominant role (e.g., Whitney, 2001). By embedding
target letters in symbol strings in Experiment 3, we expected to
block any influence of this hypothesized chunking mechanism,
because letters and digits would not be able to combine with other
letters and digits.
Method
Participants. Twenty-four students (2 men and 22 women) at
the University of Provence, Marseille, volunteered to participate in
the experiment. Their mean age was 20 years, and all reported
having normal or corrected-to-normal vision.
Stimuli and design. Target stimuli were the same as in the
previous experiments, except that there were no digit stimuli;
Figure 3. Serial position functions for letters, symbols, and digits in Experiment 2 in which the different
stimulus categories were randomly intermixed during the experiment. Data are percentage correct for the
two-alternative forced-choice task, and vertical bars represent standard errors.
SERIAL POSITION EFFECTS
485
Table 2
Main Statistical Results for Experiment 2
Effect
F1
␩2
F2
␩2
Target type
Target position
Target Type ⫻ Target Position
Only first and second positions:
Target Type ⫻ Target Position
Only fourth and fifth positions:
Target Type ⫻ Target Position
F1(2, 46) ⫽ 6.26ⴱⴱ
F1(4, 92) ⫽ 8.80ⴱⴱⴱ
F1(8, 184) ⫽ 2.40ⴱ
0.02
0.11
0.03
F2 ⬍ 1
F2(4, 96) ⫽ 16.80ⴱⴱⴱ
F2(8, 96) ⫽ 2.29ⴱ
0.19
0.05
F1(2, 46) ⫽ 2.68
0.03
F2(2, 24) ⫽ 4.05ⴱ
0.05
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
ⴱⴱⴱ
F1 ⬍ 1
F2 ⬍ 1
p ⬍ .001.
therefore, there were only 180 trials. Also, the nine target letters
were presented in a string of symbols, and the nine target symbol
characters were now presented in a string of letters.
Procedure. Participants practiced the task during 20 practice
trials, randomly chosen out of the stimulus list of 180 trials. All
other components of the procedure were exactly the same as in
Experiment 2. The experiment lasted approximately 20 min.
Results
Mean accuracies for each target type and target position are
plotted in Figure 4, and the results of a 2 (target type) ⫻ 5 (target
position) ANOVA are shown in Table 3. Accuracy for symbol
targets (63.1%) was somewhat higher than accuracy for letter
targets (57.6%). The main effect of target position was significant,
but this did not interact significantly with target type. However,
there was a trend toward an interaction effect when only the first
and second positions were analyzed. Accuracy for letter targets
was higher in Position 1 than Position 2, whereas the opposite was
true for symbol targets.
For the letter stimuli, the sum of a linear, cubic, and quartic
trend in the serial position function explained 97% of the observed
variance, with 37%, 36%, and 24% for each trend, respectively.
However, none of these trends were significant. For the symbol
stimuli, there was a significant linear effect, F1(1, 23) ⫽ 4.74, p ⬍
.05; F2(1, 8) ⫽ 6.39, p ⬍ .05, a less robust quadratic effect, F1(1,
23) ⫽ 3.48, p ⬍ .1; F2(1, 8) ⫽ 6.26, p ⬍ .05, and a less robust
quartic effect, F1(1, 23) ⫽ 4.74, p ⬍ .05; F2(1, 8) ⫽ 5.17, p ⬍ .1.
These three effects explained, respectively, 27%, 38%, and 24% of
the observed variance.
Discussion
The results of Experiment 3 allowed us to rule out an automatically
triggered attentional bias account of the initial-position advantage
found for letter-in-string perception compared with symbols. If this
attentional bias was triggered only on presentation of a string of letters
(or digits)— hence explaining why there is no initial-position advantage for symbols—then we expected to remove the advantage for
letter targets by inserting them in strings of symbols, and we expected
to create an advantage for symbol targets by inserting them in strings
of letters. However, although the results were somewhat weaker in
Experiment 3, an ANOVA that tested only the first and second
positions did show a trend toward an interaction. This reflected the
fact that performance tended to be better at the first position compared
with the second position for letter targets, whereas symbol targets
showed a difference in the opposite direction. An inspection of
Figure 4 shows that the general pattern found in Experiments 1 and 2
was replicated in Experiment 3.
Figure 4. Serial position functions in Experiment 3 for letter targets embedded in symbol arrays and symbol
targets embedded in letter arrays. Data are percentage correct for the two-alternative forced-choice task, and
vertical bars represent standard errors.
TYDGAT AND GRAINGER
486
Table 3
Main Statistical Results for Experiment 3
Effect
F1
␩2
F2
␩2
Target type
Target position
Target Type ⫻ Target Position
Only first and second positions:
Target Type ⫻ Target Position
Only fourth and fifth positions:
Target Type ⫻ Target Position
F1(1, 23) ⫽ 14.83ⴱⴱⴱ
F1(4, 92) ⫽ 3.56ⴱ
F1(4, 92) ⫽ 1.50
0.05
0.05
0.02
F2(1, 16) ⫽ 2.24
F2(4, 64) ⫽ 3.07ⴱ
F2(4, 64) ⫽ 1.25
0.06
0.07
0.03
F1(1, 23) ⫽ 3.74
0.03
F2(1, 16) ⫽ 3.03
0.03
ⴱ
p ⬍ .05.
ⴱⴱⴱ
F1 ⬍ 1
F2 ⬍ 1
p ⬍ .001.
This pattern was further confirmed by a combined analysis of
Experiments 2 and 3, including only the data for letter and symbol
targets from Experiment 2. The results of this joint ANOVA are
shown in Table 4. There was no sign of a Target Position ⫻ Experiment interaction. Most important, the critical Target Type ⫻ Target
Position interaction was significant in the combined analysis, and
once again this interaction was shown to be carried completely by the
difference in performance across Positions 1 and 2. There was an
initial-position advantage (relative to the second position) for letter
targets and a disadvantage for symbol targets; these were independent
of the nature of the other items in the string.
Experiment 3 also allows us to reject a chunking account of the
different serial position functions for letters and symbols. According
to this account, the critical difference between a string of letters and
a string of symbols is that only the former are automatically subject to
a chunking process whereby higher order letter combinations are
formed. We predicted that embedding letter targets in strings of
symbols would disrupt such a chunking process and remove any
differences in performance for letter and symbol targets. The results of
Experiment 3 do not support this prediction. It would appear that
some other mechanism, also specific to the processing of stimuli that
typically appear in strings, is the source of the distinct serial position
functions obtained for letters and symbols.
Experiment 4
Experiments 1–3 further confirmed the seminal findings of
Mason (Mason, 1982; Mason & Katz, 1976) and Hammond and
Green (1982) that symbol stimuli show distinct serial position
functions compared with both letters and digits. The most parsimonious explanation for this systematic pattern of effects is that
letters and digits benefit from a specialized processing system that
endows an advantage for elements at the first position of a string.
Therefore, a combination of visual acuity and an initial-position
advantage for letters and digits provides a complete account of the
pattern of results found in Experiments 1–3.
According to this account of the results of Experiments 1–3, if we
remove the central character in the stimuli tested in the previous
experiments, then letter stimuli should show improved performance at
Position 4 as opposed to Position 2, because Position 4 is now at the
beginning of a string of two letters. On the other hand, symbol stimuli
should not be affected by this manipulation, because visual acuity
should not be affected by removing the central character. To test this,
Experiment 4 examined identification of letters and symbols in conditions that were identical to those of Experiment 2, except that the
central letter/symbol was removed. Therefore, on every trial, participants were presented with four letters or four symbols, two to the left
of fixation and two to the right of fixation, in Positions 1, 2, 4, and 5
of the previous experiments.
Method
Participants. Twenty-four students (6 men and 18 women) at
the University of Provence, Marseille, volunteered to participate in
the experiment. Their mean age was 21 years, and all reported
having normal or corrected-to-normal vision.
Stimuli and design. Stimuli were the same as in Experiment 2,
although there were no digit characters, and no character was
Table 4
Main Statistical Results for the Combined Analysis of Experiments 2 and 3
F2
␩2
0.01
0.02
F2(1, 16) ⫽ 1.24
F2 ⬍ 1
F2(4, 64) ⫽ 12.55ⴱⴱⴱ
F2 ⬍ 1
F2(4, 64) ⫽ 3.63ⴱⴱ
F1(4, 184) ⫽ 1.79
0.01
F2(4, 64) ⫽ 1.22
0.02
F1(1, 46) ⫽ 7.19ⴱ
0.03
F2(1, 16) ⫽ 8.21ⴱ
0.04
Effect
F1
Experiment
Target type
Target position
Experiment ⫻ Target Position
Target Type ⫻ Target Position
Experiment ⫻ Target Type ⫻
Target Position
Only first and second positions:
Target Type ⫻ Target Position
Only fourth and fifth positions:
Target Type ⫻ Target Position
F1 ⬍ 1
F1(1, 46) ⫽ 2.00
F1(4, 184) ⫽ 9.25ⴱⴱⴱ
F1 ⬍ 1
F1(4, 184) ⫽ 3.53ⴱⴱ
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
ⴱⴱⴱ
p ⬍ .001.
