Noise Health Home 

[Download PDF]
Year : 2007  |  Volume : 9  |  Issue : 34  |  Page : 15--24

The development of the noise sensitivity questionnaire

Martin Schutte, Anke Marks, Edna Wenning, Barbara Griefahn 
 Institut für Arbeitsphysiologie an der Universität Dortmund (IfADo), Ardeystraße, Dortmund, Germany

Correspondence Address:
Martin Schutte
Institut für Arbeitsphysiologie an der Universität Dortmund, Ardeystraße 67, D-44139 Dortmund


The existing questionnaires for determining the noise sensitivity of individuals provide information only about global noise sensitivity, although empirical data suggest that measuring noise sensitivity for different situations in daily life might be more logical. Therefore, the «DQ»Noise-Sensitivity-Questionnaire«DQ» (NoiSeQ) was developed to measure global noise sensitivity as well as the sensitivity of five domains of daily life, namely, leisure, work, habitation, communication, and sleep. The assessment of the measurement characteristics was based on the Generalizability (G) theory. The results of the G-study (N = 66) proved that a single application of the questionnaire is sufficient for determining an individual's noise sensitivity. Furthermore, the ratings are age and gender independent. The subsequently conducted Decision (D)-study (N = 288) provides information on the reliability of NoiSeQ. If the questionnaire is used for measuring global noise sensitivity, the reliability (relative and absolute G-coefficient) reaches a value above 0.90. According to ISO 10075-3, the questionnaire satisfies the precision level 1 «DQ»accurate measurement«DQ» in this case. The G-coefficients for all the subscales exceed the lower limit 0.70, with the exception of subscale leisure, which did not prove satisfactory. However, this subscale can reach a reliability of more than 0.70 if additional items are included. The validity of the instrument was proven for the subscales habitation (N = 72) and work (N = 72). In both the studies, the participants were asked to rate the annoyance in the presence of several rail and traffic noise scenarios. The subjects were characterized as low and high noise sensitive according to their sensitivity values obtained from NoiSeQ. In conclusion, a significant difference in annoyance rates was observed between the low and high noise sensitive groups for both the subscales habitation and work. This data support the validity of NoiSeQ.

How to cite this article:
Schutte M, Marks A, Wenning E, Griefahn B. The development of the noise sensitivity questionnaire.Noise Health 2007;9:15-24

How to cite this URL:
Schutte M, Marks A, Wenning E, Griefahn B. The development of the noise sensitivity questionnaire. Noise Health [serial online] 2007 [cited 2023 Jan 27 ];9:15-24
Available from:

Full Text

The subjective experience of annoyance represents the most frequent human reaction to traffic noise. Different levels of annoyance show considerable inter-individual variations and are ascribed to the differences in the noise sensitivity. Noise sensitivity is considered as a stable personality trait, which affects an individuals' reactivity toward noise sources. According to the results of psycho-acoustic studies, noise sensitivity has no relation to auditory acuity but reflects a judgmental, evaluative predisposition towards the perception of sounds. [1]

In Germany, various questionnaires for the measurement of noise sensitivity such as the Weinstein-scale (WSS) [2] and the Fragebogen zur Erfassung der individuellen Lδrmempfindlichkeit (LEF) [3],[4] are easily available in long and short versions. All instruments require the respondent to indicate the degree of affirmation to several statements describing various affective, cognitive, perceptual, and behavioral reactions to noises that occur in daily life. The level of noise sensitivity is calculated as the total of the rating values for the different items. The currently available questionnaires for the determination of individual noise sensitivity focus on global noise sensitivity. Nevertheless, the analysis of the factorial structure of the LEF leads to a multidimensional solution indicating that separate measurements for different areas of daily life might be more appropriate when determining the level of noise sensitivity. Therefore, the "Noise-Sensitivity-Questionnaire" (NoiSeQ) was developed to measure global noise sensitivity as well as the sensitivity for different domains of daily life.

