Community Annoyance Due to Settleable Dust: Influential Factors in Air Pollution Perception

Abstract

Subjective assessments of air pollution annoyance reveal that individuals’ focus on specific risks is influenced by their attachment to place, beliefs, values, and behavior rather than the composition or toxic effects of air pollutants. Additionally, the social context plays a role in shaping how communities react to and perceive air pollution impacts. This study examines residents’ environmental perceptions regarding the effects of settleable particles before and after the interruption of a large industrial source in the southern region of Espírito Santo, Brazil (South America). A second objective was to model the relationship between air pollution annoyance and other perceived variables under both scenarios. Data were collected through surveys conducted before and after the interruption of the industrial plant. The Pearson chi-square test and ordinal logistic regression model analyzed the data. Results indicate a shift in residents’ concerns with a focus on social and well-being issues. We also found a small number of items relating to dust annoyance and home ownership that can be used to predict the air pollution impact for individual community members. The findings show that settleable particles are directly perceived by exposed populations and significantly affect community health and quality of life.

1. Introduction

Heavy industries located near residential areas can cause annoyance to residents due to air pollution, which is perceived through impaired visibility, unpleasant odors, or dust [1,2,3], for instance. Although this may or may not cause direct health problems, it significantly affects quality of life [4,5]. The World Health Organization (WHO) defines health as a “state of complete physical, mental, and social well-being, and not simply the absence of disease or illness” [6].

Therefore, annoyance caused by air pollution should be considered a public health issue.

Dust consists of particulate matter (PM) with a broad particle size range from a few nanometers to several micrometers [7]. It is generated through natural processes (such as soil erosion, vegetation, and sea spray) and human activities (including construction sites, unpaved roads, combustion processes, and various industrial operations). Typically, particles larger than 10 µm settle close to the source, while smaller particles remain suspended and can travel long distances. Settleable particulate matter (SPM), of which particles larger than PM₁₀ are the bulk constituent in mass, is inconvenient to public welfare [8,9]. SPM on the ground or surfaces causes soiling in residential and urban environments, leading to material degradation or the need for constant cleaning. This can impair the use and enjoyment of property and disrupt normal societal activities [8,10,11].

Many studies have shown a significant relationship between the PM concentration or deposition rate and annoyance levels. Research conducted in several European cities has identified PM_2.5 and NO₂ as important determinants of perceived air pollution annoyance [12]. Evidence from northern Europe has also shown exposure–response relationships between vehicular pollution and annoyance using regression-based approaches [13]. Other authors have focused on perceptual and visual dimensions of air pollution. Associations between air quality perception and visual impairment caused by suspended particles have been reported [14], and the presence of visible industrial plumes has been shown to influence well-being and emotional responses [15]. In addition, the relevance of settleable particle deposition rates has been discussed in the context of developing annoyance-related air quality guidelines [11]. Studies have also investigated broader psychosocial outcomes. Links between particulate matter exposure and subjective well-being indicators, such as life satisfaction and happiness, have been identified [2,16]. Multivariate analyses have further indicated that air pollution-related annoyance is associated with qualitative variables, including the perceived importance of air quality, perceived exposure to industrial risk, evaluation of air quality, and perceived air pollution [17]. Exposure response analyses based on questionnaire data and measured concentrations of total suspended particles (TSP and PM₁₀) have shown that higher PM levels are associated with an increased probability of annoyance [8]. Despite these contributions, most investigations have been conducted under relatively stable emission scenarios, resulting in limited evidence on how abrupt interruptions of industrial activities affect both objective air quality indicators and community perceptions, particularly when social and economic dimensions are involved. An intervention study conducted near a sintering plant on the east coast of Sweden evaluated community perceptions regarding the effectiveness of emission reduction measures, changes in dust levels, risk perception, and reported health symptoms using questionnaire-based assessments [18]. Their analysis, conducted using path analysis techniques, indicated that the intervention was successful as planned. After the sintering plant was closed, the environment was perceived as being less dusty, the residents were more positive in their risk perception, and they reported less annoyance due to dust, soot and odorous substances. Nonetheless, their study did not evaluate whether the community’s perception of reduced dust and improvement in health symptoms was affected by the social and economic impacts on the community resulting from the industrial intervention. This study examines a distinctive scenario involving an environmental disaster that led to a five-year suspension of a pelletizing industry’s operations in a city on the southern coast of Brazil. It aims to assess the community’s perceptions of settleable particles reduction resulting from the operational shutdown and its effects on risk perception and routine habits as well as social, emotional, and economic conditions. A second objective is to identify the relationship between air pollution annoyance and other significant perceived variables, considering both research scenarios. This study was only possible because a few years before the disaster, a survey had been conducted in the study area surrounding the pelletizing plant, collecting data on public annoyance caused by settleable particles [19].

2. Materials and Methods

2.1. Study Region

The cities of Anchieta and Guarapari in southern Brazil form an urban and industrialized area that includes a natural gas production plant, a pelletizing industry, and a port for exporting goods (Figure 1). The study region has two distinct seasons: the first is from September to March (Summer) characterized by high precipitation and north/northeast wind direction, and the second is from April to August (winter) with prevailing lower precipitation and southern wind direction.

