The impact of road traffic noise on cognitive performance in attention-based tasks depends on noise level even within moderate-level ranges

Abstract

Little empirical evidence is available regarding the effects of road traffic noise on cognitive performance in adults, although traffic noise can be heard at many offices and home office workplaces. Our study tested the impact of road traffic noise at different levels (50 dB(A), 60 dB(A), 70 dB(A)) on performance in three tasks that differed with respect to their dependency on attentional and storage functions, as follows: The Stroop task, in which performance relied predominantly on attentional functions (e.g., inhibition of automated responses; Experiment 1: n = 24); a non-automated multistage mental arithmetic task calling for both attentional and storage functions (Exp. 2: n = 18); and verbal serial recall, which placed a burden predominantly on storage functions (Experiment 3: n = 18). Better performance was observed during moderate road traffic noise at 50 dB(A) compared to loud traffic noise at 70 dB(A) in attention-based tasks (Experiments 1-2). This contrasted with the effects of irrelevant speech (60 dB(A)), which was included in the experiments as a well-explored and common noise source in office settings. A disturbance impact of background speech was only given in the two tasks that called for storage functions (Experiments 2-3). In addition to the performance data, subjective annoyance ratings were collected. Consistent with the level effect of road traffic noise found in the performance data, a moderate road traffic noise at 50 dB(A) was perceived as significantly less annoying than a loud road traffic noise at 70 dB(A), which was found, however, independently of the task at hand. Furthermore, the background sound condition with the highest detrimental performance effect in a task was also rated as most annoying in this task, i.e., traffic noise at 70 dB(A) in the Stroop task, and background speech in the mental arithmetic and serial recall tasks.

Introduction

The most common source of environmental noise is road traffic, which steadily increased over the last years. Road traffic noise can be overheard in many offices and home office workplaces, in particular if windows are open or tilted. While a multitude of studies exist on the acute and chronic effects of traffic noise on children's cognitive performance (cp.[1,2] for reviews), little evidence is available of its effects on adults (e.g.[3,4,5,6]). In fact, it is still an open question whether the level of road traffic noise plays a significant role in cognitive noise effects in adults, as no corresponding empirical study is available. However, level is the most often discussed physical dimension of traffic noise for several reasons, including the level orientation of legal traffic noise regulations and noise abatement measures, the link of high traffic noise levels and health threats (e.g., hypotension, coronary heart disease, and stroke cp.,[7] for a meta-analysis, and[8] for an overview), and the decisive role of noise level on perceived annoyance (e.g.,[9,10,11]). The present study aimed to contribute systematic empirical evidence of the effects of road traffic noise on cognitive performance in adults. As it is widely accepted in cognitive noise research that noise effects depend on both task and sound characteristics (cp., e.g.[12,13] for comprehensive reviews), we focused on different levels of road traffic noise (50-70 dB(A)) and on performance in three exemplarily selected tasks that varied regarding the burden placed on attentional and storage functions.

Although not testing specifically road traffic noise, some early studies on cognitive noise effects have varied the level of continuous broadband noise and found corresponding effects on performance. Specifically, loud noise was found to impair cognitive performance to a greater extent than was moderate noise (e.g.[14,15,16,17,18,19]). In these studies, however, the levels of moderate noise conditions were about 70-80 dB(A), whereas loud noise conditions were about 85-100 dB(A). These higher levels are generally nonexistent in urban road traffic noise, and are no longer tested in noise studies due to ethical reasons, namely the corresponding risk to hearing. Nevertheless, the applied tasks are informative concerning cognitive processes that are sensitive to such high noise levels, and may thus also apply to more moderate-level ranges. The least common denominator is most probably the dependency of task performance on attention functions. Attention is the cognitive ability to focus on a certain task or certain aspects of it (sustained attention), to split processing resources between tasks or task aspects (divided attention), and to inhibit automated responses. Congruently, in the studies mentioned above there were applied a vigilance test,[14] a complex choice reaction task,[15] nonautomated mental arithmetics,[16] and the Stroop task, in which automated responses need to be inhibited.[17,18,19]

