Coherence of the Irrelevant-Sound Effect: Individual Profiles of Short-term Memory and Susceptibility to Task-Irrelevant Materials

Abstract

We examined individual and developmental differences in the disruptive effects of irrelevant sounds on serial recall of printed lists. Experiment 1 examined adults (N=205) receiving 8-item lists to be recalled. Although susceptibility to disruption of recall by irrelevant sounds was only slightly related to memory span, regression analyses documented highly reliable individual differences in this susceptibility, across speech and tone distractors, even with variance from span level removed. Experiment 2 included adults (n=64) and 8-year-old children (n=63) receiving lists of a length equal to a predetermined span and one item shorter (span - 1). We again found significant relationships between measures of span and susceptibility to irrelevant sounds, although only in two of the measures of span. We conclude that some of the cognitive processes helpful in performing a span task may not be beneficial in the presence of irrelevant sounds.

The disruptive effects of irrelevant sounds on immediate serial recall are of considerable interest. This disruption is typically referred to as the irrelevant speech effect (Colle & Welsh, 1976; Salamé & Baddeley, 1982) but also the irrelevant sound effect (ISE; Jones & Macken, 1993) given that it can occur with nonspeech sounds. Investigators have noted theoretical implications of this task for understanding memory, attention, and perception (for a recent review see Neath, 2000 and commentaries by Baddeley, 2000 and Jones & Tremblay, 2000). Fundamentally, the ISE reflects an important limitation in humans' ability to determine which stimuli will enter into the mnemonic processing needed for short-term recall. To the extent that the ISE affects processes within short-term recall that are also used more widely in cognitive tasks, it is of broad significance. Yet, important questions remain.

A key question we will address is whether an individual's degree of susceptibility to the ISE is materials-specific or general. Jones and Macken (1993) found that either tones or speech could disrupt serial recall, provided that the auditory stimuli kept changing in perceptible ways (e.g., the use of multiple syllables or tone pitches as opposed to a single, repeating stimulus). According to their theoretical account (developed by Jones, 1993), the disruption occurs for the same reason for irrelevant speech sounds or tones: because the sounds corrupt the formation of a short-term episodic record.

We do not know of a theory asserting that there is no process in common between speech and tone interference. However, a theory of Neath (2000) asserts that speech interference occurs primarily because features from the speech are adopted into the phonological memory record of the printed items to be recalled. To account for irrelevant-tone effects, an additional assumption was that changes in irrelevant sounds can cause distraction (cf. Cowan, 1995).

In light of Neath’s (2000) approach, speech-based and tone-based ISEs theoretically could occur for largely different reasons (speech effects, because of feature adoption; tone effects, because of distraction). Different sets of individuals might be most susceptible to speech interference versus tone interference and the correlation between them would be low. In contrast, a strong correlation between speech-based and tone-based ISEs, indicating that some individuals are more susceptible than others to both kinds of interference, would help to constrain Neath's model, by suggesting that the same mechanism should be used to account for much of the variance in both types of ISE (as would be posited also by Jones, 1993). We address this question with a large sample of adults and children using regression analyses.

A second question we will address (in addition to the materials-specific versus general basis of disruption by irrelevant sounds) is about the role of attentional distraction. If distraction is a component of the ISE (as suggested by Cowan, 1995; see also Buchner, in press), then one would expect that individuals with better attention-related skills would be better able to avoid distraction. One way that attention-related skills have been examined in previous literature is through the measurement of working memory capacity, in which individuals engage in tasks that require some elements of additional processing while attempting to retain information for later recall (Daneman & Carpenter, 1980). Given that individuals with better working-memory abilities appear to be better at using attention to resist interference (e.g., Kane & Engle, 2003), one could similarly expect that individuals with better serial recall or working memory would demonstrate smaller ISEs.

In contrast to this second expectation, though, two studies have demonstrated that correlations between irrelevant-speech effects and serial recall are nonsignificant despite large individual differences in susceptibility to irrelevant speech (Ellermeier & Zimmer, 1997; Neath, Farley, & Surprenant, 2003). A third study (Beaman, in press) similarly indicates no significant correlation between irrelevant speech and the level of performance in an often used measure of working memory capacity known as operation span, which taxes attention by requiring that the participants carry out an arithmetic operation before each item to be retained for later recall (Turner & Engle, 1989).

The absence of these correlations in adults stands in contrast, however, to what has been found in a developmental study with children and adults (Elliott, 2002). Along with the well-established increase in memory span with age, a significant decrease in the magnitude of the ISE was obtained. The magnitude of this effect was especially striking for speech interference (i.e., the irrelevant-speech effect was particularly large in children compared to adults). The results of this study were interpreted with respect to the role of attention in the ISE. As children develop, according to the interpretation, they gain better control of their attentional abilities, and this leads to a reduction in the disruption shown by irrelevant sounds.

According to certain assumptions, one would also expect negative correlations between span and the magnitude of the ISE within an age group. The key such assumption is that the factors influencing span and the ISE are the same or are correlated with one another within an age group. The link to the developmental evidence depends on the additional assumptions that developmental changes in span (and in the magnitude of the ISE) depend on changes in the same factors that are important in distinguishing between individuals within an age group. Given the developmental evidence, accepting these assumptions leads to a theory in which individuals with higher spans should be less susceptible to the ISE than lower-span individuals. One can see, though, that there are several ways in which the assumptions could be wrong. For example, development includes not only an improvement in attention (Lane & Pearson, 1982; Zuckier & Hagan, 1978), but also an improvement in covert verbal rehearsal (Cowan & Kail, 1996; Flavell, Beach, & Chinsky, 1966; Ornstein & Naus, 1978). It is always possible that span and diminution of the ISE depend on attention and rehearsal in different ways or that these two factors play different roles in developmental versus individual-difference data. These are assumptions worth exploring and the absence of significant correlations in adult studies therefore warrants further investigation.