Although the area is also a tourist destination with many wild beaches, tourism plays a less significant role in the regional economy. According to the Brazilian Institute of Geography and Statistics (IBGE), the region had a population of 16,362 inhabitants as of the 2022 Demographic Census. The industrialization of this region has negatively impacted the environment, directly influencing the quality of life and causing health issues [20]. The authors interviewed workers at the pelletizing plant with the aim of identifying the perceived risk factors and workload. The processes with air pollution emissions most frequently reported by workers were the movement of pellets to the conveyor belts and to the yard, storage in piles, pellet recovery and transport to the port. Exposure to heat, noise, sounds and air pollution in the form of particles, soot and toxic gases resulting from some processes can cause physiological consequences, such as respiratory, skin and eye problems, nasal obstruction, allergies, asthma attacks, coughing, and chest pain. According to the authors, these negative charges and risks, which turn into physiological symptoms, are confirmed in data from the Espírito Santo State Health Department on deaths resulting from neoplasms.

In 2012, a survey conducted in this region indicated that about 80% of the population reported annoyance caused by air pollution (with the majority identifying settleable particles as the main form of air pollution in the region) [19]. In that study, respondents reported having to keep windows closed to keep out dust from outside, frequently going to the doctor because of respiratory problems, and not having knowledge about air quality monitoring in the region. The authors presented an empirical correlation model in which a significant relationship was found between the discomfort caused by dust and gender, the occurrence of respiratory problems, risk perception and assessment, and the importance of air quality.

In November 2015, a major tailings dam rupture occurred at a mining plant in the city of Mariana, approximately 400 km away from Anchieta and Guarapari, which was reported as the largest environmental disaster in Brazil’s history imposing an unprecedented risk to biodiversity [21,22]. This mining plant produced the iron ore used by the pelletizing company in Anchieta. Consequently, following the disaster in Mariana, the pelletizing industry in Anchieta halted pellet production, and the port used for exporting these products became inactive. This series of events significantly impacted the employment and economic conditions of both Anchieta and Guarapari as well as considerably decreased the air pollution in the region.

2.2. Data Sets

Two structured face-to-face surveys were conducted in urban regions in March 2014 and March 2017, respectively, prior to and after the interruption of the pelletizing plant operation. Figure 2 presents the wind rose and monthly precipitation graphs for the survey periods: the wind rose for (a) March 2014 and (b) March 2017, and the monthly precipitation graphs for (c) 2014 and (d) 2017 during the years in which the survey was carried out.

The wind roses shown in Figure 2 were generated using meteorological data from a station operated by the Brazilian National Institute of Meteorology (INMET) located at 20.27° S latitude and 40.31° W longitude. This station is situated in a nearby municipality and represents the closest location with continuous data available for the selected periods in a region with meteorological characteristics similar to the study area.

The surveys were conducted using the same questionnaire previously applied in a study carried out in the study region [19]. The same questionnaire was also used for other studies conducted in different regions [11,17]. The questionnaire comprised closed-ended questions with response options as a Likert scale including both (categorical and numerical scales). For example, one question asked: “How annoyed do you feel by dust from outside?”. The response options were: 1—Not annoyed, 2—Slightly annoyed, 3—Moderate annoyed, 4—Very annoyed, and 5—Extremely annoyed.

It is important to explain that all variables analyzed in this study represent the following questionnaire constructs: 1— Assessment and monitoring of air quality; 2—Discomfort caused by pollution and dust; 3—Impacts and losses caused in daily life and health; 4—Identification of sources of air and dust pollution; 5—Sociodemographic aspects.

The residents were selected randomly in the sub-regions located in a radius of 1.5 km around six air quality monitoring stations operated by the local environmental protection agency: Anchieta-sede (E-1), Belo Horizonte (E-2), Guanabara (E-3), Maemba (E-4), Meaipe (E-5), and Ubu (E-6) as shown in Figure 1. A simple stratified random sampling with proportional allocation was used [23]. The sample sizes satisfy the following equations:

$$n\ge N{\left[1+{\displaystyle \frac{(N-1)}{p(1-p)}}{\left({\displaystyle \frac{d}{Z_{\alpha }}}\right)}^2\right]}^{-1}$$

(1)

and

$$n_i=\left({\displaystyle \frac{N_i}{N}}\right)n,i=1,\dots ,6.$$

(2)

In (1) and (2), n is the sample size, N is the total population size, $1-\alpha$ is the confidence level, $Z_{\alpha }$ is the $(1-\alpha /2)$-th percentile of the probability distribution of the data, p is the population proportion, d is the margin of error equal to 0.05, $N_i$ is the population size in the i-th sub-region, and $n_i$ is the sample size in the i-th sub-region. Due to the population’s known variability (in 2012 survey) $p=0.8$ was considered along with an acceptable margin of error of $d=0.05$. Thus, the minimum sample size calculated was 130 respondents. However, we decided to be conservative and considered all respondents who agreed to participate (answering the questionnaire), as shown in

Table 1. Population of sub-regions and surveys.

