Document B8nkg1DzboOXomXgGjo7Mkwwm

Benzene exposure in the petroleum refining industry ROBERT C. SPEAR, STEVE SELVIN, JANE SCHULMAN and MARClE FRANCIS Northern California Occupational Health Center, School of Public Health: University of California, Berkeley, California 94720 Introduction In the latter part of 1984, the authors were asked by the American Petroleum Instituteto conduct a study of benzene exposure data available from some of their member companies. These data had been collected by company hygienists over the years 1978to 1984.The goals of the study were to: 1. Characterize the distribution of exposures within work operations and job categories. 2. Estimate the proportions of individuals exposed to benzene at or above levels ranging from 0.5 to 2.0 ppm. 3. Investigate the usefulness and suitabilityof specific statistical distributions as summary descriptors of exposure, e.g., normal, lognormal, etc. 4. Investigate intrapersonalvariation as a proportion of total variation in exposure levels within job categories. 5. Assess the implications of the resuits as they relate to air sampling strategies and the definition of compliancewith OSHA standards. These goals can be broadly classified into descriptive, analytical, and policyrelated aspects of the study. This paper deals first with a description of the benzene exposure levels in various parts of the industry during the time period in question. An analytical framework is presented which was found to be necessary to characterize exposures adequately and to deal with issues of exposure variability within and between workers. The policy implicationsof these Benzeneexposure data, Submitted by nine petroleum refining companies, were studiedwith the general objectiveof characterizingthe distribution of exposures within work operations and job categories. The data were collected by company hygienists over the period of years between 1978 and 1984. All measurements were of personal exposures determined using charcoal tubes or organic vapor monitors. Of the 123 location and unit specific job groups studied, most eighthour time-weighted average (TWA) exposures were below 1.0 ppm. For some groups, however, the variability in exposure was such that ten percent or more of exposuresexceeded 1.0 ppm. For nineteenof the most highly exposed groups, sufficient data were available to study the variability in exposure associated with individualworkers versus the common work environment. Results indicate that there were some cases in which variability in exposure was mostly asso- ciated with the environment and others in which it was mostly associated with differences between workers, but it was most often approximately evenly split betweenthe two sources. This analysis providesguidance in determiningwhether control strategies should be targeted at the work environment or at work requirements or practices of individual workers within the group. Some shortterm exposure data were included for study. It was found that most 15-minute TWA exposures were less than 1.0 ppm, but again, there were groups with highly variable exposure with some measurements in excess of 5 ppm. These were often in jobs involving loading and unloading operations of barges or tanker trucks. Spear, R.C.; Selvin, S.; Schulman, J.; Francis, M.: Benzene exposure in the petroleum refining industry. Appl. Ind. Hyg. 2355-163; 1987. analyses, in thecontext of item5 above, will be treated in a subsequent paper. The data base The data base of Dersonal exposure measurements which was available for analysis was comprised principally of eight-hourtime-weighted average W A ) values, although some short-term exposure measurementswere submitted. An early decisionwas made to focus on the eight-hour data to avoid the additional variability that would be introduced by different samplingdurations."' All measurements were collected with either charcoal tubes or passiveorganic vapor monitors. Data were submitted by nine petroleum refining companies. The measurementswere originally collected for many different reasons by the company hwienists; for the majority of the data, itwas impracticalto consider separately those samples collected under routine conditions versus those collected to characterize exposures during unusual circumstances. In that regard, respiratory protection was used in some of these exposure situations although data on the frequency of respirator use were not analyzed. In general, each measurement was classified by worker identifier, location, unit, and job. The fact that more than one observation was APPL IND. HYG. VOL 2, NO. 4 * JULY 1987 0882-8032/87m40155-09$2.50 0 ACGIH 1987 155 I ' collected on most workers presented the possibility of estimating the varia- bility in cxposure due to the common work environment versus that associ- ated with a particular individual. Early inthe study, considerableeffort w;1s given to working with the company hygienists in an attempt to define job categories that would at least approximate equal exposure groups, i.e., groups of workers whose exposures could be expected to be similar based on the natureand location of their jobs. In some companies, the specificity of job titles made it necessary to pool several job categories into a single exposure group for analysis. In the end, all job groups were specific to a particular refinery, a unit within the refinery, and an operation within the unit. For example, one particular job group was comprised of all benzene production operators (24 persons and 58 measurements) on the benzene and cyclohexane unit at location 8. Descriptive results: eight-hour TWA measurements Once an understanding had been reached on how to specify job groups, the companies were asked to submit eight-hour exposure data for all such groups for which there were 30 or more measurements collected from ten or more workers. There