We re-examine these correlations with modifications of previously-used methods. We examined correlations between serial-recall performance and the effect of irrelevant tones, whereas previous studies have examined only correlations using speech interference. If a speech-specific process contributes a great deal to the effect of the ISE, as some have theorized (Salamé & Baddeley, 1982; Neath, 2000), and if this speech-specific effect is unrelated to attention, then this effect might obscure a smaller, attention-related effect that might correlate with span. With interfering tones, any obscuring effect of a speech-specific process would be eliminated so that the possibility of a correlation between an attention-related process and level of performance in serial recall could be examined more clearly.

Also, in the correlational studies using serial recall, the measures of performance were somewhat unconventional. In the study of Ellermeier and Zimmer (1997), no correlation between the ISE and memory capacity was obtained (N=72, r = 0.01). However, the measure of capacity was nonstandard. Given that recall in a quiet control condition was used to calculate the magnitude of the ISE, a separate, pink noise condition was used as the measure of memory capacity. In the study of Neath et al. (2003), the quiet control condition was used for the performance measure and a proportional measure of the irrelevant-speech effect was used. Specifically, the speech - quiet difference score was divided by the performance level in quiet and multiplied by 100 to produce a percentage effect. With N=100, the percentage effect of irrelevant speech was found not to correlate with the level of performance in quiet. However, in neither of these studies was the measure of performance a separate memory-span task. In our Experiment 1, it was.

EXPERIMENT 1: MEMORY SPAN AND IRRELEVANT SOUNDS

This large-scale experiment comprises six sub-experiments (1a – 1f) conducted originally for a different purpose. We had been trying to understand a finding that the disruptive effect of tones presented during the visual items to be recalled occurred only sporadically in procedures that also included a long retention interval between the stimuli and the recall cue. In the process of investigating that finding (which we still do not understand), we ran control experiments in which no retention interval was included. These showed ISEs with both speech and tones and were combined here to allow a large-scale correlational investigation. The questions under investigation were (1) whether the speech- and tone-based ISEs are related, and (2) whether there are at least modest correlations between ISEs and memory span.

Method

Participants

A total of 211 participants were originally assessed, but two males and two females reported hearing loss. One female reported taking pain medicine, and a computer problem excluded another female. In the final sample, N = 205 (157 female, 48 male). The participants were enrolled in Psychology classes at the University of Missouri and Louisiana State University. All reported normal or corrected-to-normal vision and hearing, and were native speakers of English.

Apparatus, Stimuli, and Procedure

Overview

Six sub-experiments (1a – 1f) all used a fixed list length of 8 items with immediate recall. The only differences among them were in how the tones were selected and how many tones were used on each trial (see below). None of the analyses indicated any differences in the disruption caused by the variations of the tone conditions. This is consistent with previous research by Tremblay and Jones (1998) showing that the “token set size” does not increase the amount of disruption when the number of tokens is larger than two.

Participants were seated in a quiet room or booth for an immediate visual-span assessment, a delayed visual-span assessment, and then the irrelevant-sounds task. In all three tasks, lists of digits from the set 1–9 (randomly selected without replacement for each list) were presented on a computer screen for immediate recall. The digits in a list were presented one at a time in the center of the screen, in 30-point font, at the rate of one per second, with no inter-stimulus interval. Responses were typed using the computer keypad as described below.

Immediate span task

Participants first carried out three practice trials with three-digit lists. These were followed by test lists that began with three digits per list. Four lists were presented at each list length, with length increasing until the point at which a participant failed to answer at least one list out of the four correctly. The range of list lengths was from three to nine items.

A fixation cross appeared in the center of the screen for 750 ms prior to the first digit appearing. After the set of digits was presented for a particular list, a row of white line segments appeared on the screen, cueing participants to type in the response. The number of lines matched the number of items presented in a given list. When the response was completed, the computer queried if a change was needed, because participants were not allowed to backspace during digit entry. If participants responded “yes” by pressing the “y” key, they were given a chance to change the responses. Participants responded “no” by pressing the “n” key. After the prompt to indicate yes or no (and possible digit entry), the computer continued on to the next trial. The presentation of the next list was initiated by the participant pressing the spacebar, and the next trial began immediately. Instructions for this task were to remember the numbers in the order in which they were presented, and to guess if unsure. Participants were told not to speak the numbers aloud or to whisper to themselves, but to "think about the numbers inside your head."

Delayed Visual Span Task

This task was identical to the immediate span task with one exception. After seeing the digits, participants waited through a retention interval before they were cued to respond. The retention interval was equal in duration to the duration for the visual list presentation. For example, if a participant was shown a 6-item list for recall, they then saw a black screen for 6 s before the recall cue (white lines) appeared. Instructions were to recall the numbers in order, as in the previous task. In addition, they were told that extra time would be given before they were asked to provide their answer and they were encouraged to think actively about the numbers during this time, "inside your head."

Serial Recall with Irrelevant Sounds

Each trial included an 8-digit list visually presented. In Sub-experiments 1a, 1b, and 1c, there were three auditory conditions (words, tones, and silence). Participants first completed 3 trials in a practice block, and then 9 test blocks of 3 trials each, for a total of 30 trials. In Sub-experiments 1d, 1e, and 1f, there were four auditory conditions (words, 4 tones, 8 tones, and silence), and there were 4 trials in a practice block followed by 9 test blocks of 4 trials each, for a total of 40 trials. In all experiments, the auditory conditions were randomly ordered within each block of trials.

Further details about the sub-experiments are as follows. In Sub-experiment 1a (N=30), University of Missouri students heard 8 tones out of a possible selection of 9 on each irrelevant-tone trial, selected randomly without replacement. In Sub-experiments 1b and 1c (total N=73), Louisiana State University students heard 4 of the tones, randomized and repeated twice during the visual presentation of 8 digits to be recalled. Sub-experiments 1d (at Missouri) and 1e and 1f (at Louisiana State) (total N=101), included both the 4-tone and the 8-tone conditions. In each of these experiments, the conditions with 4 tones and with 8 tones were not significantly different in the amount of disruption to serial recall performance. For correlational purposes, therefore, the average across these two tone conditions was used in these sub-experiments.

The irrelevant, spoken words were selected randomly without replacement from the following: red, blue, green, yellow, white, tall, big, short, and long. These words were digitally recorded from a male speaker, and ranged in duration from 210 – 500 ms. The sine-wave tones were 500 ms in duration, and were selected randomly from the following set of frequencies: 87, 174, 266, 348, 529, 696, 788, 880, and 972 Hz. These were the same words and tones used by Elliott (2002) and therefore afforded a comparison across studies.