In an early attempt to explain a level dependency of adverse cognitive noise effects, Kahneman[20] suggested arousal to be the mediator between noise level and altered attention functions back in the 1970s (cp. also[21,22]). Here, arousal is understood in terms of a general cortical excitation level. Noise as sensory stimulation increases arousal, which is assumed to decrease the breadth of attention. In fact, loud noise has been found to induce performance relevant alterations of attentional functions. Hockey[23] reported that loud noise of 100 dB(A) enhanced processing of central visual stimuli, while the processing of peripheral stimuli was reduced compared to a more intermediate noise condition of 70 dB(A) (cp. also[24,25]). Consistent with this finding, a magnetic resonance imaging (MRI) study found neurophysiological differences during noise of 80 dB(A) between the processing of tasks that necessitated a wide focus of attention compared to tasks that require a narrow focus.[26] Comodulations of attentional functions and noise effects have been also reported in a series of recent publications. These studies suggest that the potential of noise events to distract attention varies with the participants’ concentration on the focal task at hand [e.g.[27,28]].

A level effect of noise on attention-based cognitive tasks might thus not be restricted to high noise levels but might also be found for more moderate and realistic road traffic noise-level ranges, for example 50-70 dB(A). Road traffic noise has not yet been evaluated from this perspective. However, level within a moderate range has been shown to be mostly irrelevant to the effects of background speech and its effects on another basic cognitive function, namely, storage functions. For this noise source, the interplay of sound signal and task characteristics has been widely explored over the last two decades for at least two reasons. First, complaints in offices are most frequently related to background speech produced by colleagues talking to each other or on the phone. The question of whether not only subjective ratings but also objective performance is reduced during background speech in office settings is of high economic interest (cp.[29]). Second, background speech affects performance in certain tasks reliably and with high effect sizes, which has been verified in particular for verbal serial recall.[30] This task is the standard measure for short-term storage — participants are asked to recall a sequence of digits immediately after presentation in exact serial order. The detrimental impact of background speech on this task has been shown to be independent of speech levels within the range of 35-85 dB(A).[31,32] This finding is due to the fact that the noise effect in serial recall is to a large extent determined by the speech signal's distinct temporal-spectral variations,[33] which, however, persist in auditory-perceptive perspective despite the lowering of the overall level. Here, the underlying cognitive mechanisms for a noise-induced performance decrement are assumed to be interference processes between the involuntary processing of the seriation of distinct auditory-perceptive speech tokens and the voluntary processing of serial information necessary to solve the verbal serial recall task (interference-by-process principle[34,35]). The recent duplex-mechanisms account of auditory distraction[36,37] differentiates such highly specific interference processes from attentional distraction as a more general mechanism for noise to affect cognitive performance. Auditory distraction results from unexpected changes in the auditory background (e.g., change in background speaker voice).[27] However, the effects of noise on other attentional functions, e.g., sustained attention or selective attention, are not accounted for by this framework for cognitive noise effects.

The example of background speech nicely demonstrates that a certain cognitive function (short-term storage) can be sensitive regarding a specific background sound feature (temporal-spectral variations) while being largely immune to another sound characteristic (level). The question of interest of the present study is whether the effects of road traffic noise on attentional functions vary with level even within a moderate range. Yet, it is a challenge in cognitive noise research that performance in a given cognitive task is not administered only by a single cognitive process or function.[38,39] For example, although performance in verbal serial recall relies particularly on storage functions, the effects of attentional distraction have been shown to be also involved in the decrement of performance during certain noise conditions (e.g.[40,41,42]). Analogously, performance in tasks that predominantly call for attentional functions might also ask for other basic cognitive functions, e.g., storage or reasoning functions. Consequently, we exemplarily chose three different cognitive tasks to test the performance effects of road traffic noise at different levels in the present study. These tasks can be assumed to vary with respect to the extent to which successful task performance relied on storage and attentional functions. In addition to performance data, subjective annoyance ratings were collected as an important yet supplemental dimension for noise evaluation.

Experiment 1 tested Stroop performance. Here, performance necessitated focussing on one dimension of the stimulus’ attributes, while inhibiting responding to another dimension. Thus, two attentional functions are indispensable here, namely selective attention and inhibiting automated responses, while the storage of any information was not necessary to solve the task. In Experiment 2, a specially designed multistage mental arithmetic task was applied in which performance relied on both storage and attentional functions. The correct final result could only be obtained by chance if intermediate results could not be upheld or if attention could not be maintained over a longer period of time (several seconds up to minutes). Thus, the attentional function of “sustained attention” while processing and manipulating the presented information was essential here. Finally, verbal serial recall was examined in Experiment 3, which is in the first instance a storage task with the participants being asked only to retain and recall the presented information without further processing or manipulation.