F1 ⬍ 1
␩2
0.00
0.07
F2 ⬍ 1
0.10
0.03
SERIAL POSITION EFFECTS
presented at the third position. Therefore, targets could appear at
four possible positions, two to the left and two to the right of
fixation, and on each trial, four letters or four symbols appeared at
these four locations. In this design, target position is defined as the
combination of two variables: visual field (left vs. right) and target
position (first vs. second).
Procedure. The procedure was the same as in Experiment 2,
except for the removal of the central character in the stimulus
strings and the pattern mask. Participants were first presented with
a practice block of 20 randomly chosen trials, followed by the 144
experimental trials. The experiment lasted about 20 min.
Results and Discussion
The mean accuracies per experimental condition are given in
Table 5. The results of a 2 (target type) ⫻ 2 (target position) ⫻ 2
(visual field) ANOVA are shown in Table 6. Mean accuracy for
letters was 63.5%, and accuracy for symbols was 60.1%. There
was a significant triple interaction reflecting the fact that the
Target Type ⫻ Target Position interaction was significant in the
left visual field (LVF) but not in the right visual field (RVF). In
line with the results of Experiments 1–3, the results of Experiment
4 showed once again a strong interaction in the identification
accuracy for letters and symbols at Serial Positions 1 and 2 in the
LVF. Although overall performance was lower for symbols compared with letters, the pattern of results for targets presented to the
right of fixation was almost identical for letters and symbols, with
a 10% advantage for the innermost target.
Figure 5 shows the results of Experiments 2 and 4 plotted
together. To examine the influence of removing the central character on performance for letters and symbols, we performed a
series of planned comparisons (Experiment 2 vs. Experiment 4) for
each serial position. For letter stimuli, there was a significant effect
at Position 1 in the RVF, F1(1, 46) ⫽ 10.97, p ⬍ .01, ␩2 ⫽ 0.19;
F2(1, 16) ⫽ 6.99, p ⬍ .05, ␩2 ⫽ 0.30. For symbol stimuli, on the
other hand, there was a significant difference at both Position 2 in
the LVF, F1(1, 46) ⫽ 5.46, p ⬍ .05, ␩2 ⫽ 0.11; F2(1, 16) ⫽ 4.93,
p ⬍ .05, ␩2 ⫽ 0.24, and Position 1 in the RVF, F1(1, 46) ⫽ 6.01,
p ⬍ .05, ␩2 ⫽ 0.12; F2(1, 16) ⫽ 4.63, p ⬍ .05, ␩2 ⫽ 0.22.
In line with the predictions of the initial-position advantage for
letter stimuli, performance at Position 4 (innermost letter in the
RVF) was significantly greater in Experiment 4 than in Experi-
Table 5
Results (Percentage Correct Identification) in Experiment 4 for
Letter and Symbol Targets
Accuracy
Letters
Symbols
Visual
field/position
%
SE
%
SE
Left/1
Left/2
Right/1
Right/2
69.9
63.4
65.3
56.5
2.9
2.7
2.0
2.3
57.4
66.7
63.2
53.2
2.7
2.4
2.6
2.5
Note. Targets were presented at the same positions as the first, second,
fourth, and fifth positions in Experiments 1–3, thus giving two positions in
the left visual field and two positions in the right visual field.
487
ment 2, whereas performance at all other positions remained
statistically equivalent across experiments. However, there was
also a significant increase in performance at Position 2 (innermost
letter in the LVF) and Position 4 for symbol stimuli, which was not
predicted by this account. Therefore, it appears from the combined
results of Experiments 2 and 4 that a simple combination of visual
acuity and an initial-position advantage for letters cannot provide
a satisfactory account of performance for letters and symbols in
these two experiments. This combined analysis clearly suggests
that the presence of a symbol character at the central location in
Experiment 2 exerted an inhibitory influence on the identification
of symbols presented at Positions 2 and 4. This, therefore, reinstates crowding as one major factor determining the visibility of
characters presented in strings and further raises the possibility that
differential effects of crowding might underlie the different serial
position functions for letters, digits, and symbols. The results of
Experiment 4 further suggest that visual field differences in crowding effects might be part of the overall story.
Experiment 5
One critical difference with respect to prior observations of
outer-position effects for letters is that the bulk of the advantage
was carried by the first letter in our experiments, with little
evidence for a final-letter advantage.2 This result is all the more
important given the nature of the paradigm we used, which, if
anything, should have reduced sequential, beginning-to-end biases
compared with the letter search task. Experiment 5 was designed
to examine one possible reason for this divergence with respect to
prior research on serial position functions for letter stimuli in
data-limited identification tasks.
The 2AFC task used in Experiments 1– 4 has the advantage of
eliminating errors that occur when the target item has been identified
but another item in the string is reported (location errors). These
are eliminated in our 2AFC procedure because the alternative item
provided for response is never present in the string. Thus, the
2AFC procedure provides a means of estimating how well a given
item has been identified at a given position in the absence of
contamination from such location errors. This implies that the
first-position advantage that we have systematically found for
letters and digits, but not for symbols, was not due to improved
encoding of positional information for letters and digits that appear
at the beginning of strings. On the other hand, the final-position
advantage found for letter and digit stimuli in prior research, but
not observed in Experiments 1– 4, could well have been due to a
larger number of location errors when identifying interior letters/
digits than final letters/digits. There is evidence from experiments
using the bar-probe technique that this is, indeed, the case. First, it
has been shown that use of a backward mask (as was the case in
the present experiments) has a greater influence on accuracy of
report from inner positions compared with outer positions
(Mewhort & Campbell, 1978). Second, it has been shown that the
2
It should nevertheless be noted that one recent study of serial position
effects in the target search paradigm (Pitchford et al., 2008) did not find
such a pronounced final-position advantage as in the earlier work of Mason
(1982). The results of Pitchford et al. (2008) suggested that positional letter
frequency might be one factor determining the size of outer-position
effects.
TYDGAT AND GRAINGER
488
Table 6
Main Statistical Results for Experiment 4
Effect
Target type
Target position
Visual field
Target Type ⫻ Target Position
Target Type ⫻ Visual Field
Target Position ⫻ Visual Field
Target Type ⫻ Target Position ⫻
Visual Field
Left visual field: Target Type ⫻
Target Position
Right visual field: Target Type ⫻
Target Position
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
␩2
F1
ⴱⴱⴱ
F1(1, 23)
F1(1, 92)
F1(1, 23)
F1(1, 23)
F1 ⬍ 1
F1(1, 23)
⫽
⫽
⫽
⫽
4.99ⴱ
8.70ⴱⴱ
3.97
7.13ⴱ
F1(1, 23) ⫽ 7.12ⴱ
F1(1, 23) ⫽ 14.23
0.04
F2 ⬍ 1
F2(1, 16)
F2(1, 16)
F2(1, 16)
F2 ⬍ 1
F2(1, 16)
0.03
F2(1, 16) ⫽ 12.13ⴱⴱ
0.02
0.02
0.03
0.02
⫽ 11.73ⴱⴱ
ⴱⴱⴱ
F1 ⬍ 1
␩2
F2
0.09
⫽ 6.51ⴱ
⫽ 6.57ⴱ
⫽ 5.43ⴱ
0.03
0.04
0.02
⫽ 19.69ⴱⴱⴱ
0.05
F2(1, 16) ⫽ 20.44
ⴱⴱⴱ
0.03
0.12
F2 ⬍ 1
p ⬍ .001.
use of a backward mask increases the number of location errors as
opposed to intrusion (item) errors in the bar-probe identification
task (Mewhort & Campbell, 1978; Mewhort, Campbell, Marchetti,
& Campbell, 1981; Mewhort, Huntley, & Duff-Fraser, 1993). If
the final-position advantage found in studies using a partial report
bar-probe procedure was due to the larger number of positional
errors for inner positions compared with the final position, then we
would expect to observe this advantage with exactly the same
stimuli and procedure as in Experiment 2, but with the bar-probe
partial report procedure replacing the 2AFC procedure.