 Development of the NoiSeQ

Based on the multidimensional pattern attained for the LEF, [3] five areas can be regarded as relevant for measuring noise sensitivity, namely, leisure, work, habitation, communication, and sleep. Firstly, the item collection was accomplished taking into consideration the items of the WSS and LEF questionnaires. The items were partly reformulated to achieve a better content-related understanding. The development of the subscales necessitated the creation of completely new situational descriptions. Subsequently, five judges assigned the items to the following five categories: leisure, work, habitation, communication, and sleep. Only those items that were concordantly classified by all judges in only one of the five categories were taken into further consideration. For each subscale, a random sample of seven items was taken from the respective category. Consequently, the final version of the questionnaire comprised a total of 35 items (Annex 1). The respondents were asked to indicate the extent to which the items applied to their attitudes using a four-level rating scale (strongly agree = 3, slightly agree = 2, slightly disagree = 1, and strongly disagree = 0). To calculate the characteristic value for the global noise sensitivity, the average of the rating values of all 35 items had to be calculated and for the subscales, the mean value based on the ratings of the corresponding seven items needed to be calculated.

 Testing reliability

The NoiSeQ reliability analysis is based on Generalizability (G) Theory. In comparison to classical test theory, this procedure has the advantage that it allows the estimation of the magnitude of multiple sources of measurement error (so-called facets), and thus, facilitates the separation of the major sources of error. [5],[6],[7] One important consideration of G-Theory is that although a measurement instrument is applied under specific conditions the user makes generalizations of the results under similar conditions. [8] Accordingly, a measurement represents a sample taken from the universe of all possible measurements. The average value of all these measurements-the universe score-would, therefore, represent an ideal base for statements concerning the level of noise sensitivity. However, this value cannot be observed. A reliability analysis based on G-theory gives information about how well conclusions from the observed measurement to the universal score can be drawn. First, the object of measurement needs to be specified. Since NoiSeQ should provide information concerning individual noise sensitivity the individuals themselves are the object of measurement. Furthermore, the application of G-theory necessitates a precise description of all facets and conditions the user of an instrument would want to generalize. Therefore, the extent to which the different facets influence the measurements must be determined. This information is proved via G-study, which is based on an analysis of variance approach (ANOVA). For this purpose, random, fixed, or mixed ANOVA models can be used. A facet is considered as a random effect if its conditions are a random sample from the corresponding population. A facet is considered to be fixed if the design takes into consideration all possible conditions of that facet (e.g., gender). The contribution of each facet to the measurement error is determined via the estimation of variance components. [9] The D-study supplies information not only about the expenditure involved with the application of the measurement procedure but also about alternative measurement designs that would result in reliable measurement values. There are two different parameters describing the reliability of an instrument. The relative G-coefficient (ρ2 ) indicates how well an observed score is likely to locate individuals (or conditions), relative to other members of the corresponding population, on a 0-to-1 scale. The absolute G-coefficient (φ) also indicates how well an observed score is likely to locate individuals (or conditions) without regard to others in the respective population on a 0-to-1 scale.

Based on empirical findings, it is demonstrated that noise sensitivity could be influenced by age and gender [10],[11] and that both variables represent possible sources of measurement error and should be included in the G-study. Since noise sensitivity is defined as a stable characteristic of a person it will be logical to determine whether the measurements are time dependent. Furthermore, it must be clarified whether situational shifts exist in noise sensitivity restricting the informational value of the NoiSeQ score, which indicates the level of global noise sensitivity. Although the subscales were developed in order to achieve a high content-related homogeneity in each case, inconsistencies in ratings could not be ruled out. Thus, the items should also be considered as a potential source of error.

The different facets could not be combined completely since a person belongs only to one gender and one age group and an item was associated with only one such subscale. Therefore, the G-study was based on a nested design where an individual was nested within an age group and gender type, and the items were nested within subscales. Further, the properties of the ANOVA-model must be specified. Based on G-Theory, the effects can be specified as random. But such a definition does not make sense regarding gender and subscale, because male and female represent the whole population concerning the facet gender gender and the noise sensitivity for the five daily situations cannot be generalized with other situations. Therefore, these two facets are 'fixed' facets.