All of the sounds in a trial were presented with a one-second onset-to-onset period, with onsets synchronized with the visual stimuli. The sounds were presented over audiological headphones and sound levels, measured with a Quest sound-level meter and earphone coupler, all fell within the range of 62–68 dB(A).

The instructions in this task were to concentrate on remembering the digits in order and to ignore any sounds that were heard during any portion of the task. Participants were asked not to say anything aloud during the task. Halfway through the task, participants were offered a break.

Results and Discussion

Means

Means for every sub-experiment and for the full sample are presented in Table 1. Two measures of span were calculated to assess the number of lists answered correctly at each list length. The cumulative span measure was created by finding the highest list length at which a participant answered all four lists correctly, and then adding 0.25 for each additional list above that list length (see Elliott, 2002). The second measure, referred to as maximum span, was calculated as the highest list length at which at least one list was answered correctly.

Table 1.

Means (and Standard Errors) in the Six Sub-Experiment Studies Within Experiment 1, Separately and Together

	Sub-Experiment
Measure	1a	1b	1c	1d	1e	1f	All (1a–1f)
N	31	30	43	31	30	40	205
Span Task
DelCum	7.4 (0.20)	6.5 (0.13)	6.6 (0.16)	6.5 (0.19)	6.7 (0.12)	6.7 (0.14)	6.7 (0.07)
DelMax	8.3 (0.19)	7.6 (0.18)	7.6 (0.16)	7.5 (0.22)	7.7 (0.15)	7.8 (0.15)	7.7 (0.07)
ImmCum	7.1 (0.19)	6.2 (0.17)	6.3 (0.13)	6.3 (0.18)	6.7 (0.16)	6.5 (0.16)	6.5 (0.07)
ImmMax	7.8 (0.20)	7.3 (0.20)	7.4 (0.18)	7.3 (0.20)	7.9 (0.19)	7.4 (0.17)	7.5 (0.08)
List-Recall Task
Silence	0.80 (0.03)b	0.64 (0.03)c	0.68 (0.02)c	0.64 (0.04)b	0.73 (0.03)c	0.73 (0.02)c	0.71 (0.01)c
Speech	0.73 (0.04)a	0.52 (0.03)a	0.57 (0.03)a	0.56 (0.04)a	0.59 (0.03)a	0.58 (0.03)a	0.59 (0.01)a
4 Tones		0.59 (0.03)b	0.64 (0.03)b	0.63 (0.03)b	0.66 (0.03)b	0.66 (0.02)b
8 Tones	0.78 (0.03)b			0.60 (0.03)b	0.67 (0.03)b	0.65 (0.02)b
All Tones	0.78 (0.03)	0.59 (0.03)	0.64 (0.03)	0.62 (0.03)	0.66 (0.03)	0.65 (0.02)	0.66 (0.01)b

Note. Span tests measures: DelCum = delayed test, cumulative score; DelMax = delayed test, maximum score; ImmCum = immediate test, cumulative score; ImmMax = immediate test, maximum score.

Within each of the six sub-experiments, list-recall condition scores marked with different superscript letters (a–c) are significantly different in Newman-Keuls post-hoc tests.

Correlations

Following the previous literature on individual differences in the ISE, difference scores were created to compare the magnitude of the irrelevant-speech effect and the irrelevant-tone effect. In each case, the sound condition performance level was subtracted from the level of performance in the silent control condition. These derived variables were approximately normally distributed.

These two indices were entered into a correlational analysis with the four measures of span, the silent control condition, the speech condition, and the tone condition (shown in Table 2). The generally high correlations between different span scores demonstrates that they were reliable, as the table shows. The generally low correlations between the ISE effects and the immediate spans is in keeping with past results (Ellermeier & Zimmer, 1997; Neath et al., 2003).

Table 2.

Correlations Between Span Measures and ISE Measures in Experiment 1

	1	2	3	4	5	6	7	8	9
DelCum (1)	0.84	0.97	0.81	0.66	0.26	0.10	0.89	0.76	0.79
DelMax (2)	0.81*	0.82	0.72	0.62	0.40	0.33	0.83	0.61	0.69
ImmCum (3)	0.68*	0.60*	0.84	1.00	0.12	0.00	0.81	0.72	0.75
ImmMax (4)	0.55*	0.51*	0.85*	0.83	0.30	0.00	0.70	0.63	0.59
ToneEffect (5)	0.10	0.16*	0.05	0.12	0.20	1.00	0.81	−0.27	−0.76
SpeechEffect (6)	0.05	0.17*	0.00	0.00	0.59*	0.33	0.47	−0.89	−0.25
Silence (7)	0.72*	0.67*	0.66*	0.57*	0.32*	0.24*	0.81	0.93	0.98
Speech (8)	0.63*	0.50*	0.60*	0.52*	−0.11	−0.46*	0.75*	0.82	0.99
Tones (9)	0.66*	0.57*	0.63*	0.49*	−0.31*	−0.13	0.80*	0.82*	0.84

Note. DelCum = delayed test, cumulative score; DelMax = delayed test, maximum score; ImmCum = immediate test, cumulative score; ImmMax = immediate test, maximum score. Scores on the diagonal represent Cronbach’s Alpha reliability coefficients. Scores above the diagonal represent the correlations after correction for attenuation. These were not evaluated for statistical significance.

^*p < .05.

To evaluate the reliability of all of the variables, separate reliability analyses were conducted following the methodology of Ellermeier and Zimmer (1997). On the basis of the raw measures, which all had reliabilities of .81 and above, proportion correct values from adjacent speech and silence trials were used to create speech effect difference scores for each participant individually, resulting in nine difference scores per participant. These nine scores were used to calculate Cronbach’s alpha with a resulting reliability estimate, α = 0.33. In comparison, Ellermeier and Zimmer had a higher number of trials per participant than the current study, which resulted in twenty scores for each of their 72 participants and a higher reliability estimate, α = 0.55.

