Environ. Sci. Technol. 2006, 40, 6903-6911
Measured Concentrations of VOCs in Several Non-Residential Microenvironments in the United States M I R A N D A M . L O H , * ,† E. ANDRES HOUSEMAN,† GEORGE M. GRAY,† JONATHAN I. LEVY,† JOHN D. SPENGLER,† AND D E B O R A H H . B E N N E T T †,‡ Harvard School of Public Health, Boston, Massachusetts 02215 and University of California, Davis, California 95616
Individuals spend about 25% of their time in non-residential indoor microenvironments. For some of these microenvironments, particularly stores and restaurants, exposures to volatile organic compounds (VOCs), have not been well characterized. In the Boston Exposure Assessment in Microenvironments (BEAM) study, sampling using scripted activities was conducted in stores, restaurants, and transportation in the summer of 2003 and winters of 2004 and 2005. A suite of VOCs including hydrocarbons, several chlorinated compounds, and aldehydes was analyzed. Nine store types were sampled using a composite design to enable a greater number of stores to be visited. Stores had higher concentrations of formaldehyde, toluene, ethylbenzene, xylenes, and styrene than other microenvironments, particularly in certain store types. Geometric mean formaldehyde levels were highest in the housewares stores, at 53 µg/m3 (95% CI ) 43, 66). Geometric mean toluene levels were highest in multipurpose stores, at 76 µg/ m3 (95% CI ) 50, 118). The levels observed in stores were several times higher than levels found in transportation microenvironments, and indicate strong indoor sources. In contrast, benzene did not have significantly higher levels in stores than typically found outdoors. Concentrations of formaldehyde and benzene, ethylbenzene, xylenes, and styrene were strongly influenced by the presence of smoking in the dining microenvironment. Chloroform levels were higher in restaurants than in other microenvironments, with a geometric mean of 1.1 µg/m3 (95% CI ) 0.7, 1.8). The VOC concentrations found in stores and restaurants in this study are a potentially important source of exposure for sensitive individuals or people who work in these microenvironments.
Introduction Volatile organic compounds (VOCs) can induce a range of health effects, including irritation to the eyes, mucous membranes, skin, and respiratory tract, effects on the nervous system, cancer, and liver and kidney toxicity (1). The United States Environmental Protection Agency (EPA) estimated that * Corresponding author phone: (617) 384-8815; fax: (617) 3848854; e-mail:
[email protected]. † Harvard School of Public Health. ‡ University of California, Davis. 10.1021/es060197g CCC: $33.50 Published on Web 10/19/2006
2006 American Chemical Society
over half of excess cancer risk from outdoor concentrations of hazardous air pollutants can be attributed to several VOCs, in particular formaldehyde, benzene, and 1,3-butadiene (2). Outdoor concentrations, however, have been shown to contribute only partially to total human risk from VOCs (310). For several carcinogens, indoor home levels are often higher than outdoor levels, due to additional sources indoors. The Total Exposure Assessment Methodology (TEAM) studies showed that personal exposure to certain compounds was most influenced by specific activities and indoor sources. For example, 1,4-dichlorobenzene was associated with use of mothballs and deodorizers, chloroform was associated with washing clothes and dishes, benzene, styrene, and other hydrocarbons were associated with smoking, and perchloroethylene was associated with visiting dry cleaners (4, 11). Source apportionment analyses of indoor and personal exposures and emissions studies have found that formaldehyde and acetaldehyde were associated with building materials and consumer products; styrene was associated with flooring adhesives, rubber products, and plastics; ethylbenzene and xylenes were from paints, cleaning products, and building materials; and perchloroethylene was associated with dry cleaning and textile treatments (10, 1218). Methylene chloride, toluene, 1,3-butadiene, and trichloroethene also have indoor sources (7). One could therefore hypothesize that stores selling building materials and consumer products would have elevated concentrations of VOCs compared with the outdoors. The store VOC levels would also be influenced by the different mixes of products found in different store type (e.g. hardware versus grocery stores). While VOC exposures in the home have been widely studied, people spend 25%, on average, of their time indoors away from home. A cross-sectional 24-hour survey study found that about 6% of time across the U.S. population is spent in a vehicle, 5% in an office or factory, and 2% in a bar or restaurant, with 11% in other indoor locations (19). Transportation and offices have been previously investigated (20-33). Less data exist on air concentrations of VOCs in shops and restaurants, where exposures could be important if concentrations are high or for people who work there. Previous studies of VOCs in shops and restaurants have been done by Kim et al. in the United Kingdom, and Lee et al. and Guo et al. in Hong Kong. In Hong Kong, restaurants and shopping malls had the highest indoor concentrations for formaldehyde, and, compared to homes in these studies, were also high for benzene, toluene, ethylbenzene, and xylenes (BTEX) (34, 35). Kim et al. sampled in department stores, perfume shops, offices, laboratories, cinemas, libraries, pubs, restaurants, and various transportation microenvironments (36). They found the highest VOC concentrations in transportation microenvironments and pubs. While there have been studies in the United States measuring particulate matter in restaurants and bars, VOCs have not been widely measured in these locations. One aspect of the Boston Exposure Assessment in Microenvironments (BEAM) study is to provide a more complete view of personal exposure to VOCs by monitoring in several non-residential microenvironments. In this study, we specifically characterized the distribution of VOCs in stores, restaurants, and transportation modes in the Boston metropolitan area. We expanded upon the store types visited in previous studies, including hardware stores, grocery stores, furniture stores, drug stores, housewares stores, sporting goods stores, department stores, electronics stores, and VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6903
TABLE 1. Number of Samples and Visits To Each Microenvironment
a
microenvironment
seasons sampled
number of samples
number of places
total number of visits
hardware multipurpose grocery drug stores sporting goods furniture stores housewares stores department stores electronics dining nonsmoking dining smoking transportation multipurpose
summer,a winterb summer,a winterb summera summera summera summera summera winterb winterb winterb winterb summer,a winterb winterc
12, 11 8, 7 12 7 7 6 7 5 7 13 7 21 28
17, 15 12, 9 16 8 14 11 16 10 9 13 7 5 modes 3
22, 19 14, 9 20 11 14 14 17 15 10 13 7 21 28
2003.
b
2004. c 2005.
multipurpose stores. We used a composite sampling design for stores, visiting multiple locations using the same sampler, allowing us to increase the number of places sampled. This type of sampling is typically used in medical screening tests and contaminated site remediation assessments (37, 38) and is a means of cost-effectively gathering information from a wide range of sources. In this paper, we compare VOC concentrations across microenvironments and consider variability within microenvironment types, including a sub-study to explore both day-to-day and within day variability at multipurpose stores. We compare microenvironmental concentrations with concentrations of VOCs in the outdoors and in other microenvironment studies. We assess whether compounds that are considered predominantly of indoor origin or those with mixed indoor and outdoor origins based on studies in homes had high concentrations in the BEAM microenvironments relative to typical outdoor concentrations.