Method
Participants. Twenty-four students (4 men and 20 women) at
the University of Provence, Marseille, were paid 5€ to participate
in the experiment. Their mean age was 23 years, and all reported
having normal or corrected-to-normal vision.
Stimuli and design. Stimuli and design were exactly the same
as in Experiment 2.
Procedure. The procedure was identical to that of Experiment
2, except for a change in the report procedure. The 2AFC task was
replaced by a bar-probe partial report task. The backward mask
was accompanied by two hyphen marks positioned above and
below one of the five positions in the string (hence mimicking the
position of the two alternative letters in the 2AFC procedure).
Participants were asked to decide as accurately as possible which
character they had seen at this position, by choosing from nine
letters, nine digits, and nine symbols indicated on the keyboard in
front of the participant. They responded by pressing one of these
27 marked response keys on the keyboard.
Results
Two participants had to be excluded from the analyses because
they had not followed the task instructions. For the remaining 22
participants, mean accuracy for each target type and target position
is presented in Figure 6. A 3 (target type) ⫻ 5 (target position)
ANOVA was performed on the accuracy data. The main statistical
results are presented in Table 7. Performance for symbol targets
was lower than for letter and digit targets, with 36.8%, 60.7%, and
61.0% mean accuracy, respectively.3 Target position significantly
influenced performance and interacted significantly with target
type, reflecting the different serial position function obtained for
symbol stimuli compared with both letters and digits. The Target
Type ⫻ Target Position interaction was significant in an analysis
restricted to the first and second positions and in an analysis with
only the fourth and fifth positions.
Trends analyses were performed to examine the best fitting
serial position functions for the different types of stimuli. For
letters, there was a significant linear effect, F1(1, 21) ⫽ 4.61, p ⬍
.05; F2(1, 8) ⫽ 4.67, p ⬍ .1, quadratic effect, F1(1, 21) ⫽ 15.39,
p ⬍ .001; F2(1, 8) ⫽ 21.18, p ⬍ .01, and quartic effect, F1(1,
21) ⫽ 47.59, p ⬍ .001; F2(1, 8) ⫽ 27.47, p ⬍ .001. These effects
accounted for 12%, 40%, and 46% of the variance, respectively.
For digits, such linear, quadratic, and quartic effects explained 5%,
16%, and 79% of the variance, respectively. Again, all of these
functions were significant, with F1(1, 21) ⫽ 4.48, p ⬍ .05; F2(1,
8) ⫽ 12.62, p ⬍ .01, and F1(1, 21) ⫽ 8.17, p ⬍ .01; F2(1, 8) ⫽
15.32, p ⬍ .01, and F1(1, 21) ⫽ 104.17, p ⬍ .001; F2(1, 8) ⫽
76.65, p ⬍ .001, respectively. For symbols, we found a cubic
effect, F1(1, 21) ⫽ 9.96, p ⬍ .01; F2(1, 8) ⫽ 3.60, p ⬍ .1, and a
quartic effect, F1(1, 21) ⫽ 42.26, p ⬍ .001; F2(1, 8) ⫽ 29.68, p ⬍
.001. These functions explained 11% and 76% of the variance,
respectively.
In the partial report task used in Experiment 5, errors can be
categorized as intrusion errors—when the erroneous response was
not a member of the stimulus string— or location errors—when the
erroneous response was present at another location in the string.
3
Because of the low overall level of accuracy for symbol stimuli, we
analyzed the results of a group of 6 participants with the highest performance for symbols. In this group, accuracy for symbol targets at Positions
2, 3, and 4 was statistically equivalent to accuracy with digits and letters at
these positions, F1(1, 5) ⫽ 2.72, p ⫽ .16; F2 ⬍ 1. For all positions, the
main effect of target type was significant, F1(2, 10) ⫽ 31.90, p ⬍ .001,
␩2 ⫽ 0.12; F2(2, 24) ⫽ 5.26, p ⬍ .05, ␩2 ⫽ 0.11, as were the main effect
of target position, F1(4, 20) ⫽ 3.22, p ⬍ .05, ␩2 ⫽ 0.23; F2(4, 96) ⫽ 13.58,
p ⬍ .001, ␩2 ⫽ 0.21, and the interaction, F1(8, 40) ⫽ 4.82, p ⬍ .001, ␩2 ⫽
0.10; F2(8, 96) ⫽ 2.86, p ⬍ .01, ␩2 ⫽ 0.09. The interaction was clearly
driven by the lower levels of accuracy for symbol stimuli compared with
letters and digits at Position 1, F1(1, 5) ⫽ 52.70, p ⬍ .001, ␩2 ⫽ 0.91; F2(1,
24) ⫽ 4.35, p ⬍ .05, ␩2 ⫽ 0.46, and at Position 5, F1(1, 5) ⫽ 11.75, p ⬍
.05, ␩2 ⫽ 0.70; F2(1, 24) ⫽ 3.66, p ⬍ .1, ␩2 ⫽ 0.40.
SERIAL POSITION EFFECTS
489
Figure 5. Percentage correct for the two-alternative forced-choice task for letter and symbol targets in
Experiment 4 (Positions 4 and 5 correspond to Positions 1 and 2 in the right visual field) and for the same
positions in Experiment 2. In Experiment 2, there was also a character present at Position 3 in the array, whereas
no character was present at this position in Experiment 4. Vertical bars represent standard errors. Expt. ⫽
Experiment.
Performance for letter, symbol, and digit targets is broken down
by error type in Figure 7. Separate ANOVAs were performed
for location errors and intrusion errors (see Table 7). When only
location errors were counted as errors, mean accuracies for
letters, symbols, and digits were 72.1%, 60.3%, and 70.3%,
respectively. When only intrusion errors were counted as errors,
mean accuracies for letters, symbols, and digits were 88.5%,
76.5%, and 90.7%, respectively. The main effects and interactions were significant in both analyses. As can be seen in
Figure 7, for location errors, the interaction was mostly driven
by the lower levels of accuracy for symbols as opposed to
letters and digits at Positions 1 and 5. For intrusion errors, one
can also note a sizeable difference at Position 2 for symbols
relative to letters and digits.
Discussion
The results of Experiment 5 show that replacing 2AFC with a
postcued partial report procedure reinstated the final-letter advantage for letters and digits. These stimuli then generated the classic
W-shaped serial position function for letter and digit identification
accuracy obtained with the bar-probe procedure (e.g., Haber &
Standing, 1969; Merikle, Lowe, & Coltheart, 1971), the equivalent
of the M-shaped function typically found for letter and digit
detection latencies (e.g., Hammond & Green, 1982; Mason, 1982).
Because the only difference between Experiment 2 and Experiment 5 was the switch from a 2AFC procedure to a bar-probe
procedure, this suggests that the final-position advantage found in
Experiment 5 was due to the larger number of location errors in
Position 4 compared with Position 5. Such location errors would
not arise in the 2AFC task because the alternative response was
never a part of the target string. Furthermore, one would expect a
greater number of location errors at inner positions compared with
outer positions because the latter have only a single flanking
character. An analysis of the different types of error (intrusion vs.
location) provided some support for this analysis. Location errors
did indeed show a more exaggerated W-shaped serial position
function than intrusion errors.
However, one aspect of the comparison of Experiments 2 and 5
does not fit an explanation of differences in performance in the 2AFC
and bar-probe procedures being uniquely driven by the greater
number of location errors in the latter. That aspect is the overall
estimated superior performance in the bar-probe task over the
2AFC. To compare performance in the bar-probe task of Experiment 5 to performance in the 2AFC task of Experiment 2, a graph
of transformed partial report accuracies was plotted by applying an
equation for transforming partial report into approximate 2AFC
Figure 6. Serial position functions for letters, symbols, and digits in Experiment 5. Data are percentages correct
with a bar-probe partial report procedure, and vertical bars represent standard errors.