Each participant also contributed 9 trials in the tones condition (for the experiments containing more than one type of tone condition, the average of the tone trials was used). The same method was used to calculate the reliability of the tone effect difference score, with α = 0.20. Given the low reliability of the difference scores, below we examine our hypotheses with regressions based on raw scores, which were much more reliable as shown in Table 2.

Unlike the past studies, there were significant, albeit modest, correlations; in particular, between the maximum measure of delayed span and both the speech effect and the tone effect (r = 0.17 and 0.16, respectively, with N = 205; with this number of participants, an r = 0.14 would be statistically significant). The direction of correlation was an unanticipated one in which individuals with a higher span had larger ISEs. However, these correlations became nonsignificant when the four individuals with the lowest delayed maximal spans (5 items) were removed from the sample. One possible interpretation of the effect is that the levels of serial-recall performance in these four individuals were so low in the silent condition (M = .40, SD = .09, as opposed to a much higher mean for the entire sample, M = .71, SD = .17) that they reflected only a basic form of memory that is resistant to disruption (e.g., the focus of attention; see Cowan, 2001).

We analyzed our ability to detect correlations in this sample, using N=201 after the four individuals with the lowest delayed maximal spans were removed. Without those participants, the correlation between the maximum measure of delayed span and the speech effect was r = .10, and the resulting power was low (power = .29). To obtain a more acceptable level of power, such as the standard .80, a minimum correlation, r = .20, would be required.

Regressions

The high correlation between the speech effect and tone effect (r = .59; see Table 2) is potentially of central interest but cannot be cleanly interpreted, given that both effects were calculated using the same silent control condition. To overcome this problem, we entered the ISE variables and silent control condition into stepwise regression analyses, shown in Table 3.

Table 3.

Regression Analyses in Experiments 1 and 2 Using Difference (Effect) Scores

Age group	Dependent Measure	Regression Step 1	Regression Step 2
Experiment 1 (N = 205)
Adults	Tone effect	Silence	Speech effect
		R2 = .11, p < 0.01	ΔR2 = .28, p < 0.01
Adults	Speech effect	Silence	Tone Effect
		R2 = .06, p < 0.01	ΔR2 = .29, p < 0.01
Experiment 2 (N = 64 adults, 63 children)
Adults	Tone effect	Silence	Speech Effect
		R2 = .17, p < 0.01	ΔR2 = .11, p < 0.01
Adults	Speech effect	Silence	Tone Effect
		R2 = .12, p < 0.01	ΔR2 = .12, p < 0.01
Children	Tone effect	Silence	Speech Effect
		R2 = .25, p < 0.001	ΔR2 = .14, p < 0.001
Children	Speech effect	Silence	Tone Effect
		R2 = .10, p < 0.05	ΔR2 = .16, p < 0.001

Note: The data for Experiment 2 included both span and span − 1 list lengths averaged together.

In the first such analysis, the dependent variable was the tone effect. As the table shows, a significant amount of variance was shared by the tone effect and the silent control condition, which was entered first into the regression equation. What is important in the analysis is that the speech effect, entered after the silent condition, still accounted for a very substantial and significant additional amount of variance in the tone effect, Δ R² = 0.28. Similarly, when the speech effect was the dependent variable and the silent control condition was removed first, the tone effect still accounted for a substantial and significant additional amount of variance, Δ R² = 0.29.

These regressions indicate that susceptibilities to the speech and tone effects are closely related, for reasons having nothing to do with the use of a common control condition to calculate the effects. Almost identical results were obtained when the four “outlier” participants with a delayed maximal span of only 5 items (see above) were omitted.

Figure 1 further illustrates, quite explicitly, that the susceptibility to irrelevant sounds is unrelated to memory span. It is a scatter plot of the irrelevant-speech effect as a function of the irrelevant-tones effect, with different symbols to indicate the immediate span (maximal measure) for each individual. It is clear that there is a strong relation between irrelevant-speech and irrelevant-tone effects and that, for each span level, there is a range of susceptibility levels across individuals.

Figure 1.

Scatter plot of the irrelevant-tones effect as a function of the irrelevant-speech effect in Experiment 1 (in proportion difference scores). Separate symbols (graph parameter) represent individuals with different immediate maximal span (ImmMax) scores.

Inasmuch as the reliability estimates for the speech and tone effects were much lower than the estimates for the raw scores, additional regression analyses were conducted using the raw scores from the speech and tone conditions in the place of the difference scores. The results are shown in Table 4. The pattern is very similar to the analyses using the difference scores. Variability in the irrelevant-tone condition was predicted first by the silent control condition; and the irrelevant-speech condition, entered after the silent condition, accounted for significant additional variance. Similarly, the irrelevant-tone condition predicted significant variance in the irrelevant-speech condition even after the relation to the silent condition was taken out.

Table 4.

Regression Analyses in Experiments 1 and 2 Using Condition Scores

Age group	Dependent Measure	Regression Step 1	Regression Step 2
Experiment 1 (N = 205)
Adults	Tones	Silence	Speech
		R2 = .61, p < 0.01	ΔR2 = .11, p < 0.01
Adults	Speech	Silence	Tones
		R2 = .56, p < 0.01	ΔR2 = .13, p < 0.01
Experiment 2 (N = 64 adults, 63 children)
Adults	Tones	Silence	Speech
		R2 = .22, p < 0.01	ΔR2 = .11, p < 0.01
Adults	Speech	Silence	Tones
		R2 = .19, p < 0.01	ΔR2 = .11, p < 0.01
Children	Tones	Silence	Speech
		R2 = .36, p < 0.001	ΔR2 = .12, p < 0.001
Children	Speech	Silence	Tones
		R2 = .26, p < 0.001	ΔR2 = .13, p < 0.001

Note: The data for Experiment 2 included both span and span − 1 list lengths averaged together.

We also carried out regressions on each type of span measure using serial recall performance in the presence of silence, speech, or tones as the predictors (see Table 5). There were some significant regression effects of the irrelevant-sound conditions after the silent-condition variation was taken out, but they accounted for only very small amounts of variance in the span measures.

Table 5.