Materials and Methods Study Design. Microenvironments were categorized into retail stores (further classified according to the types of products sold), smoking or nonsmoking restaurants or bars, and five modes of transport. A given store type was sampled in the summer of 2003, the winter of 2004, or during both seasons (Table 1). Dining samples were taken in winter 2004. Transportation microenvironments were sampled in both seasons. In the winter of 2005, we conducted additional sampling in multipurpose stores to quantify temporal variability. All sampling occurred in the greater Boston metropolitan area. We chose different sampling designs depending on the microenvironment, since our goals were different in each. For stores, we used a composite design because we were interested in estimating mean concentrations in many different types of stores. For a given number of samples, more locations can be visited with less time spent in each location, improving estimation of the mean without increasing sampling time or analytical costs. The variance can also be estimated, although this estimate is less robust than that of the mean (39). In composite sampling, it must be assumed that pooling the data collected (e.g., the air or other media) on a single sample will not interfere with the compounds of interest (38). We did not use composite samples for restaurants or for transportation. Field staff collected samples by wearing personal samplers while performing activities relevant to the microenvironment being monitored. In retail stores, staff would walk around the store as if shopping, spending an equal amount of time in the various parts of the store. In restaurants, staff would be seated at a table as a customer. 6904
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 22, 2006
Retail Stores. Store types included hardware stores, grocery stores, furniture stores, drug stores, housewares stores, sporting goods stores, department stores, electronics stores, and multipurpose stores. Housewares stores were defined as selling kitchen, bedroom, and bath items for the home. Multipurpose stores were large stores that carry a variety of goods, including clothing, food, toys, electronics and hardware products. Department stores sold primarily clothing and items for the home. One of several composite designs was used for each store type. Each sampling tube contained air from 1, 2, or 3 locations. The standard error for estimating the mean and variance depended on the particular combination of sites per sample and visits per site. See the Supporting Information for more details on the composite sample methodology. Multipurpose Store Variability Sampling. Additional noncomposite sampling was performed in winter 2005 at multipurpose stores to examine within-store and betweenstore variability. Three different stores were examined in different parts of the Boston metropolitan area, with a total of 10 individual samples per store on different days (referred to as the day-to-day set). In addition to these samples, samples were collected every other hour from 10 a.m. to 5 p.m. in one store for 2 days (the within-day set). Dining. Samples were taken in both smoking and nonsmoking restaurants and bars. Sampling time at each location was 1.5 hours, since each sample had only one location (see Table 1). Instead of consumer products, we expect VOC levels to be influenced by smoking, size of the establishment, and activities such as cooking and cleaning. Transportation. Three-hour noncomposite samples were taken during morning and evening rush hours in each of the transportation modes simultaneously: highway car, urban car, bus, train, and walking (summer only). The urban car and walking routes closely followed the bus route. Both the highway car and train routes traveled between Boston and the suburban town of Beverly, north of Boston. Sample Collection. The personal sampling unit consisted of one VOC sorbent tube (Perkin-Elmer) and a DNPH aldehyde cartridge (Waters, Inc.) connected to a pump (BGI, AFC 400s) and battery (BGI 8x Sanyo HR 4/3 FAU), along with a temperature/relative humidity HOBO (Onset Computer), in a backpack. Sorbent tubes were analyzed for benzene, toluene, ethylbenzene, and the xylenes (BTEX), 1,3butadiene, styrene, and several chlorinated compounds, as well as MTBE and several alkanes as mobile source markers. Cartridges were analyzed for formaldehyde and acetaldehyde. Flow rates depended on the sampling time and were adjusted to maintain target volumes of 10 L for sorbent tubes and 100 L (200 L if sample was 3 h or longer in duration) across samples of the same microenvironmental type. Rates ranged from
167 mL/min (VOC) and 1.7 L/min (aldehyde) for 1 h samples to 56 mL/min (VOC) and 556 mL/min (aldehyde) for 3 h samples. For composite samples, equal volumes of air were collected at each location. Field blanks were taken on 11% of samples, and all compounds except for 1,3-butadiene, methylene chloride, and trichloroethene had less than 10% of samples below the detection limits. Details of the sample analysis and QA/QC can be found in the Supporting Information. Data Analysis Methods. Composite Estimation. In order to estimate the geometric means of each store type using the composite samples, we assumed that the dataset consisted of the averages of lognormal variables (Yij), as shown in the model below:
Y hi )
1 ki
ki
∑Y
ij
(1)
j)1
where Y h i ) the concentration on sample i, ki ) the number of sites sampled on sample i, and Yij ) the concentration at site j on sample i. The Yijs were assumed independent and identically distributed for a given store type. We remark that the sample concentrations are assumed to be sums of lognormal variables, not products, and the resulting probability distributions can be obtained only from convolution integrals. Statistical inference on the parameters of the underlying lognormal distributions is therefore complicated. To address this issue, and also to account for the leftcensoring typical of environmental concentration data, we used Markov-Chain Monte Carlo (MCMC) techniques (40) implemented in WinBUGS Version 1.4 (41, 42) for statistical analysis. Although a Bayesian framework is required for WinBUGS, we used non-informative prior distributions for all estimated parameters. Numerical problems prevented estimation in one case (m,p-xylene in hardware stores, where there was an outlier of 1760 µg/m3), in this case we used maximum likelihood, ignoring the issue of left-censoring. In all cases, we obtained estimates of the geometric means and standard deviations for the Yijs for each microenvironment. Programs are available from the authors upon request. The coefficient of variation (CV), the ratio of the standard deviation to the mean, was calculated to compare the distributions of each microenvironment using the following equation:
CV ) x[exp(σ2) - 1]
(2)
where σ ) log[GSD(Yij)]. This formula takes advantage of the fact that the CV of a lognormal distribution depends only on its variance parameter, σ2 ) Var[log(Y)]. Multipurpose Store Variability Data. A random effects model was used to apportion the variance between stores and within stores (43). The model assumes that there is an underlying population mean, µ, for all multipurpose stores, and that the concentration in each store is a function of µ and the variance between stores and within the same store on different days. As shown in eq 3,
log(Yij) ) µ + ai + bij + ijk
(3)
where log(Yij) ) the log-transformed concentration at multipurpose store i on day j, µ ) the mean concentration for all multipurpose stores, ai ) the portion of variability attributable to the location i, bij ) the portion of variability attributable to day j at location i, and ijk ) the residual variability due to error and within-day heterogeneity (k) on day j at location i. All single site multipurpose data, including day-to-day and within-day, were included in this analysis.