TYDGAT AND GRAINGER
490
Table 7
Main Statistical Results for Experiment 5
Effect
F1
␩2
F2
␩2
Target type
Target position
Target Type ⫻ Target Position
Only first and second positions:
Target Type ⫻ Target Position
Only fourth and fifth positions:
Target Type ⫻ Target Position
Location errors
Target type
Target position
Target Type ⫻ Target Position
Intrusion errors
Target type
Target position
Target Type ⫻ Target Position
F1(2, 42) ⫽ 58.47ⴱⴱⴱ
F1(4, 84) ⫽ 17.76ⴱⴱⴱ
F1(8, 168) ⫽ 13.11ⴱⴱⴱ
0.16
0.17
0.05
F2(2, 24) ⫽ 21.26ⴱⴱⴱ
F2(4, 96) ⫽ 30.78ⴱⴱⴱ
F2(8, 96) ⫽ 4.89ⴱⴱⴱ
0.26
0.28
0.09
F1(2, 42) ⫽ 21.70ⴱⴱⴱ
0.04
F2(2, 24) ⫽ 8.21ⴱⴱ
0.06
F1(2, 42) ⫽ 24.35ⴱⴱⴱ
0.05
F2(2, 24) ⫽ 9.85ⴱⴱⴱ
0.10
F1(2, 42) ⫽ 19.59ⴱⴱⴱ
F1(4, 84) ⫽ 16.90ⴱⴱⴱ
F1(8, 168) ⫽ 5.70ⴱⴱⴱ
0.06
0.19
0.04
F2(2, 24) ⫽ 5.91ⴱⴱ
F2(4, 96) ⫽ 23.67ⴱⴱⴱ
F2(8, 96) ⫽ 2.47ⴱ
0.10
0.32
0.07
F1(2, 42) ⫽ 17.05ⴱⴱⴱ
F1(4, 84) ⫽ 6.14ⴱⴱⴱ
F1(8, 168) ⫽ 4.05ⴱⴱⴱ
0.17
0.03
0.03
F2(2, 24) ⫽ 19.34ⴱⴱⴱ
F2(4, 96) ⫽ 4.42ⴱⴱ
F2(8, 96) ⫽ 2.22ⴱ
0.33
0.06
0.06
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
ⴱⴱⴱ
p ⬍ .001.
accuracy.4 Figure 8 compares these transformed data to the results
of Experiment 2. If the different serial position functions found
with the 2AFC and bar-probe procedures were due to a greater
number of location errors in the bar-probe task, then a decrease in
performance relative to the 2AFC data of Experiment 2 would be
expected, yet the opposite pattern was found.
This suggests that some other factor caused the shift in performance for final letters and digits when switching from a 2AFC to
a bar-probe partial report procedure. One possibility is that the
alternative items presented simultaneously with the backward
mask in Experiments 1– 4 affected performance in this task. This
could arise through meta-contrast masking (Breitmeyer & Ogmen,
2000) or through some other form of lateral interference or crowding (Huckauf & Heller, 2004), with the two alternative characters
causing an additional interference effect adding to that of the
backward mask. The interfering effect of two characters would be
greater than that caused by the two bar-probe stimuli used in
Experiment 5 because of the greater visual complexity of the
characters and their closer spatial proximity to the target item (see
Figure 1). This would therefore account for the overall lower levels
of performance in Experiment 5 compared with Experiment 2.
iment 6, and one within-subjects variable was added. This additional variable consisted of the two different report tasks. In one
block of the experiment, a 2AFC task was used. In the other block
of the experiment, a partial report task was used. The order of these
blocks was counterbalanced across participants.
Procedure. The procedure was almost identical to that of Experiment 5, except for the use of two different report tasks. The backward
mask was always accompanied by two hyphen marks positioned
above and below one of the five positions in the string. On the sixth
line below the backward mask, the task was indicated on each trial. In
the 2AFC block of the experiment, the task indication consisted of
two response alternatives, with the word ou (French for or) in between
(e.g., T ou B). Participants responded by pressing either the leftward
arrow key (for the alternative on the left) or the rightward arrow key
(for the alternative on the right) on the keyboard. In the partial report
block of the experiment, participants saw the task indication quoi
(French for what) and had to choose among the letters and symbols
indicated on the keyboard. Participants responded by pressing 1 of 18
marked response keys on the keyboard. At the start of every task
block, participants practiced the (2AFC or partial report) task during
20 trials. Both blocks consisted of 180 experimental trials. The experiment lasted approximately 30 min.
Experiment 6
Experiment 6 was designed to test the interference hypothesis by
presenting the two alternative letters in the 2AFC task at a different
location at the same time as a bar probe indicating the location for
report. Participants were tested with both procedures (2AFC and bar
probe) in two different blocks, and the same bar-probe stimulus was
used to indicate position of report in both procedures.
Method
Participants. Twenty-two students (5 men and 17 women) at
the University of Provence, Marseille, were paid 5€ to participate
in the experiment. Their mean age was 21 years, and all reported
having normal or corrected-to-normal vision.
Stimuli and design. Stimuli and design were the same as in
Experiment 2, except that there were no digit characters in Exper-
Results
One participant had to be excluded from the analyses because of
very low performance. For the remaining 21 participants, mean
accuracy for each target type, target position, and report task is
presented in Figure 9. The results of a 2 (target type) ⫻ 5 (target
position) ⫻ 2 (report task) ANOVA on the accuracy data are
shown in Table 8. Letter targets were identified more accurately
4
Performance on 2AFC can be estimated from partial report performance by assuming that in the 2AFC task, participants either identify the
target and respond correctly or fail to identify the target and guess. Under
these assumptions, 2AFC performance (FC) is calculated from percentage
correct partial report performance (PR) with the following equation: FC ⫽
PR ⫹ [(100 ⫺ PR)/2].
SERIAL POSITION EFFECTS
491
Figure 7. Serial position functions for letters, symbols, and digits in Experiment 5 after categorizing errors into
location errors (left) and intrusion errors (right). Vertical bars represent standard errors.
than symbol targets, with 62.5% and 51.4% correct identifications,
respectively. There was a main effect of target position, and
performance was higher with the 2AFC procedure compared with
(uncorrected) partial report. The three-way interaction and all
two-way interactions were significant. The interactions with task
simply reflect the fact that all effects were exaggerated in the
partial report task given that the possible range of accuracy was
twice as large with this procedure. In both tasks, there was a
significantly more pronounced W-shaped serial position function
for letter targets compared with symbol targets, and this pattern
was more exaggerated with the partial report procedure.
Trends analyses of the 2AFC data for letter targets revealed a
significant linear effect, F1(1, 20) ⫽ 4.58, p ⬍ .05; F2(1, 8) ⫽
6.48, p ⬍ .05, quadratic effect, F1(1, 20) ⫽ 4.80, p ⬍ .05; F2(1,
8) ⫽ 10.87, p ⬍ .05, and quartic effect, F1(1, 20) ⫽ 39.87, p ⬍
.001; F2(1, 8) ⫽ 41.10, p ⬍ .001. These accounted for 12%, 16%,
and 72% of the variance, respectively. For symbols in the 2AFC
task block, the quadratic and quartic functions explained the main
part of the observed variance, with 49% and 48%, respectively, but
these trends were not significant. For the partial report data with
letter targets, there was a significant linear effect, F1(1, 20) ⫽ 7.77,
p ⬍ .05; F2(1, 8) ⫽ 6.48, p ⬍ .05, a significant quadratic effect,
F1(1, 20) ⫽ 50.24, p ⬍ .001; F2(1, 8) ⫽ 30.01, p ⬍ .001, and a
significant quartic effect, F1(1, 20) ⫽ 77.42, p ⬍ .001; F2(1, 8) ⫽
50.43, p ⬍ .001. These trends explained 10%, 49%, and 38% of
the variance, respectively. For symbols, there was a significant
quartic effect, F1(1, 20) ⫽ 21.66, p ⬍ .001; F2(1, 8) ⫽ 26.08, p ⬍
.001, which accounted for about 90% of the observed variance.
The errors in the partial report task can again be categorized into
intrusion and location errors. Performance for letter and symbol
targets is broken down by error type in Figure 10, and the results
of separate ANOVAs are shown in Table 8. When only location
errors counted as errors, mean accuracies for letters and symbols
were 69.4% and 62.1%, respectively. The pattern was the same as
the overall analysis with the critical Target Type ⫻ Target Position
interaction being driven largely by the lower levels of accuracy for
symbols as opposed to letters at Positions 1 and 5. When only
intrusion errors counted as errors, mean accuracy for letter targets
was greater than for symbol targets, with 86.6% and 77.0% correct
identifications, respectively. The effects of target position were
less pronounced, but the critical interaction with target type was
significant and was again due to the superior performance for
letters compared with symbols at Positions 1 and 5.
Discussion
The results of Experiment 6 are clear-cut and provide a solution
for integrating the apparently diverging results obtained in the
previous experiments. Performance on the bar-probe procedure
was an almost exact replication of performance in Experiment 5,
Figure 8. Comparison of the transformed PR data of Experiment 5 with the 2AFC results of Experiment 2. PR data
are transformed into 2AFC data by applying the equation described in footnote 4. PR ⫽ partial report; Expt. ⫽
Experiment; 2AFC ⫽ two-alternative forced choice.