Regression Analyses for Experiment 1 Using Span Scores as the Dependent Measure

Age group	Dependent Measure	Regression Step 1	Regression Step 2	Regression Step 3
Adults	Dspan_C	Silence	Speech	Tones
		R2 = .52, p < .001	ΔR2 = .02, p < .01	ΔR2 = .01, p < .05
			Tones	Speech
			ΔR2 = .02, p < .01	ΔR2 = .00, ns
	Dspan_M	Silence	Speech	Tones
		R2 = .44, p < 0.001	ΔR2 = .00, ns	ΔR2 = .01, ns
			Tones	Speech
			ΔR2 = .00, ns	ΔR2 = .00, ns
	Ispan_C	Silence	Speech	Tones
		R2 = .44, p < 0.001	ΔR2 = .03, p < .01	ΔR2 = .01, p < .05
			Tones	Speech
			ΔR2 = .03, p < .001	ΔR2 = .00, ns
	Ispan_M	Silence	Speech	Tones
		R2 = .32, p < 0.001	ΔR2 = .02, p < .05	ΔR2 = .00, ns
			Tones	Speech
			ΔR2 = .01, ns	ΔR2 = .01, p < .05

One possible reason for this absence of correlation in the adult samples in which it has been evaluated is that avoidance of the ISE requires attention (Cowan, 1995; Buchner, in press) and therefore would correlate more highly with WM procedures in which processing and storage are combined (e.g., reading span: Daneman & Carpenter, 1980; operation span: Turner & Engle, 1989), given that these tasks seem to depend on how well attention can be controlled (e.g., Kane & Engle, 2003). Although this hypothesis was not confirmed in one recent study with adult participants (Beaman, in press), in Experiment 2 we re-examine this issue with samples of second-grade children and adults, using both irrelevant speech and irrelevant tones as in Experiment 1. In this experiment, the list length depended on the individual’s span, so that the level of difficulty could be kept comparable across age groups.

One reason to examine children, as well as adults, is that children do not engage in the systematic, effective covert verbal rehearsal that adults do (e.g., Cowan & Kail, 1996; Flavell, Beach, & Chinsky, 1966; Ornstein & Naus, 1978). The inclusion of children in Experiment 2, with rapidly changing abilities of not only rehearsal, but also attention (Lane & Pearson, 1982; Zuckier & Hagan, 1978), can help us to address the issue of correlations between measures of span and the ISE by including more individual difference variability. Furthermore, we already know that there is an increase in the ISE with speech as opposed to tone distractors in children, more than in adults (Elliott, 2002), but the relation between irrelevant-speech and irrelevant-tone effects has not been examined in this age group.

EXPERIMENT 2: VARIOUS SPAN MEASURES AND THE ISE

Method

Participants

Sixty-four adults from the University of Missouri (43 females and 21 males) and sixty-three children in the 2^nd grade from the Columbia Public School system (34 females and 29 males; ages ranged from 7 years, 1 month to 9 years, 2 months; M = 8 years, 2 months, SD = 4 months) were included in these analyses. Adults were given credit in their Psychology classes and the children were given $5 and a book for their participation. All participants reported normal or corrected-to-normal vision, normal hearing, and were native speakers of English.

Design and Procedure

Participants were asked to complete a visual span task, an auditory span task, a reading span task, an operation span task, and an irrelevant-sounds task that included both speech and tones. The visual span task was identical to the immediate span task used in Experiment 1, and the auditory span task was only different in that the digits to be recalled were presented over headphones.

The two complex span tasks were slight variations of the reading and operation span tasks used by Towse and his colleagues (Towse, Hitch, & Hutton, 1998). In the reading span task, sentences were presented that were missing the final word. Participants were to read aloud each sentence, complete it with a single word, and recall these generated words when a cue was presented (e.g., “The clown had a big smiling ___;” the expected answer was “face.”) Each trial included 2 to 7 sentences (randomly varying from 5 to 13 words long, including the completion word), with three trials at each set size. If a participant repeated all of the completions for at least one trial correctly, the computer advanced to the next set size.

The stimuli for the operation span task were arithmetic problems presented in the form of A [op] B = ? (short form) or A [op] B [op] C [op] D, where [op] was “+” or “−“, A was a number from 1 to 10, and B through D were each 0 or 1. The sum always fell in the range of 3 to 9, e.g., “4 + 1 − 1 + 1 = ?” The participant read aloud the arithmetic problem, solved the problem, and tried to retain the sum. These problems were presented in set sizes ranging from 2–7 problems. The participant verbally recalled the sums in order when a cue to recall was presented. Two lists had to be recalled correctly within a set size for the computer to advance to the next set size. If a participant recalled two consecutive lists correctly, the program went on to the next set size without presenting the third problem for that set size.

To control for the level of difficulty across the age groups, lists in the irrelevant-sounds task were presented at each individual’s measure of maximum visual span and at span minus one (span – 1). Further details on the method of the irrelevant-sounds task were reported in Elliott (2002).

Results and Discussion

Means

The means and standard errors for the span measures and the irrelevant-sound task are shown in Table 6. As in Experiment 1, difference scores were created to assess the speech and tone effects.

Table 6.

Means (and Standard Errors) in Experiment 2

	Age Group
Measure	Children			Adults
	N	Mean	SEM	N	Mean	SEM
Span Task
VspanMax	63	5.0	0.10	64	7.3	0.15
VspanCum	63	4.3	0.09	64	6.6	0.14
AspanMax	59	5.6	0.11	64	7.5	0.13
AspanCum	59	4.9	0.09	64	6.7	0.12
RspanMax	60	2.5	0.09	64	4.6	0.12
RspanCum	60	2.0	0.06	64	3.8	0.09
OspanMax	55	3.6	0.13	64	6.1	0.11
OspanCum	55	3.1	0.11	64	5.5	0.12
List-Recall Task
Span Length Lists
Silence	63	0.77a	0.03	64	0.79a	0.02
Speech	63	0.41b	0.03	64	0.70b	0.02
Tones	63	0.56c	0.03	64	0.71b	0.02
Span – 1 Length Lists
Silence	63	0.89a	0.02	64	0.87a	0.02
Speech	63	0.64b	0.03	64	0.82b	0.02
Tones	63	0.80c	0.03	64	0.90a	0.02
Average of Both List Lengths
Silence	63	0.83a	0.02	64	0.83a	0.01
Speech	63	0.52b	0.03	64	0.76b	0.02
Tones	63	0.68c	0.02	64	0.80a	0.02

Note. Span tests measures: Vspan = visual digit span, Aspan = auditory digit span, Rspan = reading span, Ospan = Operation span, "Cum" = cumulative span measure, "Max" = maximum span measure.

