Participants from nine neighborhoods in the city of Ghent (Belgium) were selected, including inner city zones and suburban areas. The choice of these neighborhoods was based on specific requirements regarding the most and least exposed façade of dwellings that will be explained in the next section. Participants were directly contacted, without prior announcement, by knocking on doors at the selected sites. Interested people who had no time at the moment of the visit were offered a second visit. The minimum age for respondents was 18 years.

A face-to-face questionnaire was performed and 100 participants were aimed at. There was a single interviewer. The survey was announced as general research concerning the living environment, and how it was affected by the architecture of their dwelling. It was stressed that people satisfied with their living environment should participate too. An important prerequisite to start the questionnaire was that participants had lived at least 1 year at the current location, in order to prevent people making a comparison relative to their previous dwelling (to prevent so-called “change effects” in addition to the “exposure effect”, see e.g., Reference [18]), and to ensure that participants were sufficiently familiar with the local noise climate.

The noise exposure at the most exposed façade was extracted from the road traffic noise map approved by the Flemish regional government for the agglomeration of Ghent, which has been reported to the European Commission in the framework of the Environmental Noise Directive (END) [19]. Locations dominated by road traffic were selected, where the most exposed façade fell in the classes 65–75 dBA for day-evening-night corrected (yearly) equivalent sound pressure levels (= L_den). About half of the neighborhoods were selected with dwellings having a clearly shielded façade (as illustrated in Figure 1), the other half with no quiet side at all (see Figure 2). The presence of a quiet side was assessed based on the noise map, in combination with analyzing building geometry and their surroundings using aerial photographs, and by site visits. This selection procedure assured that the quiet side was accessible (windows, balcony, etc.) by the inhabitants of the dwelling and that there were no disturbances that are generally not included in END noise maps such as parking lots and ventilation units.

Road traffic noise levels at the most and least exposed façade were manually selected from the noise maps at the survey locations and include a quality check of the exposure data. Since noise levels at the most exposed façade in streets are strongly source driven, their correctness is generally considered to be reasonable; only very strong deviations from the actual traffic intensity or composition would lead to significant errors in the predicted levels. For less trafficked roads, such predictions in noise maps are less reliable. However, in the current study, only façades along (very) busy roads have been selected to reach the 65–75 dBA L_den exposure range. At shielded façades, more complex sound propagation aspects need to be taken into account when calculating noise maps. These are unfortunately often neglected to limit the computational cost. Typically, this leads to an underprediction of levels at the least-exposed façade. Therefore, only a rough classification regarding level difference between the loudest and quietest side will be used during analysis.

In current urban noise mapping research, engineering models are under development to account for these inaccurate predictions at shielded locations. An interesting concept called “background noise mapping” is discussed in Reference [20].

In a first part, some general questions were posed regarding the quality of living in the neighborhood, and annoyances caused by common environmental stressors. The first question looked at the general satisfaction regarding the quality of living in the neighborhood of the dweller, with indication of some examples of parameters to be taken into account (e.g., safety, child-friendly, environment, ...). A 5-point categorical scale was offered, with the following textual descriptions: “very satisfied”, “satisfied”, “more or less satisfied”, “not satisfied”, and “not at all satisfied”. Next, it was asked if the respondent would advise friends or relatives to come live in their neighborhood when considering the quality of the living environment. A 3-point categorical scale was offered (“yes”, “no”, “undecided”). Then, the ISO-standardized question [21] was asked regarding annoyance by noise, odor and light (“If you consider the past 12 months, to what degree were you annoyed or not annoyed by the following sources”). A 5-point categorical scale (“not at all annoyed”, “slightly annoyed”, “moderately annoyed”, “strongly annoyed”, and “extremely annoyed”) was proposed for each of these types of environmental nuisances.

In a second part, more detailed information about possible sources of noise annoyance was sought. These sources are traffic (“road”, “railway”, “air”, “water”), industry and small and medium enterprises (“delivery of goods by trucks”, “construction noise”, “industrial plants”, “trade and services”), leisure activities (“music from pubs, bars, and restaurants”, “music in cars”, “outdoor concerts”, “sports activities”, “street racing”), agricultural activities, neighbors (“children playing”, “animals”, “do-it-yourself home improvement”, “loud music/television”, “gardening”, “heating, ventilation and air conditioning units”), others. For each defined subcategory of the different sources of potential noise annoyance, the same scale was used as for the general noise annoyance question as discussed in previous paragraph.