Sub-Region	Habitants	Survey 2014	Survey 2017
Meaípe (E5)	2.750	43	93
Anchieta sede (E1)	6.762	107	229
Guanabara (E3)	1.805	28	61
Maembá (E4)	2.045	32	69
Ubu (E6)	1.904	30	64
Belo Horizonte (E2)	1.906	17	37
Total	16.362	258	553

.

The same 258 respondents from 2014 were selected for the 2017 survey with the addition of new respondents (after the industry shutdown, many residents who participated in the 2014 survey lost their jobs and moved away from the region). Also, further questions about the industry closure were included in the 2017 survey. Table 1 displays the population and sample sizes (considering the proportional differences) for both surveys.

2.3. Pearson’s Chi-Square Test

Pearson’s chi-square homogeneity tests [24] was implemented to compare the responses from the two surveys and verify whether the differences are statistically significant at a the significance level of 5%. The two hypotheses of the homogeneity test are outlined below:

H₀. 

The distributions of responses in the 2014 and 2017 surveys are the same.

H₁. 

The distributions of responses in the 2014 and 2017 surveys are not the same.

The statistic for testing H₀ is

$${\mathcal{X}}^2=\sum _{i=1}^l\sum _{j=1}^r{\displaystyle \frac{{(n_{ij}-{\mu }_{ij})}^2}{{\mu }_{ij}}},$$

(3)

where $n_{ij}$ is the number of respondents (observed value) from each class $ij$ of the contingency table in which the rows correspond to the possible configurations of the covariates, and the columns are the categories of the ordinal responses. The expected number ${\mu }_{ij}$ for the class $ij$ is the product of $n=\sum n_{ij}$ by the probability of occurrence of class $ij$. Lastly, l and r are the numbers of rows and columns of the contingency table. Under H₀, the distribution of ${\mathcal{X}}^2$ is approximated by a chi-squared distribution with $(l-1)(r-1)$ degrees of freedom, and H₀ is rejected when the observed value of ${\mathcal{X}}^2$ is larger than the $(1-\alpha )$th quantile of this distribution with $\alpha =0.05$.

Subsequently, Pearson’s chi-square independence test was carried out in order to verify which variables were significant for inclusion in the regression model. That is, from the Pearson chi-square independence test results, only the categories of significant variables will be selected for inclusion in the regression model.

2.4. Ordinal Logistic Regression

The ordinal logistic regression (OLR) model is an essential statistical tool widely used to measure and quantify the relationship between an ordered response variable and one or a set of explanatory variables; see, e.g., [24,25,26,27].

Here, the response variable is the degree of annoyance Y with five categories $\left(c=5\right)$: extremely annoyed, very annoyed, moderately annoyed, little annoyed, not annoyed. The covariates ${\mathit{X}}^{⊺}=(X_1,\dots ,X_k)$ are presented in Section 3.2, and ${\mathit{X}}^{⊺}$ is the transpose of $\mathit{X}$. The conditional probabilities of Y [24] are modelized by

$${\pi }_j=\mathcal{P}(Y=j\mid \mathit{X})={\displaystyle \frac{\exp \left({\alpha }_j+{\varphi }_j{\mathit{\beta }}^{⊺}\mathit{X}\right)}{{\sum }_{i=1}^{c}exp\left({\alpha }_i+{\varphi }_i{\mathit{\beta }}^{⊺}\mathit{X}\right)}},j=1,\dots ,c,$$

(4)

where ${\alpha }_c=0$ and ${\varphi }_c=0$ are introduced to simplify the notation. Parameter ${\alpha }_j$ is the intercept related to the $j^{th}$ category of Y, while ${\mathit{\beta }}^{⊺}=({\beta }_1,\dots ,{\beta }_k)$ is the vector of coefficients independent of the categories of Y and corresponds to the effects of the covariates on the response variable. It follows from (4) that

$$log{\displaystyle \frac{{\pi }_j}{{\pi }_c}}={\alpha }_j+{\varphi }_j{\mathit{\beta }}^{⊺}\mathit{X},j=1,\dots ,c-1.$$

(5)

Therefore, for $r=1,\dots ,k$, ${\varphi }_j{\beta }_r$ is the coefficient of the explanatory variable $X_r$ in the log odds ratio (OR) for categories j and c of Y. The parameters to estimate in (4) are ${\alpha }_j,{\varphi }_j$ for $j=1,\dots ,c-1$ and $\mathit{\beta }$. To avoid identifiability difficulties, we set ${\varphi }_1=1$. The parameters are estimated by maximizing the conditional likelihood; see [28]. SPSS (version 20.0, IBM Corp., Armonk, NY, USA) and R (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria) software were used to perform the statistical data analysis.

3. Results

3.1. Comparing Before and After the Interruption in the Industrial Activities

Figure 3 shows the SPM time series measured at the air quality monitoring stations. It is clear that SPM decreased from January 2016 onwards at all monitoring stations when the full stop of the pelletizing plant operation occurred. Before January 2016, several SPM peak values larger than 10 g/m²/30 days were observed. In most countries that have an air quality standard for SPM, the limits range from 5 to 10 g/m² per 30 days [11].

Table 2. Frequency and percentage data on demographics for the 2014 and 2017 surveys (NA/NK = not answered/not know).