Within each age group, list-recall condition scores marked with different superscript letters (a–c) are significantly different in Newman-Keuls post-hoc tests.

Correlations

The correlations between the measures are shown in Table 7 for the adults and Table 8 for the children. In the adults, there was a significant correlation between the irrelevant-speech effect and maximal visual span, in the unanticipated direction (i.e., higher spans corresponding to a larger irrelevant-speech effect): r = 0.38, p = 0.002. After correcting for attenuation of the measures this was a very large correlation, r = 0.95. We did a power analysis for this adult correlation, and with N=64 (and an estimate of Rho= 0.38), the power to detect this correlation was 0.88. With this sample, r = .34 could be detected with .80 power. (In children, no such correlation was significant, r = −.03. In our sample of 63 children, to detect a correlation with a minimum value of .35, the power was = .81.)

Table 7.

Correlations Between Measures in Adult Participants in Experiment 2.

	1	2	3	4	5	6	7	8	9	10	11	12	13
VspanMax (1)	0.85	1.00	0.74	0.75	0.34	0.57	0.49	0.60	−0.36	−0.77	−0.25	0.95	−0.11
VspanCum (2)	0.91*	0.86	0.72	0.77	0.32	0.54	0.45	0.57	−0.25	−0.51	0.00	0.62	−0.37
AspanMax (3)	0.62*	0.60*	0.82	1.00	0.48	0.66	0.52	0.61	0.05	−0.19	0.06	0.41	−0.05
AspanCum (4)	0.61*	0.63*	0.86*	0.78	0.53	0.74	0.48	0.65	0.23	−0.16	0.18	0.65	0.06
RspanMax (5)	0.26*	0.25*	0.36*	0.39*	0.69	1.00	0.61	0.53	0.35	0.25	0.38	−0.03	−0.15
RspanCum (6)	0.42*	0.40*	0.48*	0.52*	0.82*	0.64	0.72	0.77	0.22	0.07	0.09	0.17	0.18
OspanMax (7)	0.40*	0.37*	0.42*	0.38*	0.45*	0.51*	0.79	1.00	0.10	−0.25	−0.08	0.57	0.30
OspanCum (8)	0.50*	0.48*	0.50*	0.52*	0.40*	0.56*	0.86*	0.82	−0.06	−0.35	−0.10	0.56	0.08
Silence (9)	−0.23	−0.16	0.03	0.14	0.20	0.12	0.06	−0.04	0.47	0.84	0.99	1.00	1.00
Speech (10)	−0.54*	−0.36*	−0.13	−0.11	0.16	0.04	−0.17	−0.24	0.44*	0.58	0.86	−1.00	−0.38
Tones (11)	−0.16	0.00	0.04	0.11	0.22	0.05	−0.05	−0.06	0.47*	0.50*	0.48	−0.50	−1.00
SpeechEffect (12)	0.38*	0.25	0.16	0.25	−0.01	0.06	0.22	0.22	0.35*	−0.69*	−0.15	0.19	1.00
ToneEffect (13)	−0.04	−0.14	−0.02	0.02	−0.05	0.06	0.11	0.03	0.41*	−0.12	−0.61*	0.46*	0.17

Note. Vspan = visual digit span, Aspan = auditory digit span, Rspan = reading span, Ospan = Operation span, "Max" = maximum span measure, "Cum" = cumulative span measure. N = 64. Data points were averaged across both span and span − 1 length lists. Scores on the diagonal represent Cronbach’s Alpha reliability coefficients, with the exception of the scores for the Ospan and Vspan tasks. For those tasks, Spearman-Brown split-half reliability estimates were calculated. Scores above the diagonal represent the correlations after correction for attenuation. These were not evaluated for statistical significance.

^*p < .05

Table 8.

Correlations Between Measures in Child Participants in Experiment 2

	1	2	3	4	5	6	7	8	9	10	11	12	13
VspanMax (1)	0.66	1.00	0.50	0.59	0.46	0.59	0.36	0.24	−0.74	−0.52	−0.86	−0.06
VspanCum (2)	0.77*	0.83	0.52	0.61	0.20	0.51	0.35	0.20	−0.40	−0.36	−0.54	0.11
AspanMax (3)	0.34*	0.39*	0.70	1.00	0.20	0.36	0.23	0.31	0.23	0.37	0.08	−0.32
AspanCum (4)	0.39*	0.45*	0.79*	0.65	0.36	0.50	0.27	0.35	0.10	0.10	−0.12	−0.06
RspanMax (5)	0.22	0.11	0.10	0.17	0.35	1.00	0.68	0.40	−0.18	0.22	0.22	−0.53
RspanCum (6)	0.30*	0.29*	0.19	0.25	0.69*	0.39	0.29	0.31	−0.27	0.04	−0.20	−0.34
OspanMax (7)	0.26	0.28*	0.16	0.19	0.16	0.16	0.78	1.00	0.01	−0.37	−0.13	0.56
OspanCum (8)	0.17	0.16	0.23	0.25	0.21	0.18	0.89*	0.77	0.08	−0.23	−0.07	0.43
Silence (9)	−0.50*	−0.30*	0.16	0.07	−0.09	−0.14	0.01	0.06	0.69	0.70	1.00	0.55
Speech (10)	−0.35*	−0.27	0.26	0.07	0.11	0.02	−0.27	−0.17	0.48*	0.69	0.88	−1.00
Tones (11)	−0.44*	−0.31*	0.04	−0.06	0.08	−0.08	−0.07	−0.04	0.56*	0.61*	0.39	−0.53
SpeechEffect (12)	−0.03	0.06	−0.16	−0.03	−0.19	−0.13	0.30*	0.23	0.28*	−0.71*	−0.21	0.37
ToneEffect (13)	−0.09	0.00	0.13	0.14	−0.18	−0.07	0.09	0.11	0.51*	−0.11	−0.43*	0.53*	0.00

Note. Vspan = visual digit span, Aspan = auditory digit span, Rspan = reading span, Ospan = Operation span, "Max" = maximum span measure, "Cum" = cumulative span measure. Based on N=52 participants with data for all of the measures. Data points were averaged across both span and span − 1 length lists. Scores on the diagonal represent Cronbach’s Alpha reliability coefficients, with the exception of the scores for the Ospan and Vspan tasks. For those tasks, Spearman-Brown split-half reliability estimates were calculated. Scores above the diagonal represent the correlations after correction for attenuation. These were not evaluated for statistical significance.