Experiment 1: Stroop Performance During Traffic Noise and Background Speech

Experiment 1 explored the effects of traffic noise of various levels on performance in the Stroop test[43] and thus in particular on inhibition processes and selective attention, whereas no information storage is necessary (see Introduction). In this task, different-colored words were displayed (blue, green, red, yellow) and were either printed in the same color as their semantic meaning (congruent item, e.g., the word “blue” displayed in blue) or in another color (incongruent item, e.g., the word “blue” displayed in red). Participants were asked to press a key corresponding to the color in which the word was printed (e.g., red) and not the word's semantic meaning (e.g., blue). (Keyboard button functions were verbally coded and not colored, e.g., the participant pressed “r” in the latter example of “blue” displayed in red).

Reading the semantics of a word is an automated process for skilled readers. Thus, in the case of an incongruent item, naming of the word's semantic meaning has to be inhibited and selective attention has to be paid to the print color as a further stimulus dimension to execute the correct response. The so-called Stroop effect describes the increase in errors and reaction times for incongruent items compared to congruent items.

Methods

Participants

Twenty-four students of the Catholic University of Eichstätt-Ingolstadt took part in Experiment 1 (17 female). The participants were 19-31 years old (median (Md) = 23 years) and were all native German speakers. All participants reported normal hearing and had answered a notice seeking participants. A small allowance was paid or credit points were given for participation.

Material and apparatus

The experiment was run on a Pentium I PC using the experimental software ERTS (Beringer, BeriSoft Cooperation, Frankfurt, Germany).

As stimuli, the words blue, green, red, and yellow (in German: blau, grün, rot, gelb) were printed in one of these colors. Words were presented in the middle of the screen in randomized order (2000 ms on, 100 ms off; font: Arial, 48 pt). A fixation cross preceded each stimulus (500 ms on, 500 ms off).

Altogether, the following six sound conditions were included: A silence condition (overall control condition), irrelevant speech, and four traffic noise conditions. The irrelevant speech consisted of German narration and was thus semantically meaningful to our German participants (Büchner, G., 1988. Helmut Griem spricht Georg Büchner. Teldec Record Service. ASIN: B0000245SH). This speech was played back at L_eq = 60 dB(A) (Speech 60 dB(A)). All traffic noise conditions comprised recordings of motorcars in traffic light situations (drive away) or passing by at 20-50 km/h (19-31 mph). The traffic noise condition with 2000 passings per hour varied in level (L_eq = 50 vs. 60 vs. 70 dB(A): Traffic₂₀₀₀ 50 dB(A), Traffic₂₀₀₀ 60 dB(A), Traffic₂₀₀₀ 70 dB(A)). This variance was achieved by amplifying or attenuating the traffic noise signal Traffic₂₀₀₀ 60 dB(A) using the software Artemis 6.0.200 (Head Acoustics GmbH, Herzogenrath). Additionally, a traffic noise condition comprising 100 passings per hour was included and played back at L_eq = 60 dB(A) (Traffic₁₀₀ 60 dB(A)). The quoted sound pressure levels refer to an energy-equivalent sound pressure level, L_eq, averaged over presentation duration and measured using an artificial ear (Brüel & Kjaer 4153) and sound pressure meter (Brüel & Kjaer 2231). All background sounds were presented diotically using a Sony CDP-103 CD-player and Sennheiser HD 600 headphones.

Procedure

The experiment was carried out in single sessions in a double-walled IAC soundproof booth (IAC Acoustics, Niederkrüchten, Germany) at the Catholic University of Eichstätt-Ingolstadt. The experimental session began with written instructions. Background sounds were stated to be irrelevant for the task; therefore, they were to be ignored and no questions concerning the sounds would be asked.