The study sample consisted of 66 individuals, of which 33 were women and 33 men; all were divided into three different groups based on their ages [Table 1]. The participants answered the questionnaire twice at an interval of 14 days.

A six-factorial mixed-effect ANOVA was used to analyze the data (factor 1: gender, factor 2: age, factor 3: person nested within the factors gender and age, factor 4: occasion, factor 5: subscales, factor 6: items nested within the factor subscales).

[Table 2] shows the calculated mean squares, the estimated variance components, their standard errors as well as the explained proportion of variance in each case. According to the table, 1.8% of variance can be traced back to the facet gender indicating that men and women do not differ substantially in their rating level. Since only 3.6% of variance can be ascribed to age, the age factor also has no marked influence. In contrast, 18% of variance is caused by inter individual differences; it must be considered that this component contains the interactions G * P, A * P, and G * A * P to an unknown extent. The estimated variance component for the facet occasion (σ2 = 0.0) justifies the assumption that the ratings are not time dependent. The main effect for the subscales is small. Consequently, the subscales are comparable with respect to their mean values. A 13% of variance can be traced back to the facet item, indicating that some inhomogeneity exist with regard to the content. But it must be considered that this component inseparably includes the interaction S * I. From the remaining sources of variance, the interaction P * S: G: A explains nearly 6% of variance. The relative standing of the individuals differed from scale-to-scale. Nevertheless, it must be regarded that this component includes additionally the interactions P * S * G, P * S * A as well as P * S * G * A.

Despite these confounding effects, the result justifies the conclusion that noise sensitivity depends on the five domains of daily life. The component for the interaction P * I: G * S * A binding nearly 25% of variance contains further fractions of interactions. Nevertheless, this component shows that the ratings of the individuals differ with particular item. The residual variance, which contains the interaction P * M * I: G * S * A as well as all other uncontrolled effects, explains 26% of data variance.

Based on the results of the G-study, NoiSeQ does not depend on the age and gender of the respondents because the measurements are not influenced by these facets. Furthermore, the recording of noise sensitivity can be confined to a single measurement due to the extremely small time-dependent variation of the ratings. However, the level of global noise sensitivity depends on a particular daily situation. Accordingly, the effects resulting from the facets subscale and item must be taken into consideration further, since the estimated variance components deviate substantially from zero as the corresponding standard errors verify.


The D-study aimed at getting information concerning the reliability of the global score of the NoiSeQ taking into consideration only the relevant sources of measurement error, namely, the facets item and subscale whereby the individuals represent the object of measurement. The D-study is based on a study sample comprising 288 men and women belonging to different age groups. A mixed model ANOVA with subscales as a fixed effect and persons and items as random effects was used to analyse the data [Table 3]. Further, the items were nested within the levels of the factor subscale. Accordingly, 20% of data variability is caused by inter individual differences. A 16% of variance can be traced back to the factor items. Therefore, the evaluations depend on the situational descriptions expressed by the single items. The differences between persons vary in dependence on the kind of domain of daily life as the non negligible proportion of variance indicates, which can be traced back to the interaction P * S (10%). The most striking variance component resulted for the residual. 54 % of data variance can be traced back to interaction P * I: S and further uncontrolled effects. The estimations of variance components show no instabilities as the calculated standard errors verify.

The reliability of the measured overall noise sensitivity can be indicated via the relative G-coefficient reflecting how well a person can be located relative to other members of the population. Starting from the estimated variance components, the relative G-coefficient (ρ2 ) takes a value of 0.93 [Table 4], which is above the lower limit of 0.90 recommended by ISO 10075-3 for high precision measurements. The absolute G-coefficient (f) takes a value of 0.91, which also exceeds the critical value of 0.90 [Table 4].

Furthermore, the confidence intervals for the overall noise sensitivity take values of were ±0.26 (α = 0.05) and ±0.34 (α = 0.01). In order to facilitate the classification of the measured global noise sensitivity the quantiles of the distribution of the global noise sensitivity scores are listed in [Table 5].

Overall the NoiSeQ allows relative and absolute conclusions concerning the individual level of global noise sensitivity with high precision.