^*p < .05

One possible explanation for this correlation between the irrelevant-speech effect and maximal visual span in adults is that irrelevant speech may interfere with rehearsal, as many theorists agree (Jones, 1993; Beaman & Jones, 1997,1998). The explanation would be as follows. If one examines the number of items recalled in the silent control condition, it increases markedly with span size. For individuals with spans (and list lengths in the ISE procedure) of 5, 6, 7, 8, and 9, respectively (n = 3, 16, 17, 13, and 15, respectively), the mean numbers of items correct were 4.44, 5.04, 5.31, 6.12, and 7.15. In contrast, in the irrelevant-speech condition, the numbers of items correct in these subgroups were much more nearly identical: 4.56, 4.55, 5.15, 5.12, and 5.39. Thus, irrelevant speech greatly diminished the ability of individuals with a longer span to correctly recall the longer lists that they received in this experiment. Given that the ISE was calculated as a difference between silent and speech conditions, this would have resulted in larger ISEs for more capable participants.

In the children, there was a finding of a significant correlation between the maximum measure of operation span (OspanMax) and the irrelevant-speech effect difference score, r = .30, as shown in Table 8. When this correlation was adjusted for attenuation, it increased to r = .56. This may suggest, as discussed above, that those children with more sophisticated rehearsal processes were more susceptible to the ISE. However, no other correlations between measures of simple or complex span and the Speech Effect were significant in the children so it is difficult to clearly interpret the correlation with OspanMax. (As mentioned above, for the power to detect a significant correlation to reach the acceptable level of .80 the value of r needed to be at least .35). Regardless of the exact mechanism underlying this correlation, the most important point is that it was not in the expected direction. It was the higher-span individuals who had larger irrelevant-speech effects, and that requires a revised theoretical analysis, which we will offer in the General Discussion section.

Regressions

We conducted the same regression analyses on the data separately for the adults and the second graders. For the adults, the results were comparable to those obtained in Experiment 1. In accounting for the tone effect, after variation due to the silent control condition was removed, there was considerable residual variance due to the speech effect and, conversely, in accounting for the speech effect, after the variation due to the silent control condition was removed, there was considerable residual variance due to the tone effect. This same pattern was also seen in the children, indicating that these effects have shared variance in both the children and the adults (see Table 3). Analyses in which the condition means were used instead of the difference scores also produced the same significant effects as in Experiment 1, for both adults and children (Table 4)

As with Experiment 1, another series of regression analyses were conducted using the measures of memory span (simple and complex measures in this experiment) as the dependent variables to determine if the paucity of significant relationships between our span measures and the effect scores was the result of the unreliability of the effect scores. We found that this was not the case, as the analyses using raw measures roughly paralleled our findings with the effect scores. Using the average of the two list lengths, the only regression analyses to reach significance in the adults were when the immediate visual span measures were used (referred to as VspanMax and VspanCum to differentiate these from the auditory span measures). When VspanCum was used as the dependent measure, after silence was taken out in the first step of the analyses, for the speech condition ΔR² = 0.10, p < 0.01, and the variance explained by the tone condition was not significant. When done in the reverse order the tone condition did not explain significant variance but the speech condition resulted in ΔR² = 0.14, p < 0.01. When VspanMax was used as the dependent measure the tone condition once more did not explain a significant amount of variance in either order. The speech condition accounted for ΔR² = 0.24, p < 0.01, when it came before the tone condition, and ΔR² = 0.25, p < 0.01 when following the tone condition. Notice from the raw correlation (in Table 7) that individuals with higher spans actually had lower performance in the irrelevant-speech condition.

The same analyses were conducted with the children’s data. In line with the correlational analyses, the only significant finding in the regressions was for OspanMax, the maximum operation span. The silent control condition accounted for no variance in OspanMax and the irrelevant-speech condition, entered subsequently, accounted for ΔR² = 0.10, p < .05. Again, as in the visual spans in adults, notice from the raw correlation (in Table 8) that children with higher OspanMax scores actually had lower performance levels in the irrelevant-speech condition.

It appears that the only way to account for this pattern of findings, which includes larger ISEs for higher-span individuals even after the silent-condition variance is removed, is by an appeal to individual differences in strategies or processing styles. For example, it may be that using rehearsal is a good strategy for span performance but a bad strategy for remembering a printed list in the presence of speech interference; perhaps, in the presence of speech interference, rehearsal is worse than some alternative strategy such as simply trying to focus attention on the items to be remembered (cf. Cowan, 2001). If individuals tend not to be aware of the need to switch strategies in the presence of irrelevant speech, that could result in poorer performance in the irrelevant-speech condition in individuals with higher spans. Yet, it is not clear why these correlations were not found more consistently.

General Discussion

This study is the first to use a large sample and regression to investigate individual differences in the disruption of serial recall by irrelevant sounds. We now have evidence that the degree of susceptibility to the ISE is a reliable individual trait that has a substantial materials-general component, as opposed to materials-specific. Results of these experiments indicate that there is a strong relation between irrelevant-tone and irrelevant-speech effects, independent of span level.

For the most part, we confirmed the previous (Beaman, in press; Ellermeier & Zimmer, 1997; Neath et al., 2003) indications that this susceptibility to the ISE is unrelated to memory span. However, we also demonstrated the existence of various levels of susceptibility to the ISE, consistent across speech and tone effects, for participants at each level of span (see Figure 1). Thus, clearly, the absence of a correlation between the ISE and span cannot be attributed to unreliability of the ISE.