Each participant went through two testing sessions (appr. 60 min each) on different days, both encompassing three sound conditions. The sequence of sound conditions was counterbalanced over participants. In each of the two testing session, 24 test trials were given during silence. Then, 192 trials had to be solved during each sound condition. Half of the trials consisted of incongruent words (semantic meaning differed from the color in which it was printed), the other trials consisted of congruent words. These trials were presented in randomized order for each participant. Participants had to press “r” on a German keyboard if the print color of a word was rot (red), “g” for gelb (yellow), “b” (pasted over the ‘/’ key) for blau (blue), and “ü” for grün (green).

After each sound condition, a pause of 3 min followed in whcih participants were asked to answer the question “How annoying were the background sounds in this experimental block to you?” The question was printed on a piece of paper for each sound condition and was answered on a 5-point scale with the verbal anchors “not at all,” “a little,” “middle,” “rather,” and “extremely” (in German: “gar nicht,” “kaum,” “mittelmäßig,” “ziemlich,” and “außerordentlich”).

Results

Each word for which the print color was not indicated counted as an error. Performance data were analyzed using a 2-factorial analysis of variance (ANOVA)¹ with the variables congruency (incongruent word, congruent word) and sound condition as within-subject factors. The interaction between these two factors was significant [F(5, 115) = 3.14, mean squared error (MSE) = 0.00014, p = .011, partial η² = 0.12], while the main effect on sound narrowly missed the 5% α-error level [F(3.4, 78.3) = 2.51, MSE = 0.00032, p = .06, partial η² = 0.10]. The main effect on congruency was significant [F(1, 23) = 26.49, MSE = 0.00085, p <.001; partial η² = 0.54]. The latter finding indicates the Stroop effect, i.e., more errors for incongruent words (M = 3.7%, standard error (SE) = 0.5%) than for congruent words (M = 1.9%, SE = 0.2%).

Because of the significant interaction of congruency and sound condition, performance data were analysed separately for incongruent and congruent items using 1-factorial ANOVAs on the within-subjects factor sound condition. No significant sound effect was given for congruent words [F(5, 115) <1], but a significant effect did exist for incongruent words [F(5, 115) = 4.29, MSE = 0.00022, p = .001, partial η² = 0.16]. The latter effect was due to significantly more errors during the loudest traffic noise condition, Traffic₂₀₀₀ 70 dB(A), compared to the silence and other sound conditions (p < .05, two-tailed, 0.29 ≤ Cohen's d ≤ 0.58), as t-tests with α-error adjustment following Benjamini-Hochberg revealed.[44,45] Although performance during the Traffic₂₀₀₀ 50 dB(A) condition was significantly better than that during the loud traffic noise condition (70 dB(A)), it still reduced Stroop performance in tendency compared to silence (p = .05, two-tailed, Cohen's d = 0.36), contrary to the two 60 dB(A) traffic noise conditions (p ≥.08, two-tailed) and to speech (p = .37, two-tailed). Only one other comparison reached statistical significance, although with a small effect size: Performance during Traffic₂₀₀₀ 60 dB(A) was significantly better that than during Traffic₂₀₀₀ 50 dB(A) (p = .04, two-tailed, Cohen's d = 0.22). The left panel of Figure 1 depicts mean errors with standard errors for incongruent stimuli.

Figure 1.

Impact of traffic noise varying in level on incongruent item performance in the Stroop test and on subjectively rated annoyance in Experiment 1 (n = 24). Means and standard errors of relative error rates and annoyance ratings are plotted

To note, enhanced error rates for incongruent words were not due to a speed-accuracy trade-off.[46] This phenomenon describes a strategy to produce faster responses while making more errors. On the contrary, reaction times for incongruent words (M = 695 ms, SE = 29 ms) were about 100 ms slower than reaction times for congruent words (M = 596 ms, SE = 20 ms). This effect was statistically significant as verified by a significant main effect on congruency [F(1, 23) = 60.75, MSE = 11696.43, p <.001, partial η² = 0.732] based on a 2 × 6 ANOVA on the within-subjects factors congruency and sound. The main effect of sound was not significant [F(5, 115) <1], which also holds true for the interaction of sound and congruency [F(2.1, 48.1) <1].