Seasonal Differences. Hardware and multipurpose stores were studied in both winter and summer. To test for seasonal differences, a Wilcoxon rank sum test (p < 0.05) was performed. Dining Data. These samples were not composite samples and therefore the assumption of additivity of lognormal variables on each sample does not apply. A regression analysis was conducted using information on restaurant characteristics. Benzene, 1,3-butadiene, toluene, styrene, and chloroform were chosen for the regression analysis. The first four compounds are hypothesized to be primarily due to smoking and possibly traffic, while chloroform is hypothesized to be related to water use or cleaning in the restaurant. Transportation Data. The transportation data were examined to see if there were differences between morning and evening, and by season using a Wilcoxon signed rank test (p < 0.05) and by mode using a Kruskal Wallis test (p < 0.05). In the results, the BEAM transportation data (excluding the train route, which would not be representative of ambient concentrations) are used as an outdoor comparison because a direct outdoor measurement for each indoor store measurement could not be taken due to logistical constraints. The transportation samples can be considered representative of outdoor concentrations near the stores and restaurants, which are often located near busy roadways, similar to the urban and highway routes traveled in the study. Correlation Analysis. Correlation matrices between all compounds were calculated to examine relationships between compounds and possible source groups. We included 2,2,4-trimethylpentane (2,2,4-TMP) as a potential mobile source marker. Spearman correlations were calculated on nontransformed and nonweighted concentrations in the data. Data were divided into microenvironment types (stores, dining, and transportation) to keep matrices as homogeneous as possible. Correlations of 0.65 or more were considered highly correlated. All statistical analyses were done with the R statistical package, version 2.2.0 (http://cran.r-project.org/).
Results and Discussion The geometric mean, maximum, and coefficient of variation (CV) for all stores and dining locations are shown in Table 2. In particular, toluene is twice as high in stores than in dining, while the opposite is true for benzene and acetaldehyde. The maximum values for some compounds are much higher than the geometric mean values. Retail Stores and Transportation. Figure 1 compares the geometric means from the composite analysis for each retail store type and the BEAM transportation measurements (see Supporting Information for geometric mean values, coefficients of variation, and minimum and maximum values). For the transportation data, there were no significant differences by season, transportation mode, or time of day for most compounds. For compounds where there was a difference (MTBE, formaldehyde, and chloroform for season and chloroform and perchloroethylene for mode), the concentrations and variation between samples were much lower than in indoor environments, therefore we felt it reasonable to average these values as a representative outdoor comparison. We did not include the train data as we wanted to use on-road modes only to more closely reflect road concentrations. We expect the VOC concentrations in stores to be most influenced by the presence of a large number of new products and other sources of emissions (e.g., use of motorized objects or cleaning). In particular, formaldehyde and toluene had higher concentrations in stores than in transportation, indicating strong indoor sources for these compounds. As seen in Figure 1, formaldehyde geometric mean concentrations ranged from 6.0 µg/m3 (95% CI ) 2.6, 13) in electronics VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6905
TABLE 2. Summary Statistics for All Store and Dining Samples Stores compound formaldehydec,d acetaldehyded 1,3-butadienec MTBE benzene toluene ethylbenzene m,p-xyleneb o-xylened methylene chlorided chloroformc carbon tetrachloridec trichloroethene perchloroethylene 1,4-dichlorobenzene styrenec,d
geometric mean 19.6 10.8 0.21 2.26 1.77 33.3 4.04 10.7 4.06 2.24 0.44 0.90 0.43 1.41 2.40 3.04
95% CI 16.9 9.45 0.14 1.81 1.58 27.6 3.24 9.4 3.34 1.70 0.36 0.80 0.33 1.17 1.74 2.47
max
22.6 90.6 12.4 71.7 0.30 2.20 2.79 46.5 1.98 7.92 40.1 520 5.06 107 12.2 1765 4.89 62.1 2.86 123 0.53 5.51 1.01 7.53 0.57 115 1.70 43.8 3.17 50.3 3.72 33.2
Dining CV 1.27 1.11 1.74 1.9 0.80 2.01 2.68 2.98 2.02 3.28 1.93 0.85 3.69 1.76 5.01 2.09
95% CI 1.09 0.94 1.13 1.46 0.67 1.63 2.07 2.45 1.61 2.34 1.48 0.70 2.58 1.41 3.33 1.62
1.52 1.33 3.27 2.6 0.96 2.55 3.68 3.73 2.65 5.09 2.62 1.03 5.94 2.32 9.07 2.87
geometric mean
95% CI
14.3 24.4 1.05 2.40 3.07 14.9 1.90 6.23 2.02 0.67 1.11 0.65 0.23 2.09 1.45 1.19
9.83 20.3 15.0 39.4 0.48 2.20 1.70 3.36 2.18 4.30 8.7 25.3 1.39 2.57 4.35 9.01 1.49 2.76 0.15 2.11 0.69 1.83 0.57 0.73 0.08 0.63 0.96 4.60 0.63 3.31 0.87 1.64
max
CV
47.8 1.02 185 1.61 35.5 4.73 9.94 0.91 22.8 0.91 622 1.97 8.52 0.79 28.6 0.97 10.1 0.80 90.3 25.5 8.32 1.56 1.02 0.27 118 10.29 83.4 4.86 171 5.92 6.67 0.88
95% CI 0.7 1.69 1.0 3.44 2.0 22.8 0.6 1.46 0.6 1.47 1.2 4.46 0.6 1.23 0.7 1.59 0.6 1.26 3.5 4927 1.0 3.13 0.2 0.38 2.8 215 2.1 20.95 2.4 34.41 0.6 1.42
a Only analyzed for in the winter. b m,p-Xylene store data were analyzed using a non-Bayesian method. c Significant (p < 0.05) difference by season in hardware stores according to Wilcoxon rank sum test. d Significant (p < 0.05) difference by season in multipurpose stores according to Wilcoxon rank sum test.