However, the reliability of the five subscales must be proved. Accordingly, a 2-factorial ANOVA was accomplished for each subscale (factor 1: person and factor 2: item), whereby in each case all facets were treated as random effects.

The results show that the subscales sleep, communication, habitation, and work are much more responsive to inter-individual differences than the subscale leisure; for the subscale leisure the proportion of variance explained by the factor person is 17%, whereas the corresponding proportion of variance for the other four scales is 25% and more [Table 6].

Considering the subscales communication, work, and leisure, the proportion of variance explained by the facet item is more than 19% indicating non-negligible differences in item difficulty, which is less important when considering the corresponding results for the scales sleep (7.6%) and habitation (9.7%). The largest fraction of variance that is more than 50% is explained by the residual, which includes the interaction P * I as well as all other uncontrolled effects. The standard errors of the variance components denote that the estimations are quite stable.

The relative G-coefficient of the subscale sleep is ρ2 = 0.84 which is in accordance with measurement precision class 2 according to ISO 10075-3 [Table 7]. The reliability of the subscale communication (ρ2 = 0.79), habitation (ρ2 = 0.77), and work (ρ2 = 0.77) is slightly smaller, but conforms to the measurement precision class 3 in each case. The relative G-coefficient of the subscale leisure (ρ2 = 0.67) falls below the lower limit value of 0.70. The absolute G-coefficient of the subscale sleep fulfills the requirements of measurement precision class 2 [Table 7]. The subscales communication, habitation, and work are applicable if orienting measurements are intended. The absolute G-coefficient of the subscale leisure is far below the lower limit value of 0.70.

The improvement of the reliability of this scale can be technically achieved by increasing the number of items. Based on the available estimated variance components, a reliability of at least 0.70 postulates 9 (relative G-coefficient) respectively 12 items (absolute G-coefficient). The confidence intervals for the sensitivity scores of the subscales [Table 7] were ±0.55 or ±0.56 ( p = 0.05) or they varied between ±0.72 and ±0.74 ( p = 0.01) with exception of the confidence intervals of the subscale leisure which was ±0.68 ( p = 0.05) and ±0.90 ( p = 0.01).

[Table 8] represents the quantiles of the distribution of the scores for the five daily situations in order to facilitate the interpretation and classification of the measured noise sensitivity level.

Results of first tests based on the validity of NoiSeQ

Since some of the items of NoiSeQ are assembled from other existing noise sensitivity measuring questionnaires such as the LEF [3] or WSS, [2] the determination of internal validity was set aside. Furthermore, the determination of internal validity would have been possible only for the global noise sensitivity score.

Therefore, the validity of the questionnaire was proved by comparing individuals with low or high noise sensitivity based on their results obtained from NoiSeQ. Noise sensitivity is defined as a persons' condition enhancing their reactivity to noise; based on this definition, the validity of the NoiSeQ could be tested in experiments analyzing the effect of noise and noise sensitivity on annoyance. Accordingly, there should be a verifiable significant difference in the evaluation of the annoyance of traffic noise between individuals with low and high noise sensitivity, since the empirical findings show that noise sensitivity directly influences annoyance. [12],[13] Furthermore, interactive effects of the factors sound level and noise sensitivity on annoyance are conceivable as suggested by a meta-analysis that revealed the relation between noise sensitivity and reactions to noise. [14]

 Validity of the subscale habitation

First, the validity of the subscale habitation was analyzed. For creating noise conditions, road and rail traffic noise scenarios (only passing vehicles) were created. The scenarios were of 3 minutes each. Each scenario was available at 8 different levels (L Aeq = 40, 46, 52, 58, 64, 70, 76, 82 dB). A 5.1 Dolby-Surround system was used to create the noises. Annoyance was recorded using an adapted version of the rating scale recommended by the International Commission on the Biological Effects of Noise (ICBEN): "How much did you feel annoyed or disturbed in total by the noise?" (ISO/TS 15666). [15] The ratings were measured by means of a vertically oriented rating scale consisting of five categories. The labels followed the original suggestion of ICBEN. Each category was subdivided into 10 graduations resulting in a 50-point rating scale. Similar to the category loudness scale, [16] the participants were instructed to choose the first appropriate category and then to make a second finer choice. The chosen grade was represented as a number. The numbers from 1 to 10 referred to "not at all," the digits from 11 to 20 referred to "slightly" and so on with 50 being the maximum.