TYDGAT AND GRAINGER
492
Figure 9. Serial position functions for letters and symbols in Experiment 6 with a 2AFC procedure (left) and
a PR procedure (right). In both cases, position of report was indicated by bar probes. Vertical bars represent
standard errors. 2AFC ⫽ two-alternative forced choice; PR ⫽ partial report.
thus demonstrating the robustness of the effects. Performance on
the 2AFC procedure, on the other hand, differed from the pattern
seen in Experiments 1–3, because there was a clear final-position
advantage for letter stimuli. Because the only procedural difference in Experiment 6 was the way in which the two alternative
letters were presented, this strongly suggests that interference from
the alternative letters was the critical factor at play. Therefore, the
absence of a final-position advantage for letter stimuli in Experiments 1–3 was likely due to the extra interference generated by the
two alternative letters at this specific position. In the General
Discussion, we describe in detail exactly how one simple mechanism can account for this surprising selective effect of interference
from the two alternatives in the 2AFC task.
Furthermore, in the absence of interference from the alternative
letters, performance was very similar across the two procedures
when applying the equation for transforming partial report into
approximate 2AFC accuracy described in footnote 4. Figure 11
shows the transformed bar-probe data in comparison with the
2AFC data. There was still a small advantage for the partial report
procedure, but this can be accounted for by assuming that partic-
ipants also selected randomly from the nine alternative letters or
symbols when they could not identify the target but knew which
category it belonged to. This small advantage for performance with
the partial report procedure appears to have been greatest for
performance at the first position. An analysis of the different types
of error (location vs. intrusion; see Figure 10) suggests that this
might have been due to the reduced number of location errors at
this position in the partial report bar-probe task.
General Discussion
In the present study, we examined serial position functions for
letters, digits, and symbols, using data-limited identification paradigms. Our results confirm previous results obtained with a
response-limited target search task (Hammond & Green, 1982;
Mason, 1982; Mason & Katz, 1976), showing distinct serial position functions for symbol stimuli compared with both letters and
digits. These results strongly suggest that the distinct serial position functions reflect fundamental differences in how strings of
Table 8
Main Statistical Results for Experiment 6 Obtained With the Two-Alternative Forced-Choice (2AFC) Procedure and the Partial
Report (PR) Bar-Probe Procedure
Effect
Target type
Target position
Task
Target Type ⫻ Target Position
Target Type ⫻ Task
Target Position ⫻ Task
Target Type ⫻ Target Position ⫻ Task
2AFC task: Target Type ⫻ Target Position
PR task: Target Type ⫻ Target Position
Location errors (PR task)
Target type
Target position
Target Type ⫻ Target Position
Intrusion errors (PR task)
Target type
Target position
Target Type ⫻ Target Position
ⴱ
p ⬍ .05.
ⴱⴱ
p ⬍ .01.
␩2
F1
ⴱⴱⴱ
p ⬍ .001.
␩2
F2
57.12ⴱⴱⴱ
12.85ⴱⴱⴱ
53.93ⴱⴱⴱ
20.82ⴱⴱⴱ
22.87ⴱⴱⴱ
9.85 ⴱⴱⴱ
3.48ⴱ
6.92ⴱⴱⴱ
16.72ⴱⴱⴱ
0.06
0.09
0.17
0.05
0.02
0.02
0.01
0.06
0.08
F2(1,
F2(4,
F2(1,
F2(4,
F2(1,
F2(4,
F2(4,
F2(4,
F2(4,
F1(1, 20) ⫽ 12.59ⴱⴱ
F1(4, 80) ⫽ 15.68ⴱⴱⴱ
F1(4, 80) ⫽ 7.72ⴱⴱⴱ
0.04
0.23
0.06
F2(1, 16) ⫽ 2.06
F2(4, 64) ⫽ 14.29ⴱⴱⴱ
F2(4, 64) ⫽ 3.84ⴱⴱ
0.05
0.25
0.07
F1(1, 20) ⫽ 17.82ⴱⴱⴱ
F1(4, 80) ⫽ 1.50
F1(4, 80) ⫽ 4.60ⴱⴱ
0.10
0.01
0.04
F2(1, 16) ⫽ 10.03ⴱⴱ
F2 ⬍ 1
F2(4, 64) ⫽ 2.52ⴱ
0.19
0.02
0.07
F1(1,
F1(4,
F1(1,
F1(4,
F1(1,
F1(4,
F1(4,
F1(4,
F1(4,
20)
80)
20)
80)
20)
80)
80)
80)
80)
⫽
⫽
⫽
⫽
⫽
⫽
⫽
⫽
⫽
16)
64)
16)
64)
16)
64)
64)
64)
64)
⫽
⫽
⫽
⫽
⫽
⫽
⫽
⫽
⫽
7.12ⴱ
16.03ⴱⴱⴱ
105.52ⴱⴱⴱ
8.97ⴱⴱⴱ
9.67ⴱⴱ
8.93ⴱⴱⴱ
2.83ⴱ
4.40ⴱⴱ
8.42ⴱⴱ
0.08
0.13
0.24
0.07
0.02
0.03
0.01
0.09
0.12
SERIAL POSITION EFFECTS
493
Figure 10. Serial position functions for letters and symbols in the bar-probe partial report task of Experiment
6 after categorizing the errors into location errors (left) and intrusion errors (right). Vertical bars represent
standard errors.
letters, digits, and symbols are processed and are not the consequence of how a particular laboratory task is performed.
Experiment 1 found distinct serial position functions for symbol
stimuli compared with both letters and digits in a 2AFC identification task with stimuli similar to those tested by Mason (1982).
Experiments 2– 4 were designed to test several different interpretations of the different serial positions functions found in the 2AFC
task of Experiment 1 and, in particular, examined the fact that most
of this difference occurred at the first position. The same pattern
was found when randomly mixing stimulus types (Experiment 2),
thus suggesting that the different serial position functions were not
caused by an endogenous attentional bias generated when expecting a particular type of stimulus (directing attention to the beginning of the string, e.g.). The same pattern was also found in
Experiment 3 when target letters were embedded in strings of
symbols and vice versa, thus suggesting that the different serial
position functions were not caused by an exogenous attentional
bias automatically triggered on presentation of a string of letters or
digits (again orienting attention to the beginning of the string). The
results of Experiment 3 also suggest that the different serial position functions were not caused by a chunking mechanism that
attempts to create combinations of letters or digits. Any attempt to
form higher order letter combinations should have been severely
disrupted by having symbol stimuli as the surrounding context.
Contrary to the results obtained with the target search task, the
results obtained with the 2AFC task in Experiments 1–3 showed
that the difference between symbols and letters/digits mostly concerned performance for targets presented at the initial position of
strings. Experiments 5 and 6 were designed to examine why we did
not find a final-position advantage for letters and digits in the
2AFC task. Indeed, prior research examining serial position functions for letter stimuli with a bar-probe identification procedure
(postcued partial report) systematically found a W-shaped serial
position function (Haber & Standing, 1969; Merikle, Lowe, &
Coltheart, 1971; Mewhort & Campbell, 1978; Mewhort et al.,
1981, 1993). Experiment 5 found exactly this pattern for letter and
digit stimuli and further demonstrated that superior performance
for such stimuli compared with symbols arose at both the first and
the final positions in the string. Experiment 6 replicated this result
for letter and symbol stimuli and further demonstrated that the
same pattern can be obtained with a 2AFC procedure when the
alternative letters are displaced relative to the target string and a
bar-probe is used to indicate the position of report.
On the basis of the results of Experiment 6, we concluded that
the absence of a final-position advantage for letters and digits in
Experiments 1–3 was due to additional interference from the
alternative letters displayed with the postmask. This conclusion is
consistent with prior findings showing that the outer-position
Figure 11. Comparison of the transformed PR data (left) and 2AFC data (right) of Experiment 6. PR data were
transformed into 2AFC data by applying the equation described in footnote 4. PR ⫽ partial report; Expt. ⫽
Experiment; 2AFC ⫽ two-alternative forced choice.
494
TYDGAT AND GRAINGER
advantage for letter stimuli in the bar-probe task is diminished in
the presence of a backward mask that extends beyond the outer
positions (Mewhort & Campbell, 1978) and that adding parentheses before and after the letter string greatly reduces the outerposition advantage, particularly for the final position (Haber &
Standing, 1969). Finally, Experiments 5 and 6 provided evidence
that the presence of location errors in the bar-probe technique
(when another item in the display was reported instead of the
target) exaggerates the outer-position advantage compared with
the 2AFC procedure.
Next, we argue that the pattern of results found in the present
study can be understood by considering the constraints imposed on
a specialized system for processing strings of letters or digits. The
key mechanism behind this account is crowding (see Levi, 2008,
for a recent review), and the central hypothesis is that crowding
affects letter and digit stimuli differently than symbol stimuli.