The different sound conditions also affected annoyance ratings significantly, as verified by a 1-factorial ANOVA [F(5, 115) = 17.39, MSE = 1.0959, p <.001, partial η² = 0.43]. In contrast to performance data, all traffic noise conditions and background speech were rated as significantly more annoying than the working during silence condition (p <.001, two-tailed, 1.35 ≤ Cohen's d ≤ 2.66). Furthermore, the loudest traffic noise condition, Traffic₂₀₀₀ 70 dB(A), during which most errors were made in the Stroop task, was rated as significantly more annoying than the other traffic noise conditions (p ≤.004, two-tailed, 0.62 ≤ Cohen's d ≤ 0.99) as well as compared to speech (p = .002, two-tailed, Cohen's d = 0.95). No further comparisons reached the 5% α-error level (p ≥.083, two-tailed).

Experiment 2: Mental Arithmetic During Traffic Noise and Background Speech

Experiment 2 explored the effects of road traffic noise that varied in level on a modified version of the Konzentrations-Leistungs-Test (K-L-T, Concentration Performance Test).[47,48] In the present version of this mental arithmetic task, participants calculated three arithmetic problems, memorized the intermediate results, and applied a certain calculation rule to those results to obtain the final result. Specifically, three addition and subtraction problems were presented that each consisted of three summands between 1 and 9 and two operands (+ or −), (e.g. 4 − 2 − 1; 3 + 9 + 6; 6 + 5 − 3). The arithmetic problems and summands, respectively, were displayed one after the other. Participants had to memorize the intermediate results for each arithmetic problem (here: 1, 18, 8) and apply a defined calculation rule. If the first intermediate result was smaller than the third, both had to be added (here: 1 + 8 = 9), otherwise the intermediate results were subtracted. This newly calculated intermediate result was compared with the result of the second arithmetic problem (here: 18), and the foresaid calculation rule had to be reapplied to obtain the final result (here: 18 − 9 = 7). Obviously, the correct final result could only be obtained by chance when attention could not be sustained during information processing and manipulation while the intermediate results needed to be maintained.

Methods

Participants

Eighteen students of the Catholic University of Eichstätt-Ingolstadt took part in the experiment (13 female, Md = 23 years, range: 19-31 years). All aspects of participants’ recruitment, requirements and compensation were the same as Experiment 1.

Material and apparatus

This experiment was run on a Macintosh iBook G3 computer using the experimental software Psyscope 1.2.5.[49] Background sounds and their presentation conditions remained unchanged from Experiment 1. The three summands in the arithmetic problems were digits from 1 to 9. These were presented visually (font: Chicago, 50 pt) and successively in the middle of the screen. The second and third summands were preceded by a simultaneously presented operand (+ or −; font: Chicago, 50 pt).

Procedure

The testing procedure described in Experiment 1 was altered exclusively with respect to the following aspects: A trial encompassed the successive presentation of the three arithmetic problems (summand after summand) and entering the final result by the participant. The presentation of the first summand of the three arithmetic problems was announced by three rectangles that decreased in size in the middle of the screen 3 s in advance. From the second summand onwards, summands were presented in a self-paced manner by the participant pressing the space bar on the keyboard. The calculation rule, which had to be applied to the intermediate results, was plotted at the bottom of the screen throughout the trial. Participants were requested to type the final result. It was possible to correct errors until the “return” key was pressed.

Each participant went through two testing sessions on different days, both encompassing three sound conditions. One experimental session took about 90 min to complete. The sequence of sound conditions was counterbalanced over participants. In the two testing sessions, the test trial presentation started after 10 practice trials in silence, in which the intermediate results and the correct end results were given to participants. Each of the six sound conditions was presented for 15 min, and the participants had to solve as many trials as possible. Trials were randomized with respect to the number of additions and subtractions per trial, as well as the number of necessary additions or subtractions of a tens column.