In several instances, we found positive correlations between the magnitude of the ISE and different types of memory span. In Experiment 1, it was the delayed maximum span that showed this effect, for both irrelevant-speech and irrelevant-tone effects. In Experiment 2, it was the visual maximum span for adults and the operation span for children that showed it. These correlations all run contrary to the a priori expectation of a smaller ISE in more capable individuals.

In a post-hoc manner, these positive correlations can be explained in several ways. First, if high-span individuals excel because they carry out mnemonic processing such as rehearsal more efficiently than lower-span individuals, and if this mnemonic processing is interrupted by irrelevant sounds (e.g., Jones, 1993; Beaman & Jones, 1997, 1998), then larger ISEs would be expected in the individuals who did more of the mnemonic processing in the first place.

Second, though, it may be that the correlation between high spans and the ISE is a statistical artifact in that spans are related to the silent condition used to calculate the ISE. However, this cannot be the sole explanation, given the regressions showing that span variance was picked up by the irrelevant-speech and irrelevant-tone conditions even after the variance from the silent control condition was removed. To explain these regression results, an account like the one involving mnemonic processing and its interruption by the ISE appears to be warranted.

These results provide one possible explanation for the puzzling absence of negative correlations between various sorts of working-memory span and the ISE (Beaman, in press; Ellermeier & Zimmer, 1997; Neath, Farley, & Surprenant, 2003). A mechanism that, by itself, leads toward such negative correlations theoretically may still exist, but it may be balanced out (or sometimes outweighed) by the factors leading toward a positive correlation, discussed above.

The second experiment of this study supported the findings of the first, and also included a developmental comparison. Given the dramatic decrease in the magnitude of the ISEs wisth development in childhood (Elliott, 2002), along with the dramatic increase in the magnitude of span itself, it is still surprising that negative correlations between span and the ISE have not been observed within an age group. This is surprising because one might have imagined that the same mechanisms that account for development of span also account for the developmental decrease in the magnitude of the ISE. Such factors could include the developmental improvement in covert verbal rehearsal ability (e.g., Cowan & Kail, 1996; Flavell et al., 1966; Ornstein & Naus, 1978) and the developmental improvement in the control of attention (e.g., Lane & Pearson, 1982; Zuckier & Hagan, 1978). It seemed natural to suppose that the same factors would operate within an age group to cause individuals with higher spans to have smaller ISEs.

An alternative possibility reinforced by the evidence is that there are rather different influences on span and the magnitude of the ISE. For example, whereas a more sophisticated strategy of covert verbal rehearsal undoubtedly is helpful to span in a silent setting, in the presence of irrelevant speech there may be both advantages and drawbacks to such a mnemonic strategy. It is also possible that the ability to control attention is more important for overcoming the ISE than for span in a silent setting.

There are many possible ways to explain the interesting discrepancy between the developmental results (Elliott, 2002 and the present study) and individual-difference results of the present study. We present just one scenario as an illustration. Within an individual, rehearsal may be a good strategy for span tasks but a poor strategy to overcome ISEs (at least in the case of speech interference). Across age groups, however, the older participants, who are better able to rehearse, also are the individuals with more advanced attentional capabilities; and that could be the predominant effect leading to the developmental decrease in the ISE.

Let us consider the contribution of attention in somewhat more detail, as its role in processing may be complex. Previous evidence from the cross-modal Stroop paradigm, in which participants are asked to name a colored square as quickly as possible while ignoring spoken, incongruent color words (Cowan & Barron, 1987; Elliott, Cowan, & Valle-Inclan, 1998), supports the conjecture that attention is helpful to overcome at least some sorts of irrelevant-sound effects. When examined developmentally, the magnitude of the disruption in the cross-modal Stroop effect was shown to decrease with age (Hanauer & Brooks, 2003), which the authors interpreted in terms of developmental improvements in selective attention. In contrast, there is little evidence that the control of attention is a key factor in verbal span tasks, though it may be more important for tasks in which rehearsal is prevented (Cowan, 2001).

If the control of attention is important for overcoming the ISE, then one would have expected it to correlate with performance in WM tasks in which the control of attention is important. One such WM task is operation span. There is a demonstration that the ability to overcome auditory distraction is related to operation span. Conway, Cowan, and Bunting (2001) examined individuals in the upper and lower quartiles of a larger sample of participants who completed an operation span task. Then, using a dichotic listening procedure, high and low WM span participants shadowed (or repeated immediately after hearing) words presented in one ear, while ignoring words presented to the other ear. After several minutes of shadowing, the participant’s name was presented in the unattended channel. The results were striking: only 20% of high span participants reported hearing their name in the unattended channel, in contrast to 65% of low span participants. Thus, in adults, WM capacity has been shown to be an important component of avoiding distraction in a dichotic listening task. The puzzle is then the fact that, in our Experiment 2, operation span was not negatively related to the magnitude of the ISE.

The solution to this puzzle that we have hinted at above is that another, contaminating factor heavily influences the magnitude of the ISE. Individuals who have good control of attention may also tend to be those who adopt a covert verbal rehearsal strategy, but that strategy may be harmful when irrelevant speech is present. The presence of some positive correlations between the magnitude of the ISE and some span measures in the present study indicates why it would be difficult to observe a theoretically-expected negative correlation, even if the hypothetical mechanism underlying such a correlation (such as the ability to ignore irrelevant sounds) does, in fact, exist.

In conclusion, the present research contributes to the literature on the ISE by (1) presenting a confirmation of previous findings regarding the absence of a general relationship between memory span and the ISE and (2) by demonstrating a significant amount of shared variance between speech and tone effects in adults and in children. The latter findings suggest a role for a common mechanism of disruption, regardless of the nature of the irrelevant sounds. The relations between WM spans and the ISE that were sporadically obtained were in the unexpected direction (larger ISEs for more capable participants), suggesting that a strategy often used to good effect in span tasks, such as covert verbal rehearsal, may actually be counterproductive in the presence of irrelevant sounds (at least irrelevant speech). The exact nature of the mechanisms involved, and the exact roles of attention, rehearsal, and episodic encoding in the ISE, are key questions for future research.