In order to get information about the stability of the annoyance ratings, the noises with L Aeq = 40, 52, 70, 82 dB were presented twice. In total, an experimental session comprised 24 noise scenarios. Twelve scenarios of each type of traffic noise were presented in a random order. Each noise presentation was followed by a break of 15 seconds. The experiments were conducted in three laboratories with its acoustical properties adjusted as recommended for using 5.1-Dolby surround system (ITU-R BS.1116-1). [17] The loudspeakers were positioned in a circle (radius, 2 meters). The participants were made to sit in the centre of the circle. In order to guarantee a standardized experimental run, the entire procedure was computerized. The NoiSeQ, the sounds, and the annoyance rating scales were automatically presented, while the responses were entered into the computer manually. Firstly, the participants were asked to answer the NoiSeQ. Then, the participants were familiarized with the rating scales. In order to establish inter individual comparable reference systems, the road and rail traffic sound with the lowest and highest level were presented to the participants. The participants were instructed to imagine a situation at home during daytime. This experimental technique was successfully performed to determine the relationship between loudness and annoyance. [18] After presenting one of the noise scenarios, the participants were asked to estimate their level of annoyance.

The study sample comprised 72 individuals with normal hearing, mainly students of the universities of Dortmund, Eichstδtt, and Essen. Based on the individual NoiSeQ scores for the subscale habitation, 25 individuals were characterized by low and 22 by high noise sensitivity (sensitivity score ≤25. and ≥75. percentile, respectively).

A 4-factorial ANOVA was accomplished with the factor level of noise sensitivity as a between subject factor. The remaining factors, namely, type of traffic noise (road and rail traffic), sound level (L Aeq = 40, 52, 70, 82 dB), and measurement time were specified as within factors. The results showed that the factor level of noise sensitivity had a significant effect on annoyance ratings (F = 4.25; df = 1, 45; p p p p = 0.03).

Since the observed differences between the low and high sensitivity groups were consistent with the hypothesized differences concerning the annoyance ratings, the experimental results were concluded to favor the validity of NoiSeQ.

Validity of the subscale work

The validity of the subscale work was analyzed in a second experiment with a separate group of participants. A significant detectable difference between individuals with low and high noise sensitivity with regard to their annoyance ratings is hypothesized. For testing this hypothesis, the effects of different types and levels of traffic noises on annoyance during the task were investigated. For manipulating the noise conditions, all possible combinations by mixing road and rail traffic noises with sound levels of L Aeq = 34, 46, and 64 dB in each case were composed resulting in nine different noise scenarios of 5 min each. For presenting traffic sounds, a 5.1-Dolby surround system was employed. The easy and difficult versions of the grammatical reasoning test (GRT) taken from the Criterion task set were performed during noise presentations [19] . First, the participants were asked to answer NoiSeQ. Then, the participants were provided instructions with regard to the GRT tasks prior to a training session of 21 min. Feedback concerning their performance was directly provided to the individuals after the accomplishment of each task. Further, the individuals were familiarized with the rating scales for annoyance and its usage (see 4.1). Subsequently, in order to establish inter individual comparable reference systems, pink noises with the lowest (L Aeq = ~37 dB) and highest sound levels (L Aeq = ~67 dB) occurring during the experimental sessions were presented. After an interval of 15 minutes, the participants were asked to complete the GRT tasks in presence of the traffic sound, and immediately upon completion of the task, they were asked to present their annoyance ratings. The human preparedness for work is subjected to fluctuations depending on the time of the day, and therefore, the experimental design took this variable into consideration. Correspondingly, the experiments were twice accomplished in two balanced sequences, namely, in the morning and in the afternoon.