Results

Each false final result was counted as an error, and the relative error rate per participant and sound condition were calculated. Because participants worked through each trial in a self-paced manner, a different number of trials was worked out in each sound condition. Thus, performance data were analyzed nonparametrically. The left panel of Figure 2 shows error medians for each sound condition, with deviation bars depicting the first and the third quartiles. A nonparametrical 1-factorial ANOVA following Friedman on the within-subjects factor sound condition was conducted. This analysis revealed a significant sound effect (χ² = 11.90, df = 5, p = .036), which was further clarified by single comparisons using nonparametrical Wilcoxon tests. Only background speech reduced mental arithmetic performance significantly compared to silence (Z = −2.07, p = .039, two-tailed), whereas performance during each traffic noise condition was not significantly different from performance during silence (p ≥.472, two-tailed). However, the best performance was observed during the Traffic₂₀₀₀ 50 dB(A) condition, which was significantly better compared to the Traffic₂₀₀₀ 70 dB(A) (Z = −2.03, p = .043, two-tailed) and speech (Z = −2.46, p = .014, two-tailed) conditions. Of note, the sound effects on performance were not due to a speed-accuracy trade-off as a nonsignificant 1-factorial ANOVA following Friedman on reaction times verified (χ² = 3.24, df = 5, p = .663).

Figure 2.

Impact of traffic noise varying in level on mental arithmetic performance and on subjectively rated annoyance in Experiment 2 (n = 18). Medians of error rates (errors bars depicting the first and third quartile) are plotted as well as annoyance rating means with standard errors

The right panel of Figure 2 depicts subjective annoyance ratings. Here, a 1-factorial ANOVA verified a significant sound effect [F(5, 85) = 6.82, MSE = 1.357, p <.001, partial η² = 0.29]. T-tests with α-error adjustment following Benjamini-Hochberg revealed that the participants considered all sound conditions as significantly more annoying than silence (p ≤.030, two-tailed, 0.74 ≤ Cohen's d ≤ 1.35) except for the softest traffic noise condition, Traffic₂₀₀₀ 50 dB(A) (p = .189, two-tailed). Furthermore, the softest sound condition was rated as significantly less disturbing than the speech and traffic noise conditions at 60 dB(A) and 70 dB(A) (p ≤.038, two-tailed, 0.52 ≤ Cohen's d ≤ 1.00). No other comparisons reached statistical significance.

Experiment 3: Serial Recall Performance During Traffic Noise and Background Speech

Experiment 3 explored the effects of road traffic noise of different levels on verbal serial recall performance compared to irrelevant background speech. This task places a burden predominantly on storage functions because it requires successively presented digits to be recalled in the exact presentation order.

Methods

Participants

Eighteen students of the Catholic University of Eichstätt-Ingolstadt took part in the experiment (14 female). Participants were 20-39 years old (Md = 23 years). Participant recruitment, requirements and compensations were the same as in Experiments 1 and 2.

Material and apparatus

Experimental hardware as well as background sounds and their presentation conditions remained unchanged from Experiment 2. As material to be serially remembered, the digits 1 to 9 were successively presented visually in the middle of the screen in randomised order (700 ms on, 300 ms off; font: Chicago, 56 pt). The beginning of the list was announced by three rectangles that decreased in size in the middle of the screen 3 s in advance.

Procedure

The testing procedure described in Experiment 1 was altered exclusively with respect to the following aspects. A serial recall trial started with the presentation of the digit sequence to be remembered and was followed by a 10 s retention interval and digit recall. To complete the recall task, a 3 × 3 display of rectangles appeared in which the digits were randomly reordered for each trial. The task was to click the digits following the sequence in which they had been presented. After clicking, the color of the digit and their fields inverted so that the digit could not be selected again. It was not possible to correct errors or skip serial positions.

After nine practice trials in silence, test trial presentation began. 20 serial recall trials were presented successively during the six sound conditions. After 10 trials, a pause of 10 s was given. The sequence of sound conditions was counterbalanced across participants.

Results

Each digit not recalled in its previously presented serial position was scored as an error. The left panel of Figure 3 depicts mean error rates with standard errors. An ANOVA revealed a significant effect on the within-subjects factor sound condition (F(2.9, 48.6) = 5.94, MSE = 85.63, p <.001, partial η² = 0.26). This sound effect was exclusively due to the detrimental effect of background speech on serial recall performance as t-tests with Benjamini-Hochberg α-error adjustment verified. Background speech reduced performance significantly compared to silence and all four traffic noise conditions (p ≤.036; two-tailed; 0.64 ≤ Cohen's d ≤ 0.96), while no traffic noise condition increased error rates significantly compared to silence (p ≤.284, two-tailed). Note, that enhanced error rates during background speech were not the result of a speed-accuracy trade-off. Reaction times, measured from the appearance of the recall matrix until recall was completed, did not differ significantly during the six sound conditions (F(5, 85) = 1.83, p = .116).