There are two main differences, which we discuss in detail. One
involves a reduction in crowding effects by means of a change in
the size of the receptive fields of letter and digit detectors, and the
other involves a change in the shape of these receptive fields (see
Figure 12).
Crowding, Stimulus Type, and Serial Position
We hypothesize that when children learn to read words and
numbers, they develop a specialized system that is custom built to
handle the very specific nature of these stimuli (see McCandliss,
Cohen, & Dehaene, 2003, for one version of this hypothesis). Most
notably, this system would develop to optimize processing in the
extremely crowded conditions that arise with printed words and
numbers, compared with other visual objects that do not typically
occur in such a cluttered environment. One good candidate for the
specialized mechanism for processing strings of letters and digits
is Grainger and van Heuven’s (2003) alphabetic array (see Figure
12). This is a bank of alphabetic character detectors that performs
parallel independent character recognition. In this model, alphabetic character detectors simultaneously process all characters in a
string of limited length. Each character detector signals the pres-
Figure 12. Illustration of part of Grainger and van Heuven’s (2003)
model focusing on the receptive field structure of retinotopic letter detectors in the alphabetic array. The size and shape of the receptive fields is
hypothesized to change as a function of eccentricity and visual field, with
receptive field size increasing with eccentricity and becoming asymmetrical toward the left in the left visual field. The example shows processing
of a random string of consonants with fixation on the central letter.
ence of a given character at a particular location along the horizontal meridian (i.e., a retinotopic map). These location-specific
detectors then feed activation onto location-invariant orthographic
representations. It should be noted that the alphabetic array in
Grainger and van Heuven’s model is part of a long tradition of
independent parallel-channels models of letter string processing
(Estes, 1972; Gardner, 1973; McClelland & Rumelhart, 1981;
Rumelhart, 1970; Shiffrin & Geisler, 1973; Wolford, 1975). Furthermore, the notion of overlapping receptive fields for letter
detectors is also a central component in Dehaene, Cohen, Sigman,
and Vinckier’s (2005) account of visual word recognition, and
similar ideas can be found in the early models of Estes (1972),
Rumelhart (1970), and Wolford (1975). An obvious extension of
this account, in light of the present data, is that numbers are
processed with the same mechanism as letter strings (in which case
the term alphanumeric array would be more appropriate).
The central hypothesis here is that receptive field size of retinotopic letter and digit detectors has adapted to the need to
optimize processing of strings of letters and digits. The smaller the
receptive field size of these detectors, the less interference from
neighboring characters. This is the basic idea behind Bouma’s
(1970) functional field and Pelli et al.’s (2004) integration field, as
measured with a standard crowding manipulation.5 Thus, the hypothesized smaller receptive field sizes for letters and digits (Figure 12) suggest that it is a reduction in the spatial extent of
crowding for letters and digits that is the major factor driving the
different serial position functions for these stimuli compared with
symbols. The larger receptive field size for symbol detectors
would generate close to maximum crowding with only a single
flanking symbol. In this way, the first and last symbols in an array
of symbols are subject to practically the same amount of crowding
as the inner letters. Letters and digits, on the other hand, would be
subject to lower levels of crowding that would depend on the
number of flanking characters (one or two) because of the hypothesized narrower receptive fields of the detectors in the alphabetic
array. This, therefore, explains the classic W-shaped serial position
function for letters and digits.
The different spatial extent of crowding on letter, digit, and
symbol stimuli also accounts for why the 2AFC task (Experiments
1–3) did not show the standard final-position advantage for letters
and explains why this advantage was reinstated when the alternative letters were displaced relative to the target position (Experiment 6). Inner letters in a string are assumed to suffer close to
maximal crowding, whereas outer letters do not. Therefore, the
masking effects of the alternative letters in 2AFC (when positioned
next to the target) mostly affect performance at the end positions.
Furthermore, because symbols are hypothesized to suffer maximal
5
Some particularly relevant data came to our attention during the
preparation of a revision of this work. Kwon and Legge (2006) have shown
that crowding effects are stronger in children than in skilled adult readers.
This supports the notion that reduced crowding is a key ingredient in the
optimization of printed word perception. Further support for this hypothesis was provided by Kwon, Legge, and Dubbels (2007), who demonstrated that development of reading speed is linked to development of the
size of the visual span, which has in turn been shown to be determined by
crowding (Pelli et al., 2007). Finally, recent research has shown that the
spatial extent of crowding can be reduced by training observers to identify
letters closely flanked by other letters (Chung, 2007).
SERIAL POSITION EFFECTS
crowding from a single flanking stimulus, the presence of the two
alternative letters has little additional influence on the processing
of these stimuli. This account therefore explains the observed
variation in the final-position advantage for letter targets as a
function of testing procedure but does not account for the robust
advantage for the initial position that occurs over and above such
procedural differences. We return to discuss this initial-position
advantage later.
Surprisingly, little research has used the standard flanker paradigm with eccentric target locations to directly investigate the
effect of number of flanking stimuli (e.g., Bouma, 1970). We
therefore set about running these experiments ourselves with the
same set of letters and symbols as tested in the present study
(Grainger & Tydgat, 2008). Targets were presented at eccentricities equal to the first and last positions of strings in the present
study, either in isolation or accompanied by one or two flanking
characters. The results revealed that performance for letter targets
with a single flanker was closer to the no-flanker condition than to
the two-flanker condition, whereas symbol targets showed practically as much interference from a single flanker as from two
flankers and showed overall greater interference from flankers
than did letter targets. This pattern nicely confirms the predictions
derived from our crowding interpretation of the results of the
present study. Symbols are more prone to crowding effects than
are letters, and although symbols show close to maximal interference in the presence of a single flanker, letters, on the other hand,
show very little interference from a single flanker.
The Initial-Position Advantage: Sequential Processing or
Specialized Detectors?
In this section, we discuss some other mechanisms that might be
at least partly responsible for the present data pattern. The initialposition advantage found in the present study might be the reflection of a serial processing mechanism specific to stimuli that
typically appear in strings. According to such an account, processing moves continuously from the beginning of the string to the
end; therefore, items at the beginning of the string are processed
before items at the end of the string (e.g., the scanning model of
Mewhort & Popham, 1991). Whitney’s (2001) SERIOL (sequential encoding regulated by inputs to oscillations within letter units)
model provides a recent example of such an approach. In this
model, a beginning-to-end activation gradient forms the starting
point for orthographic processing (coding letter identities and letter
positions). According to Whitney, the visual acuity function
present in early visual areas (V1/V2) is transformed into a monotonically decreasing activation gradient (called the locational gradient) at the level of feature representations in V4. This locational
gradient is then imposed on a set of position-independent letter
representations, such that the first letter in the string has the highest
activation, and activation diminishes across the string until the
final letter. The activation levels of letter representations determine
their order of firing, which then determines the formation of
ordered letter sequences (contiguous and noncontiguous bigrams)
at the bigram level of representation in the model.
Whitney’s (2001) account therefore predicts a strong linear
component in the identification of strings of letters, digits, and
symbols, because the activation gradient is already present at the
level of feature representations. The present results would appear
495
to rule out this possibility. Our results therefore suggest that, if
there is indeed such a sequential processing bias, then it is most
likely initiated at the level of representations in the alphabetic
array. This could be achieved by a spatial attention mechanism that
would impose a beginning-to-end activation gradient on letter
representations. Grainger et al. (2006) suggested that such a mechanism would be particularly important for the correct assignment
of grapheme–phoneme correspondences as in Perry, Ziegler, and
Zorzi’s (2007) model of reading aloud. This would fit into a
generic dual-route framework for visual word recognition
(Grainger, Diependaele, Spinelli, Ferrand, & Farioli, 2003;
Grainger & Ziegler, 2007) in which the alphabetic array feeds
information into two different processing routes: a direct orthographic route for semantic access and an indirect route that implements sublexical spelling-to-sound conversion (e.g., through
graphemes and phonemes).
However, several aspects of the present results appear to rule out
a beginning-to-end processing bias, whatever the precise nature of
the bias. First, the results of Experiments 2 and 3 appear to rule out
endogenous and exogenous attentional biases specific to strings of
letters and digits (as discussed earlier). Second, the results of
Experiments 5 and 6 reinstated a final-position advantage for
letters and digits. This led to a diminution in the linear component
and to an increase in the quartic component in the trends analyses.