FIGURE 1. Geometric means of VOCs in BEAM stores and transportation data. PCE ) perchloroethylene, CCl4 ) carbon tetrachloride, TCE ) trichloroethene, 1,4-DCB ) 1,4-dichlorobenzene, MTBE ) methyl-tert-butyl ether. stores to 53 µg/m3 (95% CI ) 43, 66) in housewares stores. All store types exceeded transportation means, except for electronics stores. Formaldehyde was highest in housewares and furniture stores, most likely due to the sale of pressed wood products known to emit formaldehyde (17, 44, 45). In addition, secondary reactions between oxidizing agents and other VOCs, as well as certain surfaces, can contribute to aldehyde concentrations indoors (15, 46-50). Housewares and furniture stores with high levels of unsaturated hydrocarbons, such as limonene from waxes and polishes and other surface coverings, could contribute to the formaldehyde source strength (15, 17, 49, 51, 52). Toluene in stores was found to have geometric means from three to over ten times the transportation mean (Figure 6906
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 22, 2006
1), pointing to strong and numerous indoor sources. Emissions studies have found toluene in many types of products, including adhesives, cleaners, polishes, waxes, paints, packaging materials, and other solvent-containing materials (15, 16, 51). Multipurpose stores had particularly high toluene levels, with a geometric mean of 76 µg/m3 (95% CI ) 50, 118). This could be explained by the large variety and amount of products stocked in this store type. The distribution of toluene concentrations exhibited a low amount of variability, with CVs between less than 1 and 2, a sign that toluene is found commonly in stores. Acetaldehyde in stores did not display as large a difference from transportation concentrations as toluene and formaldehyde. Acetaldehyde concentrations were highest in grocery
stores, 19 µg/m3 (95% CI ) 13, 28), and lowest in electronics stores, 2.2 µg/m3 (95% CI ) 0.87, 6.1). The high concentration in grocery stores may be due to the use of acetaldehyde as a fumigant and preservative for some fruits and as a byproduct of yeast used in baking (53). Ethylbenzene, the xylenes, and styrene had higher instore concentrations in hardware stores, housewares stores, and multipurpose stores, than transportation concentrations (Figure 1). Hardware stores had the highest levels of ethylbenzene and xylenes, with mean levels about ten times the mean transportation concentration. For ethylbenzene and the xylenes, most stores were of low variability, with CVs around 1-3. We would expect a higher source strength in hardware, housewares, and multipurpose stores, since these compounds are found in solvents, paints, varnishes, caulking. and adhesives, products that are either sold or used in these stores (15-18). In other stores, the levels were close to transportation levels, indicative of fewer indoor sources in those stores. Styrene displayed a similar pattern of having the highest geometric means in hardware, housewares, and multipurpose stores. Styrene had low variability, with CVs below 1 or between 1 and 2 except for electronics stores, where the CV was 10. In stores, styrene would be found in paints, plastics, synthetic rubber products, and some cleaners and adhesives (17, 18). Methylene chloride was much higher in hardware stores and multipurpose stores than in transportation (Figure 1), indicating that certain products sold only in these stores, such as paint removers and solvents, contain this VOC (16, 17). Methylene chloride was also more variable in stores, with a maximum CV of 5 in furniture stores, which may be due to greater variability in type and number of sources among stores. Store concentrations of trichloroethene (TCE), perchloroethylene (PCE), and chloroform were also several times greater than transportation concentrations. Chloroform was highest in grocery stores, 1.1 µg/m3 (95% CI ) 0.67, 1.6), perhaps from greater use of cleaning products and hot water compared to other types of stores (11, 54). Chloroform and TCE had particularly high CVs in certain microenvironments. PCE’s highest geometric mean was in sporting goods stores, 3.0 µg/m3 (95% CI ) 1.5, 5.5), compared with 0.78 µg/m3 (95% CI ) 0.54, 1.1) in Boston transportation microenvironments. In stores, PCE may be found in dry-cleaned fabrics or in some types of degreasers, explaining the concentrations found in sporting goods stores, while TCE would most likely be found as a degreaser or in some adhesives (16). The geometric mean for 1,4-dichlorobenzene (1,4-DCB) was five times greater in hardware stores than in transportation samples and more than ten times higher in housewares stores, multipurpose, and drug stores. It was also one of the more variable compounds, with most of the store CVs above 2. 1,4-DCB is specifically found in some moth repellent and deodorizing products (4). The stores with higher levels were seen to carry 1,4-DCB products in varying amounts. In places where these products were not sold, they may still have been used, particularly in bathrooms. The only compounds where the stores did not consistently have concentrations several times higher than transportation were MTBE and benzene. MTBE and benzene are be expected to come primarily from mobile sources, since they are not used in any consumer products and thus indoor concentrations should only be due to infiltration from the outdoors. Benzene and MTBE levels in stores were comparable to those in transportation and had low variability. In terms of climatic influence, we found that concentrations in stores sampled in the winter were not as high as concentrations in summer, although the climatic influence is difficult to evaluate since only hardware and multipurpose
stores were measured in both seasons. Table 2 shows the compounds that had significant seasonal differences and the store types in which these differences were found. Formaldehyde showed significantly higher levels in summer in both types of stores, while acetaldehyde was higher in summer in multipurpose stores. Formaldehyde levels would be expected to show a strong seasonal effect, as concentrations have been found to increase with increasing temperature and relative humidity (55, 56), as is likely for acetaldehyde as well. For the other compounds, similar climatic variables may influence indoor emissions. Air exchange rates and indoor climatic regulation may account for some of the differences. Commercial buildings are generally use mechanically ventilated, and, depending on the system and cycle used, may have a larger intake of outside air. This may contribute to higher concentrations during the summer for certain compounds. This may contribute to the increased concentrations during the summer for certain compounds. It is possible that there are seasonal interior humidity differences, with winter being drier, that may contribute to emissions and reaction rates. The correlation