Figure 3.

Impact of traffic noise varying in level on verbal serial recall performance and on subjectively rated annoyance in Experiment 3 (n = 18). Means and standard errors of error rates and annoyance ratings are plotted

The right panel of Figure 3 displays subjective rating means on perceived annoyance for each sound condition. The analysis revealed a significant sound effect (F(5, 85) = 42.62, MSE = 0.446, p <.01, partial η² = 0.72). Background speech was rated as the most annoying sound condition and was significantly more annoying than all traffic noise conditions and silence (p <.001; two-tailed; 0.65 ≤ Cohen's d ≤ 3.91). In contrast to performance, traffic noise affected annoyance ratings as well; all traffic noise conditions were rated as significantly more annoying than working during silence (p <.001; two-tailed; 2.05 ≤ Cohen's d ≤ 3.34). Furthermore, the loudest traffic noise condition, Traffic₂₀₀₀ 70 dB(A), was perceived as significantly more annoying than the traffic noise conditions at 50 and 60 dB(A) (p ≤.005; two-tailed; 0.44 ≤ Cohen's d ≤ 0.90), which were not significantly different from each other (p >.072; two-tailed).

Discussion

The present study explored the effects of road traffic noise at different levels (50 dB(A), 60 dB(A), and 70 dB(A)) on cognitive performance in three exemplarily selected tasks. These tasks can be assumed to differ regarding their dependency on attentional and storage functions for successful task performance. Whereas performance in the Stroop task (Exp. 1) placed a high burden on attentional functions like selective attention and inhibition processes, no information storage was necessary. The applied mental arithmetic tasks necessitated both sustained attention during information manipulation and short-term storage (Exp. 2). Finally, verbal serial recall performance called predominantly for storage functions (Exp. 3). Subjective ratings for perceived annoyance were collected in the three experiments to supplement the performance data.

First of all, our experiments demonstrate a level effect for road traffic noise within the moderate range of 50-70 dB(A) for performance in tasks that call for attentional functions, namely the Stroop test and the nonautomated multistage mental arithmetic task. This level effect is analogous to that found for high-level continuous noise (85-100 dB(A); e.g.[14,15,16,17,18,19]). Specifically, performance during the moderate traffic noise condition at 50 dB(A) was better in the attention-based tasks (Stroop and mental arithmetic; Exp. 1-2) than during loud road traffic noise at 70 dB(A). These results speak in favor of performance in attention-based tasks being sensitive to noise level even at moderate levels.

However, the effect pattern of noise level presents itself as more complex when silence as the performance baseline is considered. Compared to silence, the loud traffic noise condition (70 dB(A)) reduced Stroop performance significantly; however, the moderate traffic noise (50 dB(A)) did so only in tendency (cp. for continuous noise,[17,18,19]). In the mental arithmetic task, the moderate traffic noise condition even enhanced performance compared to silence, whereas the loud traffic noise condition was ineffective. Thus, loud traffic noise might have resulted in an inadequately narrowed attentional focus (cp.[20,21,22], see Introduction) so the processing of relevant stimulus aspects was impaired in the Stroop test. Following this line of thought, the found performance increment in the mental arithmetic task during moderate road traffic noise would be the result of performance supporting modulation of attentional functions. In fact, comparing performance during noise and silence conditions led to inconsistent results in the past (cp.[12,13]), and some existing results even appear to be incompatible to our findings, e.g., loud noise facilitating performance in the Stroop task (e.g.[24,50]) or reducing performance in demanding mental arithmetics (e.g.,[51,52]). Consonant with this, Szalma and Hancock[13] suggested, based on their metaanalysis of cognitive noise effects contrasting performance during noise not only with silence but with the same noise at a different level. Thus, on a theoretical level, the question remains as to how the found performance effects during noise compared to silence occurred and must be passed on to future research. Do note, however, that the present finding of a level effect in terms of better performance during road traffic noise at 50 dB(A) relative to 70 dB(A) is not derogated by the question of whether an absolute performance enhancement or decrement is given compared to silence.