	2014	2017	p-Value
Gender			0.37
Male	124 (48.06%)	285 (51.54%)
Female	134 (51.94%)	268 (48.46%)
Age (years)			0.00
16–24	34 (13.18%)	77 (13.92%)
25–34	66 (25.58%)	95 (17.18%)
35–54	103 (39.92%)	210 (37.97%)
55+	47 (18.22%)	169 (30.56%)
NA/NK	8 (3.10%)	2 (0.36%)
Levels of education			0.36
Primary school	113 (43.8%)	234 (42.31%)
High school	108 (41.9%)	220 (39.78%)
University	36 (14.0%)	90 (16.27%)
NA/NK	1 (0.4%)	9 (1.63%)
Occupation			0.00
Employed	155 (60.08%)	124 (22.42%)
Unemployed	35 (13.57%)	140 (25.32%)
Retired	20 (7.75%)	94 (17.00%)
Student	12 (4.65%)	40 (7.23%)
Freelancer	29 (11.24%)	146 (26.40%)
NA/NK	7 (2.71%)	9 (1.63%)

shows the demographic characteristics of the respondents for both surveys. According to Pearson chi-square analyses (p-value), there is no significant difference between gender and level of education distributions in both samples. However, the age range and occupational situation distribution were significantly different. Regarding the age extracts, in any case, most of the respondents are over 35 years old. As for occupation, the population shifted from employed in 2014 to unemployed, retired and freelancer.

Table 3. Questions and response absolute frequencies (%) pertaining to annoyance for the 2014 and 2017 surveys (NK/NA: not known/not answered).

	2014	2017	p-Value
How annoyed do you feel by air
pollution?			<0.05
Not annoyed	6 (2.33%)	209 (37.79%)
A little	28 (10.85%)	123 (22.24%)
Moderate	68 (26.36%)	109 (19.71%)
Very	111 (43.02%)	87 (15.73%)
Extremely	45 (17.44%)	25 (4.52%)
At home, how annoyed do you feel
by dust from outside?			<0.05
Not annoyed	6 (2.33%)	229 (41.41%)
A little	75 (29.07%)	70 (12.66%)
Moderate	49 (18.99%)	93 (16.82%)
Very	58 (22.48%)	139 (25.14%)
Extremely	69 (26.74%)	20 (3.62%)
NK/NA	1 (0.39%)	2 (0.36%)
In your opinion, where does this
dust come from?			<0.05
Vehicles (cars)	34 (13.18%)	102 (18.44%)
Industrial sources	205 (79.46%)	249 (45.03%)
Unpaved streets	9 (3.49%)	92 (16.64%)
Construction work	17 (6.00%)	94 (18.00%)
Sea breeze	2 (0.78%)	21 (3.80%)
Quarry exploration	0 (0.00%)	15 (2.71%)
NK/NA	4 (1.55%)	59 (10.67%)

,

Table 4. Questions and response absolute frequencies (%) pertaining to assess air quality and industrial risk perception for 2014 and 2017 surveys (NK/NA: not known/not answered).

	2014	2017	p-Value
How do you assess air quality in your
neighborhood?			<0.05
(perception of air quality)
Terrible	33 (12.79%)	15 (2.71%)
Bad	25 (9.69%)	21 (3.80%)
Regular	150 (58.14%)	200 (36.17%)
Good	47 (18.22%)	255 (46.11%)
Excellent	3 (1.16%)	58 (10.49%)
NK/NA	0 (0.00%)	4 (0.72%)
How do you feel about industrial risk?			<0.05
(perception of industrial risk)
Nothing exposed	16 (6.20%)	193 (34.90%)
A little	36 (13.95%)	127 (22.97%)
Moderate	69 (26.74%)	111 (20.07%)
Very	92 (35.66%)	76 (13.74%)
Extremely exposed	41 (15.89%)	26 (4.70%)
NK/NA	4 (1.55%)	20 (3.62%)
How important is the quality
of the air to you?			<0.05
(air quality importance)
Extremely	144 (55.81%)	186 (33.60%)
Very important	99 (38.37%)	296 (53.50%)
Moderate importance	9 (3.49%)	56 (10.10%)
Slightly important	3 (1.16%)	4 (0.70%)
Not important at all	0 (0.00%)	7 (1.30%)
NK/NA	3 (1.16%)	4 (0.70%)
Do you think that air quality is monitored
in your neighborhood/region?			<0.05
(perception of air quality monitoring)
Never	152 (58.91%)	176 (31.83%)
Rarely	18 (6.98%)	81 (14.65%)
Sometimes	23 (8.91%)	70 (12.66%)
Frequently	3 (1.16%)	42 (7.59%)
Always	5 (1.94%)	41 (7.41%)
NK/NA	57 (22.09%)	143 (25.86%)

,

Table 5. Questions and response absolute frequencies (%) pertaining to perceived air pollution forms for the 2014 and 2017 surveys (NK/NA: not known/not answered).