A total of 72 normal hearing individuals participated in the experimental study. Based on the NoiSeQ-score for the subscale work, 7 participants were characterized by low and 22 by high noise sensitivity (sensitivity score 75. percentile, respectively).

A 5-factorial ANOVA was used to analyze the data. The factor noise sensitivity (low and high) represented a between subject factor and the remaining independent variables, namely, road noise (L Aeq = 34, 46, 64 dB), rail noise (L Aeq = 34, 46, 64 dB), time of day (morning, afternoon) and task difficulty (low and high) were specified as within subject factors.

The factor noise sensitivity had a significant effect on annoyance (F = 5.52; df = 1, 27; p = 0.03). The average annoyance rating of the low noise sensitivity individuals was 17.75, while that of high noise sensitivity individuals was 24.44. This result corresponds to the hypothesis suggesting a difference in annoyance ratings between individuals characterized by low and high noise sensitivity. Further, there is a significant interaction between the factors noise sensitivity and rail noise (F = 9.80; df = 2, 54; p Aeq = 34 dB to L Aeq = 64 dB. This result is in agreement with the hypothesis postulating a specific effect of both factors on annoyance.

Further interactions between noise sensitivity and the factor road noise, task difficulty, and time of day did not exist. Since annoyance increased with rise in sound levels and task difficulty, the significant main effect for the factor task difficulty (F = 13.85; df = 1, 27; p p p p p [20] The reliability of the subscales exceeded the lower limit value (0.70) in each case with the exception of subscale leisure, which did not fulfill this criterion. Considerable differences in the inter-individual preferences for various recreational activities could be one of the reasons for this exception. The items of the subscale leisure presumably did not cover the entire range of possible activities. Nevertheless, the reliability of this subscale can be increased to a value of 0.70 if the number of items is increased from 7 to 9 respectively 12.

Although a complete validation of the NoiSeQ was not verified, the first analyses verified the validity of the subscales habitation and work. However, it must be taken into consideration that the participants were characterized by either low or high noise sensitivity with subscale scores clearly below quartile 1 and accordingly above quartile 3 but the subjects were not characterized by a low noise sensitivity with regard to the other daily situations.

Since the expected differences in annoyance ratings between low and high noise sensitive individuals were statistically significant, the subscale habitation was regarded as valid. Nevertheless, it must be taken into consideration that the participants were instructed to imagine a particular situation at home while assessing their annoyance, and the data with regard to the participants' imaginations were not available. Thus further studies using more standardized descriptions of the imagined situations are necessary. One possibility is that all participants can be shown a film of a particular situation along with the traffic sound presented additionally.

The experimental findings obtained for the subscale work justify the hypothesis of the validity of this scale. However, it should be considered that the results were produced by performing only one type of task, namely, the grammatical reasoning test. Therefore, it is necessary to prove that similar results can be reproduced by performing other tasks as well. The additionally existing interactions between the factor rail noise and the factor noise sensitivity corresponds with results of other studies analyzing aircraft noise. [14] In these surveys, individuals characterized by low and high noise sensitivity, indeed, differ in their levels of annoyance; however, the increase in annoyance with the rise in sound levels shows a smoother and flatter gradient. [21] The existing diversity in the experimental results can be traced back to the fact that different methods were used while measuring the noise sensitivity and the procedure for identifying low and high noise sensitive individuals also varied. Nevertheless, the reason why corresponding interaction does not exist between the factor road noise and the factor noise sensitivity needs to be clarified. It was assumed that since most participants were not accustomed to rail noise, the annoyance ratings for dominant rail noise were higher than that for traffic noise. However, this explanation was not convincing since similar effects were expected from individuals with low noise sensitivity. The assumption that the temporal structure of the traffic noises brought about the interactions can be accepted since the annoyance ratings of noise sensitive individuals were determined due to the temporal structure of passing vehicles. [22] Further, the noise of vehicles passing on the road can be characterized as constant due to a constant traffic flow, whereas the rail noise is presented in an intermittent pattern and can be more and more perceivable if the level of rail noise increases. If these considerations are proven, then they can be conceived as another indicator for the validity of NoiSeQ.