analysis provides an exploration of the source patterns in stores (Tables 3 and 4). Fewer compounds were highly (>0.65) correlated in stores, compared to transportation and restaurants, indicating a diverse array of source types. For example, toluene and formaldehyde did not correlate with other compounds in stores, supporting the idea that sources are strong yet vary within each store type. In stores, ethylbenzene and the xylenes did not highly correlate with 2,2,4-TMP and MTBE, both mobile source markers. However, this lack of correlation may be driven by the high concentrations of ethybenzene and xylenes in hardware, housewares, and multipurpose stores, since other store types did not have substantially elevated levels compared to the transportation geometric mean. It is likely that this low correlation with mobile source markers is driven by the very strong indoor source strength for ethylbenzene and the xylenes in hardware, housewares, and multipurpose stores. Styrene was also correlated with ethylbenzene and xylenes in stores. There was much less correlation among the chlorinated compounds (Table 4), and much greater heterogeneity in store concentrations, indicating less common sources (see Figure 1 for geometric mean levels). Store Variability. The BEAM data had several extremely high samples, making the actual distribution difficult to characterize. For example, m,p-xylene had one hardware store sample with a value of 1760 µg/m3, although the geometric mean for hardware stores was 42.1 µg/m3. Trichloroethene also had maximum sample values more than 100 times the geometric mean values. The highest samples were composites of three different sites. Because the concentration on a tube is the average of the concentrations in each of the three stores, one store could have a concentration higher than the resulting tube average. VOC concentrations vary in a given store over time, between stores of a given type, and between different types of stores. Within one store, differences in product stocking, activities such as cleaning or renovating, traffic, or even the number of customers can influence the temporal variability of VOC concentrations. Between stores of the same type, these factors would also influence the variability, as might the quantity and classes of items, store size, or ventilation. Some of the variability may also be due to the fact that we were using personal, rather than stationary samplers. Field staff may not walk the exact same route or spend the same amount of time in each part of the store. This, however, is reflective of the true “shopping experience” and is probably more akin to the variability in people’s true exposure in these microenvironments. VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6907
TABLE 3. Correlation Patterns in Restaurantsa
a All (smoking and non-smoking) establishments are shown on the right of the diagonal while nonsmoking locations are shown on the left. Shaded cells indicate correlations above 0.65 in various microenvironments.
TABLE 4. Correlation Patterns in Stores and Transportationa
a Stores are shown on the right of the diagonal while transportation data are shown on the left. Shaded cells indicate correlations above 0.65 in various microenvironments.
The pattern of variability of stores has not been previously characterized, so we conducted a small study in multipurpose stores, examining three different stores over several weeks and one store over the course of the day over 2 days. Figure 2 shows the proportion of variance attributable to location (a), day (b), and a term encompassing both the within-day heterogeneity as well as the residual error (). For all compounds, except 1,3-butadiene, carbon tetrachloride, and methylene chloride, more of the variability can be attributed to the different days that we visited the store rather than location. Within a given day the variability seems to be low relative to between days. For 1,3-butadiene and carbon tetrachloride, the residual error term accounted for most of the variability. It is unlikely that 1,3-butadiene has sources in multipurpose stores, while carbon tetrachloride is expected to have relatively constant concentrations, so differences are most likely due to error. For methylene chloride, the 6908
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 22, 2006
proportions from each term were similar. Since these results were based on a limited sample, for one store type, and for three stores only, they may not be representative of all stores. Still, they give us an idea of the potential effect of temporal variability on store concentrations. Dining. Figure 3 shows boxplots comparing the data of smoking and nonsmoking restaurants for selected compounds. Most compounds tended to have at least one or more outliers. The dining microenvironment has lower geometric means for most compounds than the shopping microenvironment. In multivariate regression models, smoking was a significant predictor for all compounds tested except for chloroform. Restaurant capacity and interaction terms between capacity and smoking improved the fit of the model, but terms were not necessarily all significant. For chloroform, covariates related to the kitchen, such as distance to kitchen and whether
FIGURE 2. Apportionment of variance: a ) between stores; b ) between days; epsilon ) residual and within-day variation.
FIGURE 3. Comparison of selected VOCs in smoking and nonsmoking dining samples. Y-axes on a log scale. S ) smoking, NS ) nonsmoking, FORM ) formaldehyde, ACET ) acetaldehyde, TOL ) toluene, BENZ ) benzene, CHLOR ) chloroform. *One point for acetaldehyde not included (value ) 185 µg/m3). **One point for toluene not included (value ) 622 µg/m3). the kitchen was open to the dining room, showed some association, most likely due to the use of chlorinated water and cleaning products. In the correlation analysis, we saw relationships that help us distinguish sources between smoking and nonsmoking restaurants. Due to sample size restraints, we could not compare smoking restaurants separately with nonsmoking, therefore we compared nonsmoking to all restaurants. The BTEX and styrene were not highly (>0.65) correlated with 1,3-butadiene in nonsmoking dining areas but were correlated if smoking restaurants were included in the analysis. The opposite was true for the BTEX and styrene with MTBE and 2,2,4-TMP, our mobile source markers. Combined with the regression results, when present, smoking becomes the dominant indoor source for these compounds, but otherwise, the outdoor street environment has a greater influence. These results are consistent with tracer and exhaled breath from smoker studies which showed that tobacco smoke is the major contributor to indoor levels of formaldehyde, benzene, 1,3butadiene, and styrene (57-59). It is also possible that there is some contribution from cooking as these compounds are also associated with combustion.