	2014	2017	p-Value
Do you perceived air pollution through
dust, particles, flakes, etc.?			<0.05
(perceived dust)
Never	1 (0.39%)	47 (8.50%)
Rarely	15 (5.81%)	97 (17.54%)
Sometimes	57 (22.09%)	142 (25.68%)
Frequently	105 (40.70%)	149 (26.94%)
Always	80 (31.01%)	112 (20.25%)
NK/NA	0 (0.00%)	6 (10.8%)
Do you perceive air pollution through
odour/bad smell?			0.13
(perceived odour)
Never	102 (39.53%)	211 (38.16%)
Rarely	76 (29.46%)	114 (20.61%)
Sometimes	50 (19.38%)	139 (25.14%)
Frequently	12 (4.65%)	54 (9.76%)
Always	12 (4.65%)	32 (5.79%)
NK/NA	6 (2.33%)	3 (0.54%)
Do you perceive air pollution due to
air opacity/smoke?			<0.05
(perceived opacity)
Never	37 (14.34%)	273 (49.37%)
Rarely	93 (36.05%)	98 (17.72%)
Sometimes	75 (29.07%)	120 (21.70%)
Frequently	31 (12.02%)	40 (7.23%)
Always	21 (8.14%)	19 (3.44%)
NK/NA	1 (0.39%)	3 (0.54%)

and

Table 6. Questions and response absolute frequencies (%) pertaining to consequences of air pollution in daily life in the 2014 and 2017 surveys (NK/NA: not known/not answered).

	2014	2017	p-Value
How often do you clean your house to
remove the dust caused by air pollution?			<0.05
(cleaning the house)
Never	0 (0.00%)	17 (3.07%)
Rarely	2 (0.78%)	46 (8.32%)
Sometimes	25 (9.69%)	65 (11.75%)
Frequently	90 (34.88%)	183 (33.09%)
Always	140 (54.26%)	239 (43.22%)
NK/NA	1 (0.39%)	3 (0.54%)
Do you close the windows to
prevent dust from outside?			<0.05
(closing the windows)
Never	16 (6.20%)	172 (31.10%)
Rarely	64 (24.81%)	70 (12.66%)
Sometimes	82 (31.78%)	113 (20.43%)
Frequently	37 (14.34%)	93 (16.82%)
Always	58 (22.48%)	100 (18.08%)
NK/NA	1 (0.39%)	5 (0.90%)
Do you see doctor or go to health
center because of the dust?			<0.05
(visiting a doctor)
Never	55 (21.32%)	195 (35.26%)
Rarely	35 (13.57%)	118 (21.34%)
Sometimes	95 (36.82%)	151 (27.31%)
Frequently	32 (12.40%)	48 (8.68%)
Always	39 (15.12%)	36 (6.51%)
NK/NA	2 (0.78%)	5 (0.90%)
What are the main health symptoms you
experience due to dust, in your opinion?			0.64
(perceived health symptoms)
Rhinitis, sinusitis, allergy	155 (60.08%)	333 (60.22%)
Cough, throat and ear irritation	33 (12.79%)	62 (11.21%)
Shortness of breath, difficulty breathing	11 (4.26%)	36 (6.51%)
Skin peeling, dry skin	2 (0.39%)	10 (1.81%)
Others	1 (0.39%)	2 (0.36%)
NK/NA	56 (21.71%)	110 (19.89%)

present the results (frequency, percentage and p-value) of both surveys according to the following: annoyance due to air pollution and dust (Table 3), assessment of air quality and industrial risk perception (Table 4), air pollution perception (Table 5), and consequences of air pollution in quality of life and health (Table 6). Pearson chi-square analyses were performed to compare frequency distributions for the responses in each analyzed construct for both surveys. This procedure is based on [18].

Table 3 displays significant differences between the surveys (p-value less than 0.05 for all three variables). Before the industry interruption, only 2.33% of respondents reported not feeling annoyed by air pollution, while after the interruption, this number rose to 37.79%. In 2014, about 49.22% of the respondents were very and extremely annoyed by dust from outside, but in 2017, this changed to 28.76%. In 2014, the industries were identified by almost 79.46% of respondents as the primary source of dust. However, in 2017, other sources of dust such as vehicles, construction work and unpaved streets also emerged as important contributors, representing 18.44%, 18.00% and 16.64%, respectively.

Table 4 presents significant differences in responses regarding air quality, industrial risk perception, air quality importance and monitoring. In 2014, 12.79% of respondents considered the air quality to be terrible, and 18.22% considered the air quality to be good, while in 2017, only 2.71% considered the air quality to be terrible, and 46.11% considered the air quality to be good. As for industrial risk perception, the results show that the perception of risk has decreased considerably. In 2014, the percentage of very and extremely exposed was more than 50%, while in 2017, this percentage decreased to almost 19%.

In 2014 (see Table 4), about 58.91% indicated that air quality is never monitored, whilst the number dropped to 31.83% in 2017. Nevertheless, about 22–25% of the population reported not being informed about air quality monitoring in both surveys.

Table 5 presents strong differences, comparing perceived air pollution to particles/dust, odour, and air opacity. In 2014, particles/dust was identified as the main form of air pollution, frequently/always perceived by 71.73% of respondents, which was followed by air opacity (20.16%) and odour (9.30%). In 2017, 47.19% frequently/always perceived air pollution by dust, 12% perceived odour, and 7.23% the air opacity, showing a shift toward odour perception and a decrease in the perception of dust and opacity. As shown in a previous study [18], the population experienced improved air quality following the closure of an industrial sintering plant, which was largely due to the decreased presence of dust, soot, and opacity.