The results of the experiments concerning the validity of the NoiSeQ show that noise sensitivity influence annoyance. Further, the data demonstrate that noise sensitivity also has an effect on the sound level-related changes of annoyance. However, further studies are required to determine the extent to which the noise sensitivity measured for different daily situations affects annoyance. In addition, the effects of different temporal structures of traffic noise on annoyance should be analyzed to get complete information on the combined effects of noise characteristics and noise sensitivity on annoyance. Such information will help to understand the concept of noise sensitivity more precisely.


1Ellermeier W, Eigenstetter M, Zimmer K. Psychoacustic correlates of individual noise sensitivity. J Acoust Soc Am 2001;109:1464-73.
2Zimmer K, Ellermeier W. Eine deutsche Version der Lδrmempfindlichkeitsskala von Weinstein. Zeitschrift für Lδrmbekδmpfung 1997;44:107-10.
3Zimmer K, Ellermeier W. Konstruktion und Evaluation eines Fragebogens zur Erfassung der individuellen Lδrmempfindlichkeit. Diagnostica 1998;44:11-20.
4Zimmer K, Ellermeier W. Ein Kurzfragebogen zur Erfassung der Lδrmempfindlichkeit. Umweltpsychologie 1998;2:54-63.
5Brennan RL. Generalizability Theory. New York: Springer; 2001.
6Cronbach LJ, Gleser GC, Nanda H, Rajaratnam N. The dependability of behavioural measurements: Theory of generalizability for scores and profiles. John Wiley and Sons: New York; 1972.
7Shavelson RJ, Webb NM. Generalizability theory - A primer. Newbury Park: Sage Publications; 1991.
8Gleser GC, Cronbach LJ, Rajaratnam N. Generalizability of scores influenced by multiple sources of variance. Psychometrika 1965;30:395-418.
9Cornfield J, Tukey JW. Average values of mean squares in factorials. Ann Mat Stat 1956;27:907-49.
10Job RF. Noise sensitivity as a factor influencing human reaction to noise. Noise Health 1999;3:57-68.
11Stansfield SA. Noise, noise sensitivity and psychiatric disorder. Psychological Medicine: Monographs, Supplement 22. 1992.
12Kjellberg A, Landstr φm U, Tesarz M, S φderberg L, Akerlund E. The effects of non physical noise characteristics, ongoing task and noise sensitivity on annoyance and distraction due to noise at work. J Environ Psychol 1996;16:123-36.
13Van Kamp I, Soames RF, Hatfield J, Haines M, Stellato RK, Stansfeld S. The role of noise sensitivity in the noise-response relation: A comparison of three international airport studies. J Acoust Soc Am 2004;116:3471-9.
14Miedema HM, Vos H. Noise sensitivity reactions to noise and other environmental conditions. J Acoust Soc Am 2003;113:1492-504.
15ISO/TS 15666. Acoustics -Assessment of noise annoyance by means of social and socio-acoustic surveys. Genf: International Organization for Standardization;2003.
16Heller O. H φrfeldaudiometrie mit dem Verfahren der Kategorienunterteilung (KU). Psychologische Beitrδge 1985;27:478-93.
17ITU-R BS.1116-1. Methods for the subjective assessment of small impairments in audio systems including multi channel sound system. 1994-1997.
18Berglund B, Preis A, Rankin K. Relationship between loudness and annoyance for ten community sounds. Environ Int 1990;16:523-31.
19Shingledecker CA. A task battery for applied human performance assessment research, AFAMRL-TR-84-071. Ohio Wright-Patterson Air Force Base, Air Force Aerospace Medical Research Laboratory; 1984.
20ISO 10075-3. Ergonomic principles related to mental workload - Part 3: Measurement and assessment of mental workload. Genf: International Organization for Standardization; 2004.
21Guski R. Personal and social variables as co-determinants of noise annoyance. Noise Health 1999;3:45-56.
22Roberts MJ, Western AW, Webber MJ. A theory of pattern of pass by noise. J Sound Vibration 2003;262:1047-56.