In a third part, specific questions were asked regarding possible sleep disturbance by noise. Firstly, the interviewer made, with help of the respondent, a plan of the dwelling indicating the location of the different rooms. The interviewer assessed whether the bedroom was located at a quiet side of the dwelling (without further interaction with the dweller). In this way, no hints on the link between bedroom location and noise exposure were given. The number of hours of sleep per night was asked for. It was asked whether the respondent is hindered by road traffic noise during falling asleep, and if road traffic noise leads to awakening at night. A 4-point categorical answering scale (“never”, “sometimes”, “often”, “always”) was used. Specific questions were asked regarding opening or closing of the window of the bedroom: “Do you leave your bedroom window open (“always”, “never”, “only during summertime”)?”, “Do you like sleeping with open windows (“yes”, “no”)?”, “Do you close your bedroom window because of road traffic noise (“never”, “sometimes”, “often”, “always”)?”. Closing windows was shown to be an important way of coping before, influencing annoyance ratings [22].

In a fourth part, noise sensitivity was assessed using a Dutch adaptation of Weinstein’s noise-sensitivity scale [23], used previously in large-scale Flemish quality-of-life studies. This part contained 10 questions, and used a 6-point categorical scale with textual indication of the endpoints (“totally agree” and “totally disagree”). Finally, some general questions were asked about gender, age, education, professional activities, and the number of years living at the current location.

Given the limited amount of respondents in the dataset, classification has been performed in order to have sufficient occurrences in the different cells when using frequency tables. The Chi-square test has been applied to check dependence between variables. The null hypothesis states that variables are independent. In order to have a clear dependence, this null hypothesis must be rejected with a sufficient degree of certitude (= 1 − p). Odds Ratios (OR) have been considered, and logistic regression is used to predict confidence intervals. Logistic regression with a dichotomous outcome (true or false) has been used, based on both continuous and dichotomous independent variables. In order to be statistically sound, 95% confidence intervals of OR should not contain 1. Effect modifiers have been looked for by multiple logistic regression. Statistical significance of model deviance reduction when including additional variables has been checked by likelihood ratio testing (based on the Chi-square distribution). The Matlab^® statistics toolbox has been used to perform all analyses.

The total number of respondents was 100. The willingness to participate was high and equaled 70%. From these, 4% was discarded since these inhabitants had been living less than one year at their current location. The distribution over gender, age, years living at the location, education, employment and noise sensitivity is shown in Figure 3. 52% of the respondents are more than 50 years old. 53% of the respondents were female. 52% of the respondents received a higher education (combining “non-university continued education” and “university”). The median of the noise sensitivity, after linearly averaging the responses on the 10 sensitivity questions, was at 3.9 (with 1 meaning “not at all sensitive to noise”, and 6 meaning “highly sensitive to noise”). No rectification to match the Flemish population has been performed on the data. An overview of the respondent characteristics, organized in front-back level difference intervals, is shown in Table 1.

The distribution of the front façade levels (near the street side) and back façade levels are shown in Figure 4 and Table 1. In 12% of the cases, the back façade noise level exceeds the front façade level. The distribution considering the most-exposed and least exposed façade is very similar (not shown). For 69% of the survey points, the most exposed façade falls between 65 and 75 dBA, which was the aim during site selection. Survey points outside this interval arise from the specific orientation and geometry of individual dwellings where someone was willing to participate. 38% of the respondents have a front-back level difference (absolute value) less than 10 dBA. 28% of the respondents have a level difference larger than or equal to 20 dBA.

The linear (Pearson’s) correlation coefficient between front level L_den and level difference equals 0.48 (95% CI = 0.31–0.62, p = 4E-7). This positive correlation is not surprising. Since minimum levels in the urban structure most often do not drop below a certain value, large level differences could only occur when the front façade level is high. The correlation coefficient found here is rather small, since the most exposed façade is fixed in a rather limited level interval during survey point selection, while locations are specifically searched for with a large and small level difference between front and back.

An overview of the answers to questions on sleep disturbance, and window use and habits, is shown in Figure 6 and Table 1.

It was objectively observed by the interviewer that only 20% of the respondents have their bedroom window at a quiet façade. Persons having their bedroom at a quiet façade have 7.4 times (crude OR, 95% CI = 1.6–33.9) more chance of not being (at least moderately) annoyed by noise compared to the 80% of the respondents having their bedroom at an exposed side. This dependence is highly statistically significant (χ² = 8.3, p = 0.004). The predicted chance of being at least moderately annoyed by noise (using model 1, see Table 2) reduces from 45% (bedroom not at a quiet side) to 10% when the bedroom is located at a quiet side. This indirect effect of bedroom location on noise annoyance rating seems much stronger than the direct effect of level difference Q as quantified in previous section.