Table 6 presents the quotidian consequences of air pollution affecting life quality/routine habits. In 2014, 89.14% of respondents informed that they frequently/always clean the house to remove dust, and 36.82% reported frequently/always closing the windows to prevent dust entering from outside. In 2017, these frequencies slightly decreased: 76.31% reported that they frequently/always clean the house to remove dust, and 34.90% reported closing the windows to prevent dust. The frequency of closing the windows to prevent dust did not decrease considerably. The surveys were conducted in March (during summer in the southern hemisphere), which may explain this behavior, since many houses do not have air conditioning systems.

In 2014, 27.52% of respondents reported always/frequently attending the doctor for an illness caused by the presence of particles in the household, while in 2017, the percentage dropped to 15.19%. In both surveys, the main symptoms reported by the respondents that were caused by dust were rhinitis, sinusitis, and allergy.

The last two questions in 2017 survey were “Would you like the return of the industrial activities of the pelletizing plant? If yes, what are the main reasons for that?”. The results are presented in Figure 4. More than 90% of the respondents favored the return of industrial activities to the pelletizing plant. The first reasons listed were to create and keep jobs (81%) and the plant’s importance to the economy in the region (17%).

3.2. Determinants of Annoyance Due to Sedimented Dust

The independence test allows one to determine whether two categorical variables, which belong to the same sample, are independent or associated with each other [29]. The independence test was carried out between the annoyance variable and the explanatory variables (we selected the variables with significant p-values presented in Table 2, Table 3, Table 4, Table 5 and Table 6) to verify their association at a significance level of 5%.

Table 7. The variable names and their categories.

Variable Name	Code	Categories
Annoyance due to air pollution (response var.)	(Annoy)	extremely (1), very (2), moderate (3), little/nothing (4)
Air quality importance	(ImpQAR)	no importance/little/moderate (1), very (2), extremely (3)
Perception of air quality	(Qdoar)	excellent/good (1), fair (2), bad/terrible (3)
Industrial risk exposure	(Risk)	nothing/little (1), moderate (2), very/extremely (3)
Perceived dust	(PDep)	never/rarely (1), sometimes (2), often/always (3)
Annoyance due to dust inside the house	(IndustA)	nothing/little (1), moderate (2), very/extremely (3)
Cleaning the house	(Clean)	never/rarely/sometimes (1), often (2), always (3)
Closing the windows	(Close)	never/rarely (1), sometimes (2), often/always (3)
Visiting a doctor	(Pmed)	never/rarely (1), sometimes (2), often/always (3)
Perceived health symptoms	(Health)	never/rarely (1), sometimes (2), often/always (3)
Dust reduction perception	(RedDust)	nothing/little (1), moderate (2), very/extremely (3)

presents a complete list of the selected variables and their categories. These variables were subjected to the independence test, with the exception of the variable dust reduction perception, which was only tested in the 2017 survey, as it refers to the respondents’ perception after the activities at the pelletizing plant stopped. The results of the chi-square test, including degrees of freedom, number of respondents, and p-values for each variable, are presented in

Table 8. Chi-square independence tests between degree of air pollution annoyance and explanatory variables (2014 and 2017).

	2014				2017
Variable Name	${\mathbf{\chi }}^{\mathbf{2}}$	df	n	p-Value	${\mathbf{\chi }}^{\mathbf{2}}$	df	n	p-Value
Air Quality Importance	46.26	12	255	0.00	15.4	16	549	0.49
Air Quality Assessment	181.4	16	258	0.00	125.61	16	549	0.00
Industrial Risk Exposure	134.18	16	254	0.00	66.22	16	533	0.00
Dust Deposition Perception	116.90	16	258	0.00	112.13	16	547	0.00
Odor Perception	28.85	16	252	0.02	72.12	16	550	0.00
Opacity Perception	55.23	16	257	0.00	63.16	16	550	0.00
Indoor Dust Annoyance	127.62	16	257	0.00	209.42	16	551	0.00
Dust Source	42.39	20	256	0.00	23.62	50	482	0.26
House Cleaning Frequency	68.14	12	257	0.00	50.86	16	550	0.00
Closing Windows	78.84	16	257	0.00	24.94	16	548	0.07
Doctor Visits Frequency	81.34	16	256	0.00	65.01	16	548	0.00
Health Problems Frequency	53.59	16	246	0.00	96.48	16	547	0.00
Air Pollution Source	36.92	16	254	0.00	25.46	20	494	0.18
Dust Reduction Perception	-	-	-	-	38.02	16	539	0.00
Occupation	21.08	20	251	0.17	25.9	16	544	0.05
Smoking	10.87	8	256	0.21	2.67	8	548	0.95
Gender	9.60	4	258	0.05	15.00	4	553	0.02
Age Group	12.04	12	250	0.44	24.43	12	551	0.02
Education	29.56	24	257	0.20	18.27	24	544	0.78

. To ensure comparability between the two survey periods and to identify robust determinants of annoyance, only variables presenting statistically significant chi-square test results in both years were retained for further analysis. Based on this criterion, significant variables from both surveys were selected for inclusion in the regression models applied before and after the interruption of the pelletizing plant, which was in accordance with the procedure described in [30]. Subsequently, each selected variable was examined using univariate ordinal logistic regression models (see Appendix A). From these univariate analyses, variables with positive and statistically significant coefficient estimates were selected to evaluate the combined effects of multiple determinants on the degree of annoyance in the multivariate models.