Other Comparisons. We compared some of the results from BEAM stores and dining to those of Kim et al.’s department stores and restaurants and pubs in Birmingham, U.K., and Guo et al. and Lee et al.’s shopping malls and restaurants in Hong Kong (34-36). Since these studies reported arithmetic means, we assumed that our geometric mean concentrations would be slightly lower, all else being equal. Toluene levels were similar in all studies, indicating similar sources. Ethylbenzene and m,p-xylene were higher in BEAM stores, although this is driven by the higher levels in the multipurpose, housewares, and hardware stores, which were not considered in previous studies. Department stores in Kim’s study and shopping malls in Lee’s study have VOC levels similar to those of the BEAM department stores. Kim’s department stores had 1,4-dichlorobenzene levels over ten times that in BEAM, but this has been shown to be a variable compound, and may be used or sold as a moth repellent or air freshener in the U.K. department stores and but not in the U.S. department stores we evaluated. Benzene levels were similar in the Kim and Lee studies (around 10 µg/m3) while the BEAM geometric mean was about 2 µg/m3 across all stores, which was closer to the range of concentrations in VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6909
Guo’s study. One explanation for this could be that gasoline benzene levels in the United States are, on average, about 1.5%, while in Hong Kong and the U.K., the maximum permitted level of benzene was 5% before 2000, and reduced to 1% thereafter. The Kim and Lee studies were performed earlier (1999-2000), when fuels may still have still had higher benzene content. The Lee malls were in areas of high vehicle density and some had attached car parks and bus stations, which may contribute to the higher benzene concentrations. In dining, most VOC concentrations were similar, although benzene was as much as 10 times higher in Kim’s restaurants and pubs and about twice as high in Lee’s restaurants, potentially related in part to the frequency and density of smoking, differences in ventilation, and contributions from the outdoors, in these establishments relative to the restaurants we sampled. Limitations. Using composite sampling, we were able to sample a large number of stores, while keeping within logistical and budgetary constraints. By combining sites, the variability of the estimate of central tendency can be decreased with respect to the number of sampling tubes or cartridges used; however, the estimate of the variability of the distribution becomes more imprecise. We had several repeated measures, but because these were not done on all stores measured, and they were only done twice, for 1 h each, they may not provide enough information to determine the true within-store variability. To better quantify the variability in stores, particularly the frequency of outlying values, more repeated samples and samples at more locations should be done. Another limitation faced in this study method is that while there are several theoretical composite schemes possible, there are tradeoffs. Logistically, it can be challenging to sample many sites in 1 day, or to spend the amount of time required in a given location. We were also limited in our ability to do a predictive analysis using linear regression models. There were too many overlapping products in many of the store types, making it difficult to distinguish specific product sources. In restaurants, our measurement of potential predictors was more qualitative, thus leaving room for error in measurement. We also could not logistically collect a simultaneous outdoor sample or measure air exchange. However, as mentioned previously, we had data on transportation VOC concentrations, which were much lower than most of the store and some restaurant concentrations and most likely not a large contributor to indoor concentrations. Future studies would benefit from refined measurement of some of these variables. A detailed study of selected store types, particularly those with high concentrations, would also add value to this area. The BEAM study found that retail stores and restaurants can have VOC concentrations many times that of the outdoors and home environments. Although the amount of time spent in these microenvironments is less than that spent at home, people are potentially exposed to levels of concern in these places, particularly cumulatively over many visits. In addition, those who work in these microenvironments spend more time there than others, and may face an increased risk of health effects. Combining these measured concentrations with information from time-activity studies can help us better understand total population exposure to VOCs.
Acknowledgments We acknowledge funding through the American Chemical Council. Additional support was provided by the International Society for Exposure Analysis’ young investigator grant for Dr. Bennett. Thanks to R. Dodson, S. Chang, J. Maltese, S. Fang, C. Lunn, Z. Dong, C. Meng, T. Goldberg, M. Penterson, and J. Allen for assistance in the field. Additional thanks to M. Davey and J. Vallarino for field and technical support, 6910
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 22, 2006
and to B. Labrecque, S. Forsberg, and R. Weker for laboratory analysis.
Supporting Information Available Details on composite sampling methodology; sampling and analytical methodology; dining regression results; tables for geometric means, coefficient of variation, min/max, and QA/ QC. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Williams, P. L.; James, R. C.; Roberts, S. M. Principles of Toxicology: Environmental and Industrial Applications, 2nd ed.; John Wiley and Sons, Inc.: New York, 2000. (2) Woodruff, T. J.; Caldwell, J.; Cogliano, V. J.; Axelrad, D. A. Estimating cancer risk from outdoor concentrations of hazardous air pollutants in 1990. Environ. Res. 2000, 82 (3), 194-206. (3) Payne-Sturges, D. C.; Burke, T. A.; Breysse, P.; Diener-West, M.; Buckley, T. J. Personal exposure meets risk assessment: A comparison of measured and modeled exposures and risks in an urban community. Environ. Health Perspect. 2004, 112 (5), 589-598. (4) Wallace, L. A.; Pellizzari, E. D.; Hartwell, T. D.; Sparacino, C. M.; Whitmore, R. W.; Sheldon, L.; Zelon, H.; Perritt, R. L. The TEAM study: Personal exposures to toxic substances in air, drinking water, and breath of 400 residents of New Jersey, North Carolina, and North Dakota. Environ. Res. 1987, 43, 290-307. (5) Clayton, C. A.; Pellizzari, E. D.; Whitmore, R. W.; Perritt, R. L.; Quackenboss, J. J. National Human Exposure Assessment Survey (NHEXAS): Distributions and associations of lead, arsenic and volatile organic compounds in EPA Region 5. J. Exposure Anal. Environ. Epidemiol. 