Third, performance for letter and digit stimuli at Positions 2 and 4
was always very similar across all experiments, suggesting that the
linear trends were being driven almost entirely by performance at
the first position. Overall, the results of the present experiments are
more in favor of an advantage specifically for the first position of
strings of letters or digits, rather than the combination of a
beginning-to-end processing bias with an outer-position advantage
(because of reduced crowding).
Within an integrated account of the complete set of results
reported here, we hypothesize that not only do the receptive fields
of letter and digit detectors undergo a change in size because of
experience with reading words and numbers, but they are also
subject to a change in shape. This hypothesized change in shape of
receptive fields (see Figure 12) would arise to optimize processing
at the first position in strings of letters and digits. Letters in the first
position tend to provide more constraint on lexical identity than
letters in any other position in the string (e.g., Clark & O’Regan,
1999; Grainger & Jacobs, 1993). Furthermore, the first position is
critical for translating an orthographic code into a phonological
code, because the correct graphemes (letters and letter clusters
corresponding to a phoneme) can be computed only with precise
order information (Perry et al., 2007). Such optimization can be
achieved by asymmetric receptive fields that are elongated in the
direction of the initial position (i.e., to the left for languages read
from left to right). Letter and digit detectors with receptive fields
of this shape mostly suffer interference from characters immediately to their left, hence endowing an advantage for the initial
position in strings. We further hypothesize that such optimization
operates only for stimuli that are fixated and therefore only for
detectors receiving input from the LVF.
This visual-field specificity of the shape of receptive fields of
letter and digit detectors accounts for why the initial-position
advantage (LVF) is systematically larger than the final-position
advantage (RVF). If one further assumes no change in the size
(surface area) of receptive fields as a function of visual field, then
496
TYDGAT AND GRAINGER
the elongated receptive fields in the LVF would have smaller
vertical extent (because of the larger horizontal extent) than the
more circular receptive fields of the RVF. This would explain why
letters and digits at the initial position are less sensitive to the
presence of the two alternative letters in 2AFC than are letters and
digits at the final position. Bouma (1973) was among the first to
report visual field differences in letter identification in unpronounceable nonwords (note that only the first and last positions
were tested in Bouma’s study). For a fixed eccentricity, there was
a strong advantage for initial over final position in the LVF, which
diminished and even reversed (final position superior to initial
position) in the RVF.6 This led Bouma to suggest that “the spatial
extent of foveally oriented masking is smaller in the R field than
the Left” (p. 775). This is in fact an alternative version of the
present hypothesis. Because there is evidence for an inward–
outward asymmetry in crowding effects from single flankers
(Banks, Bachrach, & Larson, 1977; Banks, Larson, & Prinzmetal,
1979; Bouma, 1970; Chastain, 1982; Krumhansl & Thomas, 1977;
White, 1976), this would suggest that receptive fields are elongated to the left in the LVF and to the right in the RVF (Bouma,
1978). Thus, our hypothesized change in shape of receptive fields
for letters and digits could involve a general shift to the left
operating over this initial asymmetry. This would generate an
exaggerated elongation to the left in the LVF and a more symmetrical receptive field shape in the RVF.
Our own recent crowding experiments (Grainger & Tydgat,
2008) have provided independent support for this hypothesized
visual-field specificity of receptive fields. Letter targets presented
in the LVF or RVF (at the same eccentricity as the outer letters in
the present experiments), and flanked by a single letter either
immediately to the right or to the left of the target, showed an
asymmetric pattern of interference. Flanker position affected performance in the LVF but not in the RVF. That is, targets presented
in the LVF flanked by another letter to the left were harder to
identify than targets at the same location flanked by a letter to the
right, whereas targets in the RVF showed equal amounts of interference from left and right flankers. This is exactly the pattern
predicted by our account of the present set of results.
Conclusions
The present study examined the identification accuracy of letter,
digit, and symbol targets presented in strings. In line with prior
research with speeded target detection, the obtained serial position
functions were similar for letters and digits and were different for
symbols. In Experiments 1–3 we used a 2AFC procedure, and
most of this difference occurred for performance at the first position in strings, with much higher levels of performance for letters
and digits compared with symbols at this particular location. We
showed that this first-position advantage for letters and digits
compared with symbols was still present when the different types
of target were randomly intermixed in the experiment, as opposed
to being blocked by stimulus type. Furthermore, the initial-position
advantage for letters compared with symbols was present even
when letter targets were embedded in symbol strings and vice
versa. Experiments 5 and 6 showed that a final-position advantage
(and hence the classic W-shaped serial position function) was
obtained for letters and digits when the alternative letters presented
for 2AFC were replaced by a bar-probe stimulus. Our results
suggest that the constraints imposed by reading words and numbers force an adaptive modification of basic visual processes to
optimize identification of alphanumeric stimuli. This adaptive
modification is hypothesized to involve a change in size and a
change in shape of the receptive fields of retinotopic letter and
digit detectors.
6
Other work has examined visual field effects in letter identification in
four-letter strings (e.g., Jordan et al., 2003). In Jordan et al.’s (2003) study,
eccentricity was matched for the whole string, such that initial and final
letters had different eccentricities. In these conditions, initial and final
letters were identified with equal accuracy in both visual fields. Under the
modified receptive field hypothesis, this symmetrical effect is likely due to
the trade-off between crowding and eccentricity. The initial position in the
LVF suffers less crowding but is more eccentric than the initial position in
the RVF, and the final position in the RVF suffers less crowding but is
more eccentric than the final position in the LVF.
References
Averbach, E., & Coriell, A. S. (1961). Short-term memory in vision. Bell
Telephone Technical Journal, 40, 19 –31.
Banks, W. P., Bachrach, K., & Larson, D. W. (1977). Asymmetry of lateral
interference in visual letter identification. Perception & Psychophysics,
22, 232–240.
Banks, W. P., Larson, D. W., & Prinzmetal, W. (1979). Asymmetry of
visual interference. Perception & Psychophysics, 25, 447– 456.
Bouma, H. (1970). Interaction effects in parafoveal letter recognition.
Nature, 226, 177–178.
Bouma, H. (1973). Visual interference in the parafoveal recognition of
initial and final letters of words. Visual Research, 13, 767–782.
Bouma, H. (1978). Visual search and reading: Eye movements and functional visual field: A tutorial review. In J. Requin (Ed.), Attention and
performance VII (pp. 115–147). Hillsdale, NJ: Erlbaum.
Breitmeyer, B. G., & Ogmen, H. (2000). Recent models and findings in
visual backward masking: A comparison, review, and update. Perception & Psychophysics, 62, 1572–1595.
Butler, B. E. (1975). Selective attention and target search with brief visual
displays. Quarterly Journal of Experimental Psychology, 27, 467– 477.
Butler, B. E., & Merikle, P. M. (1973). Selective masking and processing
strategy. Quarterly Journal of Experimental Psychology, 25, 542–548.
Campbell, A. J., & Mewhort, D. J. (1980). On familiarity effects in visual
information processing. Canadian Journal of Psychology, 34, 134 –154.
Chastain, G. (1982). Confusability and interference between members of
parafoveal letter pairs. Perception & Psychophysics, 32, 576 –580.
Chung, S. T. L. (2007). Learning to identify crowded letters: Does it
improve reading speed? Vision Research, 47, 3150 –3159.
Clark, J. J., & O’Regan, J. K. (1999). Word ambiguity and the optimal
viewing position in reading. Vision Research, 39, 843– 857.
Dehaene, S., Cohen, L., Sigman, M., & Vinckier, F. (2005). The neural
code for written words: A proposal. Trends in Cognitive Sciences, 9,
335–341.
Estes, W. K. (1972). Interactions of signal and background variables in
visual processing. Perception & Psychophysics, 12, 278 –286.
Estes, W. K., Allmeyer, D. H., & Reder, S. M. (1976). Serial position
functions for letter identification at brief and extended exposure durations. Perception & Psychophysics, 19, 1–15.
Gardner, G. T. (1973). Evidence for independent parallel channels in
tachistoscopic perception. Cognitive Psychology, 4, 130 –155.
Grainger, J. (2008). Cracking the orthographic code: An introduction.
Language and Cognitive Processes, 23, 1–35.
Grainger, J., Diependaele, K., Spinelli, E., Ferrand, L., & Farioli, F. (2003).
SERIAL POSITION EFFECTS
Masked repetition and phonological priming within and across modalities. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 1256 –1269.
Grainger, J., Granier, J. P., Farioli, F., Van Assche, E., & van Heuven,
W. J. (2006). Letter position information and printed word perception:
The relative-position priming constraint. Journal of Experimental Psychology: Human Perception and Performance, 32, 865– 884.