3.3. OLR Model Results

The ordinal logistic regression model (OLR) was applied to identify the categories and variables associated with levels of annoyance due to settleable particles (SPM). According to Section 2.3, the response variable Y, the annoyance degree with five ordered categories—extremely annoyed ($Y=1$), very annoyed ($Y=2$), moderately annoyed ($Y=3$), slightly annoyed ($Y=4$) and not annoyed ($Y=5$)—became the references category. It was necessary to group the categories (little and not) because in stereotype models for ordinal variables, grouping categories with low frequency is used to ensure parameter identifiability, avoid convergence issues in the maximum likelihood estimation process, and reduce the variance of the estimates. Categories with few observations contribute to numerical instabilities and may lead to the singularity of the information matrix. When preserving the natural order of the variable, grouping enhances the robustness, efficiency, and interpretability of the model. All NK/NA responses were excluded from this analysis, resulting in a sample size $n=249$ in 2014 and $n=501$ in 2017.

The OLR models were run with the rrvglm command from the VGAM package implemented by R software Version 3.6.1. The estimations of the products ${\varphi }_j{\beta }_r$ defined in (5) and the odds ratio (OR) $exp\left({\varphi }_j{\beta }_r\right)$ [31] are presented in

Table 9. Estimated OR for the covariates in the 2014 and 2017 models.

	2014			2017
	${\mathit{\varphi }}_{\mathbf{1}}{\mathit{\beta }}_{\mathit{r}}$ (OR)	${\mathit{\varphi }}_{\mathbf{2}}{\mathit{\beta }}_{\mathit{r}}$ (OR)	${\mathit{\varphi }}_{\mathbf{3}}{\mathit{\beta }}_{\mathit{r}}$ (OR)	${\mathit{\varphi }}_{\mathbf{1}}{\mathit{\beta }}_{\mathit{r}}$ (OR)	${\mathit{\varphi }}_{\mathbf{2}}{\mathit{\beta }}_{\mathit{r}}$ (OR)	${\mathit{\varphi }}_{\mathbf{3}}{\mathit{\beta }}_{\mathit{r}}$ (OR)
ImpQAR2	1.52 (4.57)	0.90 (2.46)	0.28 (1.32)	–	–	–
ImpQAR3	1.58 (4.84)	0.94 (2.55)	0.29 (1.33)	–	–	–
Qdoar2	1.45 (14.26)	0.86 (2.36)	0.26 (1.30)	0.55 (1.74)	0.75 (2.12)	0.29 (1.34)
Qdoar3	5.01 (149.7)	2.97 (19.53)	0.91 (2.50)	1.37 (3.94)	1.86 (6.44)	0.72 (2.06)
Risk2	2.05 (7.75)	1.22 (3.37)	0.37 (1.45)	0.50 (1.64)	0.67 (1.96)	0.26 (1.30)
Risk3	3.72 (41.13)	2.21 (9.07)	0.68 (1.97)	0.54 (1.72)	0.74 (2.09)	0.29 (1.33)
PDep2	1.68 (5.38)	1.00 (2.71)	0.31 (1.36)	0.62 (1.86)	0.84 (2.33)	0.33 (1.39)
PDep3	2.99 (19.89)	1.77 (5.90)	0.55 (1.73)	1.11 (3.02)	1.50 (4.49)	0.58 (1.79)
Close2	0.09 (1.10)	0.06 (1.06)	0.02 (1.02)	–	–	–
Close3	2.31 (10.05)	1.37 (3.93)	0.42 (1.52)	–	–	–
Pmed2	1.24 (3.46)	0.74 (2.09)	0.23 (1.25)	–	–	–
Pmed3	0.86 (2.36)	0.51 (1.66)	0.16 (1.17)	–	–	–
Indust2	–	–	–	1.18 (3.24)	1.60 (4.94)	0.62 (1.86)
Indust3	–	–	–	1.90 (6.72)	2.59 (13.29)	1.00 (2.73)
Clean2	–	–	–	0.35 (1.41)	0.47 (1.60)	0.18 (1.20)
Clean3	–	–	–	0.35 (1.42)	0.47 (1.60)	0.18 (1.20)
Health2	–	–	–	0.23 (1.25)	0.31 (1.36)	0.12 (1.13)
Health3	–	–	–	0.46 (1.59)	0.63 (1.87)	0.24 (1.28)

for the data sets obtained in the 2014 and 2017 surveys, respectively. For both models, the estimated odds ratios (ORs) are associated with each of the grouped annoyance response categories: (1) extremely annoyed, (2) very annoyed, and (3) moderately annoyed.