1999, 9, 381-392. (6) Gordon, S.; Callahan, P.; Nishioka, M.; Brinkman, M.; O’Rourke, M.; Lebowitz, M.; Moschandreas, D. Residential environmental measures in the National Human Exposure Assessment Survey (NHEXAS) pilot study in Arizona: Preliminary results for pesticides and VOCs. J. Exposure Anal. Environ. Epidemiol. 1999, 9, 456-470. (7) Sax, S. N.; Bennett, D. H.; Chillrud, S. N.; Kinney, P. L.; Spengler, J. D. Differences in source emission rates of volatile organic compounds in inner-city residences of New York city and Los Angeles. J. Exposure Anal. Environ. Epidemiol. 2004, 14 Suppl 1, S95-109. (8) Edwards, R. D.; Jurvelin, J.; Saarela, K.; Jantunen, M. VOC concentrations measured in personal samples and residential indoor, outdoor and workplace microenvironments in EXPOLISHelsinki, Finland. Atmos. Environ. 2001, 35, 4531-4543. (9) Edwards, R. D.; Jurvelin, J.; Koistinen, K.; Saarela, K.; Jantunen, M. VOC source identification from personal and residential indoor, outdoor, and workplace microenvironment samples in EXPOLIS-Helsinki, Finland. Atmos. Environ. 2001, 35, 48294841. (10) Jurvelin, J.; Edwards, R. D.; Vartiainen, M.; Pasanen, P.; Jantunen, M. Residential indoor, outdoor, and workplace concentrations of carbonyl compounds: Relationships with personal exposure concentrations and correlation with sources. J. Air Waste Manage. Assoc. 2003, 53, 560-573. (11) Wallace, L. A.; Pellizzari, E. D.; Hartwell, T. D.; Davis, V.; Michael, L. C.; Whitmore, R. W. The influence of personal activities on exposure to volatile organic compounds. Environ. Res. 1989, 50 (1), 37-55. (12) Ozkaynak, H.; Ryan, P. B.; Wallace, L. A.; Nelson, W. C.; Beher, J. V. Sources and emission rates of organic chemical vapors in homes and buildings. Indoor Air ’87: Proceedings of the 4th International Conference on Indoor Air Quality and Climate, Institute for Water, Soil, and Air Hygiene, Berlin (West), Germany, 1987. Air Quality and Climate, West Berlin, Germany, 1987. (13) Anderson, M. J.; Miller, S. L.; Milford, J. B. Source apportionment of exposure to toxic volatile organic compounds using positive matrix factorization. J. Exposure Anal. Environ. Epidemiol. 2001, 11 (4), 295-307. (14) Etkin, D. S. Volatile Organic Compounds in Indoor Environments, Cutter Information Corp: Arlington, MA, 1996. (15) Nazaroff, W. W.; Weschler, C. J. Cleaning products and air fresheners: Exposure to primary and secondary air pollutants. Atmos. Environ. 2004, 38, 2841-2865. (16) Sack, T. M.; Steele, D. H.; Hammerstrom, K. A.; Remmers, J. A survey of household products for volatile organic compounds. Atmos. Environ. 1992, 26A (6), 1063-1070.
(17) Tichenor, B. A.; Mason, M. A. Organic emissions from consumer products and building materials to the indoor environment. J. Air Pollution Control Assoc. 1988, 38, 264-268. (18) Wallace, L. A.; Pellizzari, E. D.; Leaderer, B. P.; Zelon, H.; Sheldon, L. Emissions of volatile organic compounds from building materials and consumer products. Atmos. Environ. 1987, 21 (2), 385-393. (19) Kleipeis, N. E.; Nelson, W. C.; Ott, W. R.; Robinson, J. P.; Tsang, A. M.; Switzer, P.; Behar, J. V.; Hern, S. C.; Engelmann, W. H. The National Human Activity Pattern Survey (NHAPS): A resource for assessing exposure to environmental pollutants. J. Exposure Anal. Environ. Epidemiol. 2001, 11 (3), 231-252. (20) Chan, C. C.; Ozkaynak, H.; Spengler, J. D.; Sheldon, L. Driver exposure to volatile organic compounds, CO, ozone, and NO2 under different driving conditions. Environ. Sci. Technol. 1991, 25, 964-972. (21) Chan, C. C.; Spengler, J. D.; Ozkaynak, H.; Lefkopoulou, M. Commuter exposures to VOCs in Boston, Massachusetts. J. Air Waste Manage. Assoc. 1991, 41 (12), 1594-1600. (22) Batterman, S. A.; Peng, C.-Y.; Braun, J. Levels and composition of volatile organic compounds on commuting routes in Detroit, Michigan. Atmos. Environ. 2002, 36, 6015-6030. (23) Lawryk, N. J.; Lioy, P. J.; Weisel, C. P. Exposure to volatile organic compounds in the passenger compartment of automobiles during periods of normal and malfunctioning operation. J. Exposure Anal. Environ. Epidemiol. 1995, 5 (4), 511-531. (24) Lawryk, N. J.; Weisel, C. P. Concentrations of volatile organic compounds in the passenger compartments of automobiles. Environ. Sci. Technol. 1996, 30 (3), 810-816. (25) Rodes, C.; Sheldon, L.; Whitaker, D.; Clayton, A.; Fitzgerald, K.; Flanagan, J.; DiGenova, F.; Hering, S.; Frazier, C. Measuring concentrations of selected air pollutants inside California vehicles; Report 95-339; California Air Resources Board, South Coast Air Quality Management District, Sacramento, CA, 1998. (26) Weisel, C. P.; Lawryk, N. J.; Lioy, P. J. Exposure to emissions from gasoline within automobile cabins. J. Exposure Anal. Environ. Epidemiol. 1992, 2, 79-96. (27) Jo, W.; Choi, S. Vehicle occupants’ exposure to aromatic volatile organic compounds while commuting on an urban-suburban route in Korea. J. Air Waste Manage. Assoc. 1996, 46, 749-754. (28) Lau, W.; Chan, L. Commuter exposure to aromatic VOCs in public transportation modes in Hong Kong. Sci. Total Environ. 2003, 308, 143-155. (29) Lioy, P. J.; Weisel, C. P.; Jo, W. K.; Pellizzari, E.; Raymer, J. H. Microenvironmental and personal measurements of methyltertiary butyl ether (MTBE) associated with automobile use activities. J. Exposure Anal. Environ. Epidemiol. 1994, 4 (4), 427441. (30) Daisey, J. M.; Hodgson, A. T.; Fisk, W. J.; Mendell, M. J.; Ten Brinke, J. Volatile organic compounds in twelve California office buildings: Classes, concentrations, and sources. Atmos. Environ. 1994, 28 (22), 3557-3562. (31) Environmental Health and Engineering. A screening-level ranking of organic chemicals and metals found in indoor air, Report 12278; Environmental Health and Engineering, Inc., Newton, MA, 2002. (32) Sheldon, L. S.; Handy, R.; Hartwell, T.; Whitmore, R. W.; Zelon, H.; Pellizzari, E. D. Indoor Air Quality in Public Buildings, Vol I; Report EPA 600/6-88/009a; U.S. Environmental Protection Agency: Washington, DC, 1988. (33) Sheldon, L. S.; Eaton, C.; Hartwell, T. D.; Zelon, H.; Pellizzari, E. D. Indoor Air Quality in Public Buildings, Vol II; Report EPA 600/6-88/009b; U.S. Environmental Protection Agency: Washington, DC, 1988. (34) Lee, S.-C.; Guo, H.; Li, W.-M.; Chan, L.-Y. Inter-comparison of air pollutant concentrations in different indoor environments in Hong Kong. Atmos. Environ. 2002, 36, 1929-1940. (35) Guo, H.; Lee, S. C.; Chan, L. Y.; Li, W. M. Risk assessment of exposure to volatile organic compounds in different indoor environments. Environ. Res. 2004, 94 (1), 57-66. (36) Kim, Y. M.; Harrad, S.; Harrison, R. M. Concentrations and sources of VOCs in domestic and public microenvironments. Environ. Sci. Technol. 2001, 35, 997-1004. (37) Rohlf, F.; Akcakaya, H.; Ferraro, S. Optimizing composite sampling protocols. Environ. Sci. Technol. 1996, 30 (10), 28992905.