Grainger, J., & Jacobs, A. M. (1993). Masked partial-word priming in
visual word recognition: Effects of positional letter frequency. Journal
of Experimental Psychology: Human Perception and Performance, 19,
951–964.
Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual
word recognition: A multiple read-out model. Psychological Review,
103, 518 –565.
Grainger, J., & Tydgat, I. (2008). Crowding affects letters and symbols
differently. Manuscript in preparation.
Grainger, J., & van Heuven, W. (2003). Modeling letter position coding in
printed word perception. In P. Bonin (Ed.), The mental lexicon (pp.
1–24). New York: Nova Science.
Grainger, J., & Ziegler, J. (2007). Cross-code consistency effects in visual
word recognition. In E. L. Grigorenko & A. Naples (Eds.), Single-word
reading: Biological and behavioral perspectives (pp. 129 –157). Mahwah,
NJ: Erlbaum.
Greenhouse, S., & Geisser, S. (1959). On methods in the analysis of profile
data. Psychometrika, 24, 95–112.
Haber, R. N., & Standing, L. (1969). Location of errors with a poststimulus
indicator. Psychonomic Science, 17, 345–346.
Hammond, E. J., & Green, D. W. (1982). Detecting targets in letter and
non-letter arrays. Canadian Journal of Psychology, 36, 67– 82.
Huckauf, A., & Heller, D. (2004). On the relations between crowding and
visual masking. Perception & Psychophysics, 66, 584 –595.
Jordan, T. R., & Bevan, K. M. (1996). Position-specific masking and
word–letter phenomenon: Reexamining the evidence from the Reicher–
Wheeler paradigm. Journal of Experimental Psychology: Human Perception and Performance, 22, 1416 –1433.
Jordan, T. R., Patching, G. R., & Milner, A. D. (2000). Lateralized word
recognition: Assessing the role of hemispheric specialization, modes of
lexical access, and perceptual asymmetry. Journal of Experimental
Psychology: Human Perception and Performance, 26, 1192–1208.
Jordan, T. R., Patching, G. R., & Thomas, S. M. (2003). Assessing the role
of hemispheric specialization, serial-position processing, and retinal
eccentricity in lateralised word recognition. Cognitive Neuropsychology,
20, 49 –71.
Krumhansl, C. L., & Thomas, E. A. C. (1976). Extracting identity and
location information from briefly presented letter arrays. Perception &
Psychophysics, 20, 243–258.
Krumhansl, C. L., & Thomas, E. A. C. (1977). Effect of level of confusability on reporting letters from briefly presented visual displays. Perception & Psychophysics, 21, 269 –279.
Ktori, M., & Pitchford, N. J. (2008). Effect of orthographic transparency on
letter position encoding: A comparison of Greek and English monoscriptal
and biscriptal readers. Language and Cognitive Processes, 23, 258 –281.
Kwon, M. Y., & Legge, G. E. (2006). Developmental changes in the size
of the visual span for reading: Effects of crowding [Abstract]. Journal of
Vision, 6, 1003a.
Kwon, M. Y., Legge, G. E., & Dubbels, B. R. (2007). Developmental
changes in the visual span for reading. Vision Research, 47, 2889 –2890.
Levi, D. M. (2008). Crowding—An essential bottleneck for object recognition: A mini-review. Vision Research, 48, 635– 654.
Mason, M. (1975). Reading ability and letter search time: Effects of
orthographic structure defined by single-letter positional frequency.
Journal of Experimental Psychology, 104, 146 –166.
Mason, M. (1982). Recognition time for letters and nonletters: Effects of
497
serial position, array size, and processing order. Journal of Experimental
Psychology: Human Perception and Performance, 8, 724 –738.
Mason, M., & Katz, L. (1976). Visual processing of nonlinguistic strings:
Redundancy effects and reading ability. Journal of Experimental Psychology, 105, 338 –348.
McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The visual word form
area: Expertise for reading in the fusiform gyrus. Trends in Cognitive
Sciences, 13, 155–161.
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation
model of context effects in letter perception: Part 1. An account of basic
findings. Psychological Review, 88, 375– 407.
Merikle, P. M. (1974). Selective backward masking with an unpredictable
mask. Journal of Experimental Psychology, 103, 589 –591.
Merikle, P. M., & Coltheart, M. (1972). Selective forward masking. Canadian Journal of Psychology, 26, 296 –302.
Merikle, P. M., Coltheart, M., & Lowe, D. G. (1971). On the selective
effects of a patterned masking stimulus. Canadian Journal of Psychology, 25, 264 –279.
Merikle, P. M., Lowe, D. G., & Coltheart, M. (1971). Familiarity and
method of report as determinants of tachistoscopic performance. Canadian Journal of Psychology, 25, 167–174.
Mewhort, D. J. K., & Campbell, A. J. (1978). Processing spatial information and the selective-masking effect. Perception & Psychophysics, 24,
93–101.
Mewhort, D. J. K., Campbell, A. J., Marchetti, F. M., & Campbell, J. I. D.
(1981). Identification, localization, and “iconic memory”: An evaluation
of the bar-probe task. Memory & Cognition, 9, 50 – 67.
Mewhort, D. J. K., Huntley, M. F., & Duff-Fraser, H. (1993). Masking
disrupts recovery of location information. Perception & Psychophysics,
54, 759 –762.
Mewhort, D. J. K., & Popham, D. (1991). Serial recall of tachistoscopic
letter strings. In W. E. Hockley & S. Lewandowsky (Eds.), Relating
theory and data: Essays on human memory in Honor of Bennet B.
Murdock (pp. 425– 443). Hillsdale, NJ: Erlbaum.
Nazir, T. A., Ben-Boutayab, N., Decoppet, N., Deutsch, A., & Frost, R.
(2004). Reading habits, perceptual learning, and recognition of printed
words. Brain and Language, 88, 294 –311.
Pelli, D. G., Palomares, M., & Majaj, N. J. (2004). Crowding is unlike
ordinary masking: Distinguishing feature integration from detection.
Journal of Vision, 4, 1136 –1169.
Pelli, D. G., Tillman, K. A., Freeman, J., Su, M., Berger, T. D., & Majaj,
N. J. (2007). Crowding and eccentricity determine reading rate. Journal
of Vision, 7, 1–36.
Perry, C., Ziegler, J. C., & Zorzi, M. (2007). Nested incremental modeling
in the development of computational theories: The CDP⫹ model of
reading aloud. Psychological Review, 114, 273–315.
Pitchford, N. J., Ledgeway, T., & Masterson, J. (2008). Effect of orthographic processes on letter position encoding. Journal of Research in
Reading, 31, 97–116.
Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 28,
115–121.
Rumelhart, D. E. (1970). A multicomponent theory of the perception of
briefly exposed visual displays. Journal of Mathematical Psychology, 7,
191–218.
Rumelhart, D. E., & McClelland, J. L. (1982). An interactive activation
model of context effects in letter perception: Part 2. The contextual
enhancement effect and some tests and extensions of the model. Psychological Review, 89, 60 –94.
Schwantes, F. M. (1978). Stimulus position functions in tachistoscopic
identification tasks: Scanning, rehearsal, and order of report. Perception
& Psychophysics, 23, 219 –226.
Shiffrin, R. M., & Geisler, W. (1973). Visual recognition in a theory of
information processing. In R. Solso (Ed.), Contemporary issues in
498
TYDGAT AND GRAINGER
cognitive psychology: The Loyola Symposium (pp. 53–101). Washington, DC: Winston.
Stevens, M., & Grainger, J. (2003). Letter visibility and the viewing
position effect in visual word recognition. Perception & Psychophysics,
65, 133–151.
Tramer, O., Butler, B. E., & Mewhort, D. J. (1985). Evidence for scanning
with unilateral visual presentation of letters. Brain and Language, 25, 1–18.
Van der Heijden, A. H. C. (1992). Selective attention in vision. London:
Routledge.
Wheeler, D. D. (1970). Processes in word recognition. Cognitive Psychology, 1, 59 – 85.
White, M. J. (1976). Order of processing in visual perception. Canadian
Journal of Psychology, 30, 140 –156.
Whitney, C. (2001). How the brain codes the order of letters in a printed
word: The SERIOL model and selective literature review. Psychonomic
Bulletin and Review, 8, 221–243.
Wolford, G. (1975). Perturbation model for letter identification. Psychological Review, 82, 184 –199.
Wolford, G., & Hollingsworth, S. (1974). Retinal location and string
position as important variables in visual information processing. Perception & Psychophysics, 16, 437– 442.
Received February 15, 2007
Revision received April 9, 2008
Accepted April 28, 2008 䡲