When analyzing the results in Table 9, the significant variables common to both the 2014 and 2017 models include air quality perception (Qdoar2 and Qdoar3), industrial risk exposure (Risk2 and Risk3), and perceived dust (PDep2 and PDep3). Consistent across both models, there was a marked decrease in OR values, which demonstrates that the population perceived improved air quality, reduced industrial risk, and less dust after the pelletizing plant ceased operations.

4. Discussion

The comparison between surveys conducted before and after the stoppage of the pelletizing plant revealed substantial shifts in the population’s perceptions of air quality, risk, and everyday behaviors related to dust exposure. After the closure, the proportion of respondents who perceived the air as “good” or “excellent” increased markedly, while reports of strong annoyance and risk perception declined. These patterns are consistent with previous findings showing that interventions reducing industrial emissions lead to a sense of lower exposure and improved environmental conditions [18].

A relevant change also occurred in how people perceived the monitoring and control of air quality. Although a significant share of respondents still declared not being informed about local monitoring, the proportion who believed that air quality was “never monitored” decreased after the closure. This could indicate that public communication about environmental monitoring improved in the period following the shutdown—possibly influenced by the national media coverage of the Mariana environmental disaster—or, alternatively, that there remains confusion among residents between monitoring practices and regulatory control.

Behavioral habits reflected both improvements and the persistence of defensive routines. Although the frequency of cleaning activities decreased slightly, closing windows to avoid dust remained common. As the surveys were conducted in March (summer in the southern hemisphere), high temperatures and the limited availability of air-conditioning systems may help explain the persistence of this behavior. Even under improved air quality conditions, such habits may persist due to climatic factors and housing characteristics. Despite improvements in perceived air quality and reductions in annoyance, the community’s economic dependence on the plant strongly influenced public opinion about its reopening. More than 90% of respondents expressed support for resuming industrial activities, primarily citing job creation and the plant’s importance to the regional economy. This apparent contradiction highlights the tension between environmental and socioeconomic priorities. Individuals tend to value environmental quality more once basic material and security needs are satisfied [32,33]. Thus, even though residents recognized the environmental benefits of the shutdown, the perceived threat to economic stability led many to favor the return of operations.

The ordinal logistic regression models further reinforce these patterns by quantifying how perceptions of air quality and risk were associated with annoyance before and after the industrial stoppage. In the 2014 model (Table 9), the variables Qdoar2 and Qdoar3 (air quality perception) showed strong associations with annoyance. For instance, the category Qdoar3, corresponding to respondents who rated air quality as bad or terrible, had an OR of 149 (see Table 9), indicating that these respondents were 149 times more likely to report extreme annoyance due to dust. In contrast, in the 2017 model (Table 9), the corresponding value decreased substantially (OR ≈ 3.9), confirming that the association between perceived air quality and annoyance weakened after the shutdown. Such attenuation mirrors the results presented in [18], following the closure of a sintering plant in Sweden.

In 2014, the variables “visiting a doctor” (Pmed2 and Pmed3) were also significant (Table 9): respondents who often or always sought medical care were 3.4 and 2.3 times more likely, respectively, to report high annoyance. This association reflects the elevated particulate matter concentrations during the plant’s operation, which is when pollution-related symptoms and medical consultations were more frequent. Similarly, individuals who considered air quality very or extremely important (ImpQAR2 and ImpQAR3) were about four times more likely to report extreme annoyance, which is consistent with a study conducted in the same region [19].

In 2017 (Table 9), two additional variables became relevant: perceived health symptoms Health2 and Health3) and cleaning frequency (Clean2 and Clean3). Those who reported frequent health symptoms were 1.2 and 1.5 times more likely to feel extremely annoyed, which was possibly due to the higher proportion of respondents over 55 years old in that survey. This result contributes to other documented findings from studies that show an association between air pollution and the occurrence of symptoms and health effects [34,35]. The association with cleaning habits suggests that although industrial emissions declined, other sources—such as vehicular traffic and soil resuspension—continued to affect daily experiences of air pollution. This result contributes to the fact that air pollution negatively affects quality of life [4,5]. These findings collectively suggest that SPM concentration is a direct contributor to dust-related annoyance. Furthermore, it is clear that industrial emissions directly influence the perception of air pollution, subsequently impacting an individual’s assessment of health symptoms, industrial risks, air quality, and overall annoyance levels.

5. Conclusions

This study evaluated changes in community perceptions of settleable dust and air pollution annoyance before and after the interruption of a large pelletizing industry in a coastal urban–industrial region of southern Brazil. The comparison of perception surveys conducted in two distinct periods allowed the assessment of how changes in industrial activity influenced annoyance, risk perception, and everyday behaviors related to dust exposure. The results indicate that the interruption of industrial activities was associated with a reduction in perceived dust exposure, industrial risk, and annoyance levels. Variables related to perceived air quality, industrial risk exposure, and dust deposition remained associated with annoyance in both survey periods, although their influence decreased after the shutdown. Despite these environmental improvements, most residents expressed support for the resumption of industrial activities, which was mainly due to employment and regional economic considerations. These findings highlight the complex relationship between environmental quality and socioeconomic needs in industrialized communities and underline the importance of considering both dimensions when evaluating air pollution impacts.