Besides a level effect for road traffic noise, the second main result of our study is that none of the tested road traffic noise conditions at 50-70 dB(A) had an effect on serial recall. Yet, performance in this task, which relied in particular on storage functions, was impaired significantly by background speech, as is well-documented (cp., e.g.[33]). Furthermore, mental arithmetic performance was also significantly reduced during background speech (cp.[53]). This result is presumably due to the storage component of the applied multistage task because participants needed to maintain intermediate results. This view is corroborated by the fact that irrelevant speech had no effect on Stroop performance, for which no information storage was required. Do note, however, that traffic noise conditions and background speech were only controlled for level but not matched regarding temporal-spectral variability or other sound characteristics. Consequently, the present study provides no evidence that the found differentiation between speech and traffic noise effects might be due to the different origins of the noise signals.

Admittedly, the ineffectiveness of the traffic noise conditions on short-term storage in the present study was not surprising. As described in the Introduction, the prominent temporal-spectral variability of background speech is decisive for its detrimental effect on short-term storage. Consequently, the disturbance effect of background sounds on verbal serial recall can be modeled by the hearing sensation fluctuation strength F.[33] While the used speech signal of a single German speaker hits a fluctuation strength of F = 1.15 vacil, all traffic noise conditions were characterized by a very low fluctuation strength (F = 0.08-0.14 vacil). This holds true even for the signal Traffic₁₀₀ 60 dB(A), which was constituted by 100 passings and was characterized by a more prominent temporal structure as the other traffic noise signals for which 2000 passings were auralized. However, as fluctuation strength values indicate, even Traffic₁₀₀ 60 dB(A) inherited far less prominent temporal-spectral variations as the tested background speech signal. Road traffic noise with a temporal structure similar to a background speech signal impaired storage-based tasks just as the speech signal did in another study.[5]

In sum, the found results are consistent with the well-documented distinctive interplay of task and sound characteristics for cognitive noise effects to occur (cp. for other sounds and tasks, e.g[54,55]; cp. for meta-analysis and reviews[12,13]). Summarizing the found sound effect pattern, our study is consonant with a differentiation between attention-based noise effects and interference effects that has been put forward recently — at least for auditory distraction as one single aspect of attentional functions — by the duplex-mechanism account of auditory distraction (see Introduction).[36,37] However, behavioral noise effects on other attention functions, like sustained attention or selective attention, still require up-to-date theoretical explanations, which might be stimulated by ideas on modulations of attentional functions in dependency of noise level such as those proposed by Kahneman, Easterbrook, and Broadbent[20,21,22] some decades ago (Introduction).

Besides behavioral effects, our study examined the effects of the different background signals on subjective annoyance ratings as a further important dimension for the evaluation of noise effects that are supplemental to effect patterns on cognitive performance. With respect to level, evoked performance patterns and annoyance ratings go hand-in-hand: Loud traffic noise at 70 dB(A) was perceived as significantly more annoying than was soft traffic noise at 50 dB(A) in all three experiments. This level effect of annoyance is well-documented (e.g.[9,10,11]). Furthermore, the background sound condition with the highest detrimental performance effect on a task was also rated as the most annoying sound condition in this task. This finding holds true for the traffic noise at 70 dB(A) in the Stroop task (Exp. 1) and background speech in the mental arithmetic (Exp. 2) and serial recall (Exp. 3) tasks.

In contrast to the performance data, annoyance ratings reflected a general effect of the presence of background sound. In all experiments (1-3), traffic noise conditions were rated as significantly more annoying than performing during silence, even if a sound condition did not reduce task performance (with a single exception: Traffic noise of 50 dB(A) was not rated significantly differently from silence in Exp. 2). Previous studies have reported analogous findings for other tasks and sound conditions.[53,54,56] With this, a general and task-independent preference of silence compared to any other sound condition is reflected. This finding holds true even when the background sound is as soft as a whisper.[53]

In the 1990s, about 18% of the inhabitants of the European Union (EU) were exposed to equivalent traffic noise levels in the 55-65 dB(A) range during the day and a further 16% to noise levels above 65 dB(A).[57] This number most probably has not decreased since, as a significant increase of road traffic has occurred that partly offset the technological improvements made, for example, in noise emissions from individual cars. The found effects of road traffic noise on cognitive performance as well as those on subjective annoyance demonstrate the necessity of a basic research perspective while underlining the severity of the challenges posed to road traffic noise abatement.