(38) Lancaster, V.; Keller-McNulty, S. A review of composite sampling methods. J. Am. Stat. Assoc. 1998, 93 (443), 1216-1230. (39) El-Baz, A.; Nayak, T. K. Efficiency of composite sampling for estimating a lognormal distribution. Environ. Ecol. Stat. 2004, 11 (3), 283-294. (40) Gilks, W. R.; Richardson, S.; Spiegelhalter, D. J. Markov Chain Monte Carlo in Practice; Chapman and Hall: London, 1996. (41) Gilks, W. R.; Thomas, A.; Spiegelhalter, D. J. A language and program for complex Bayesian modeling. Statistician 1994, 43 (1), 169-177. (42) Spiegelhalter, D. J.; Thomas, A.; Best, N. G.; Lonn, D. Winbugs version 1.4 User Manual. MRC Biostatistics Unit, Cambridge, UK and Imperial College, London, UK. (43) Fitzmaurice, G. M.; Laird, N. M.; Ware, J. H. In Applied Longitudinal Analysis; Wiley Series in Probability and Statistics: Hoboken, NJ, 2004; pp 76-78. (44) Kelly, T. J.; Smith, D. L.; Satola, J. Emission rates of formaldehyde from materials and consumer products found in California homes. Environ. Sci. Technol. 1999, 33 (1), 81-88. (45) Brown, S. K. Chamber assessment of formaldehyde and VOC emissions from wood-based panels. Indoor Air 1999, 9, 209215. (46) Morrison, G. C.; Nazaroff, W. W. Ozone interactions with carpet: Secondary emissions of aldehydes. Environ. Sci. Technol. 2002, 34 (23), 4963-4968. (47) Reiss, R.; Ryan, P. B.; Koutrakis, P.; Tibbetts, S. J. Ozone reactive chemistry on interior latex paint. Environ. Sci. Technol. 1995, 29 (8), 1906-1912. (48) Weschler, C. J.; Hodgson, A. T.; Wolley, J. D. Indoor chemistry: Ozone, volatile organic compounds, and carpets. Environ. Sci. Technol. 1992, 26 (12), 2371-2377. (49) Weschler, C. J.; Shields, H. C. Potential reactions among indoor pollutants. Atmos. Environ. 1997, 31 (21), 3487-3495. (50) Tichenor, B. A.; Sparks, L. A.; White, J. B.; Jackson, M. D. Evaluating sources of indoor air pollution. J. Air Waste Manage. Assoc. 1990, 40 (4), 487-492. (51) Knoppel, H.; Schauenburg, H. Screening of household products for the emission of volatile organic-compounds. Environ. Int. 1989, 15 (1-6), 413-418. (52) Destaillats, H.; Lunden, M. M.; Singer, B. C.; Coleman, B. K.; Hodgson, A. T.; Weschler, C. J.; Nazaroff, W. W. Indoor secondary pollutants from household product emissions in the presence of ozone: A bench-scale chamber study. Environ. Sci. Technol. 2006, 40, 4421-4428. (53) National Toxicology Program. Report on Carcinogens, 11th ed.; U.S. Department of Health and Human Services: Washington, DC, 2005. (54) Weisel, C. P.; Chen, W. J. Exposure to chlorination by-products from hot water uses. Risk Anal. 1994, 14 (1), 101-106. (55) Reiss, R.; Ryan, P. B.; Tibbetts, S. J.; Koutrakis, P. Measurement of organic acids, aldehydes, and ketones in residential environments and their relation to ozone. J. Air Waste Manage. Assoc. 1995, 45, 811-822. (56) Salthammer, T.; Fuhrmann, F.; Kaufhold, S.; Meyer, B.; Schwarz, A. Effects of climatic parameters on formaldehyde concentrations in indoor air. Indoor Air 1995, 5 (2), 120-128. (57) Hodgson, A. T.; Daisey, J. M.; Mahanama, K. R. R.; TenBrink, J.; Alevantis, L. E. Use of volatile tracers to determine the contribution of environmental tobacco smoke to concentrations of volatile organic compounds in smoking environments. Environ. Int. 1996, 22 (3), 295-307. (58) Gordon, S. M.; Wallace, L. A.; Brinkman, M. C.; Callahan, P. J.; Kenny, D. V. Volatile organic compounds as breath biomarkers for active and passive smoking. Environ. Health Perspect. 2002, 110 (7), 689-698. (59) Wallace, L.; Pellizzari, E. D.; Hartwell, T. D.; Perritt, R. L.; Ziegenfus, R. Exposures to benzene and other volatile organic compounds from active and passive smoking. Arch. Environ. Health. 1987, 42 (5), 272-279.
Received for review January 27, 2006. Revised manuscript received August 4, 2006. Accepted August 9, 2006. ES060197G
VOL. 40, NO. 22, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6911