Including noise sensitivity and front-façade L_den level (see Table 2) in the logistic regression model significantly reduces model deviation (at the 5% significance level). The adjusted OR for bedroom location increases to 10.6 (with 95% CI = 2.0–56.0). It can therefore be concluded that the location of the sleeping room at a quiet side has an important (indirect) positive effect on the (general) noise annoyance experienced at the dwelling. This indirect quiet side mechanism seems especially relevant for noise sensitive persons. Among the highly and extremely annoyed people, no one in the current sample had the bedroom window facing the quiet side.

61% of the respondents close their bedroom window during sleeping. The answers “always open”, and “only during summer period” have been grouped to the same category in the following analysis. The fact that 80% of the respondents have their bedroom near the loudest side already suggests a logical link. The people closing their window are not happy with this; near one half (46%) of them would actually prefer to sleep with open windows. The question whether they close their bedroom window because of road traffic noise was answered affirmative (closing “sometimes”, “often”, or “always” the window) by 74% of the people. Preference with regards to opening or closing window, and closing because of road traffic noise, tend to be dependent (χ² = 1.88, p = 0.17). People closing their windows because of noise have an OR of 2.3 (95% CI = 0.7–7.7) that they do this against their actual preference as for closing windows. This is however not statistically significant at the 95% certainty level. Only 8% of the people closing their window, and doing this against their will, seem to have another reason than road traffic noise. Noise sensitive persons close their bedroom window more often because of road traffic noise. Noise sensitivity classified as a dichotomous variable (below or above 3.5) and closing bedroom window because of noise (closing “sometimes”, “often” or “always” versus “never”) yields a strong dependency (χ² = 6.38, p = 0.01; OR = 3.1 with 95% CI = 1.3–7.7).

Persons whose sleeping room is located at the quiet side of a dwelling, have a chance that is 19 times (95% CI = 2.4–148.8) as low to close their window because of street noise than people with a bedroom at a loud side. This dependence is highly statistically significant (χ² = 13.39, p = 2.5E-4). People sleeping with open window do this fully following their preference (OR = 21.8 with 95% CI = 4.8–98.6). This is in contrast with people closing bedroom windows, which is often against their wishes. The latter can be seen as a way of active coping with excessive noise levels. People create their own quiet side when the rooms are not well orientated. This analysis is however somewhat biased, since people having their sleeping room at a quiet side have a somewhat stronger preference to sleep with open window, but independency between these variables cannot be statistically significantly rejected at the 5%-significance level (χ² = 2.47, p = 0.12). A possible reason is that while choosing bedroom location after moving or during designing a newly built house, this preference could have been taken into account. Furthermore, bedroom location at a quiet side and noise sensitivity are moderately dependent (χ² = 2.68, p = 0.10): noise sensitive people seem to have taken noise into account when making such decisions.

38% of the people with a bedroom at a loud side (both closing and not-closing windows) wake up at least sometimes because of street noise. This percentage reduces to 25% for people with a bedroom at a quiet side. However, awakening (“sometimes”, “often” or “always”) at night by road traffic noise does not seem to be strongly dependent on the location of the bedroom at a quiet side (χ² = 1.10, p = 0.29; OR = 1.8 with 95% CI = 0.6–5.5). Noise sensitivity (continuous scale) is here the variable with most predictive power (see Table 2) among the ones considered, at an OR of 1.5 (with 95% CI = 1.0–2.4). Classifying noise sensitivity to a dichotomous variable (below or above a value of 3.5) shows some dependency (χ² = 2.65, p = 0.11) with self-reported awakening.

Difficulties falling asleep because of noise and the presence of the bedroom at the quiet side are somewhat more significantly dependent (independence can be rejected with 93% certainty following the chi-square test, χ² = 3.18, p = 0.07). A sleeping room at the quiet side gives an OR equal to 5.5 (95% CI = 0.7–44.1) in preventing to have (at least sometimes) difficulties falling asleep because of noise. The predicted chance of having difficulties falling asleep at least sometimes by noise reduces from 23% (bedroom not at a quiet side) to 5% when the bedroom is located at a quiet side. Noise sensitivity is not an important parameter for the latter, and hardly improves model accuracy when adding this parameter as a second independent variable in the logistic regression.

Awakening at night by noise and having problems falling asleep because of noise are clearly linked (χ² = 19.9, p < 1E-5; OR = 11.4 with 95% CI = 3.4–38.5). Awakening by street noise strongly influences the (general) noise annoyance experienced. Waking up at least sometimes by noise, and being at least moderately annoyed, both classified as a dichotomous variable, are strongly dependent (χ² = 8.38, p = 0.004; OR = 3.5 with 95% CI = 1.5–8.2). The link between difficulties falling asleep because of street noise and general noise annoyance is not at all present.