were 123 groups which met these requirementsapproximately, with some havingas few as eight workers. These data are described by giving the percent of the measurements below the limits of detection and the fraction of measurementsobserved above 0.5, 1.0, and 2.0 ppm; these are values of potential regulatory significance. Table I contains the descriptive exposure data for groups where at least ten percent of measurements exceeded 0.5 ppm, i.e., data are reported for the most highly exposed groups. The columns of Table I, labeled P(X>0.5), P(X>l.O), and P(X>2.0), are the fractions of the total number of measurements above 0.5, 1.0, and 2.0 ppm, respectively. These values are referred to below as exceedance fractions. The limitsof detectionvary by sampling medium and by analytical method. Since these factors differed from company to company and over time within companies, detection limits were reported separately for each data submission. For 30 of the 123 groups, all of the measurementswere below 0.5 ppm. For many of these 123 groups, the fraction below the lim~itof detection was also quite high. On the other end of the scale, roughly15 percent of these groups have significant fractions of the measurements in excess of 1.0 ppm. The groups with relatively high exposures and with adequate numbers of measurements will be considered below in some detail. The high exposure groups are comprised of operators of production units within the refineries, bulk transfer or loading operations, and laboratory quality control activities. There was some interest in the exposure of maintenance workers, and a special effort was made to obtain data on this employee group. The data submittedwere eight-hour W A values and are summarized in Table II. However, the natureof maintenancework is such that these groups are not unit specific. Most, in fact, are only company specific so that there can be no pretense of uniformity of exposure. Nevertheless, it should be noted that only about five percent of the measurements exceed 1.O pprn, and a large percentage of the measurements are below detection limits. Descriptive results: short-term exposures An effort was made, late in the study, to obtain data on short-term benzene exposures. In general, short-term refers to exposures measured over intervals of 5 to 30 minuteswhen a particular task or work condition leads the hygienist to expect high transient exposures. The format for data collection and analysis was, as with the eight-hourdata, focused on unit and site specific job groups. In this case, we sought at least ten measurements per group, but no further constraints were imposed. The majority of data was submitted by one company, but there are some data from a total of four companies. Table Ill contains a summary of the data in the same format as was used for the eight-hour TWA data. The exceedance fractions are based only on a count of the number of measurements above each of the limits where each measurement is treated as an independent value. In all cases, the sampling medium was charcoal tubes. Because the sampling time was not uniform, the values were normalized to a 15-minute standard. If the measurementduration was less than 15 minutes, it was assumed, following OSHA compliance practice, that the hygienist making the measurement had concluded that the task was complete and no further exposure would have occurred. Hence, the total mass collected would not have increased had sampling continued for a longer pe- riod, and the 'IWA value was thereby adjusted downward by the ratio of the sampling duration to the 15-minute standard. If the duration exceeded 15 minutes, the TWA value was left unaltered under the assumption that exposure was uniform over time, the only possible choice in the absence of other information. As can be seen from Table 111, most of the short-term exposuresarevery low. However, as with the eight-hour data, there are some groups that show relatively high exceedance fractions even at 5 ppm. Those groups showing exceedance fractions of 0.1 or above at 5 pprn are involved in tank gauging, loading and unloading operations, barge transfer operations, and tanker truck loading. For some groups in Table Ill, groups 29 to 31 for example, the limits of detection were relatively high. This leads to high proportions below the detection limit, but also high proportions of the measurementsin excess of 0.5 ppm. These entries are not in error and simply indicate short sample duration and correspondingly high detection limit situations. Analytical framework The descriptive data leaves one with the impression that, to the extent that these data are representative of the industry, most operationsare associatedwith exposures below 1.O pprn most of the time. Some operations, however, have significant fractions of their exposures above 1.O ppm. To reach any more specific conclusion on the basis of these descriptive data alone i s difficult. Indeed, the second goal of the study, which was to estimate the proportions of individualswho were exposed to various benzene levels, cannot be addressed since these data relate to the proportionof measurementsrather than of individuals. What is needed is a framework to disentangle the individualworker's contribution to overall variability from those environmental factors that contribute to the variability in exposure common to the group. Oldham and Roach were apparently the first to apply analysis of variance .156 APPL /NO. HYG. VOL 2, NO. 4 JULY 1987 TABLE I Empiric estimates of exceeding certain limits Ippm) for groups of refinery workers exposed to benzene Group 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 Description Location A, Catalytic Cracker Refining Operators Location B, Ethylene Unit Benzene Production Operators Location B, Waste Water Treatment Refining Operators Location B, Benz. & Cyclohex. Unit Benzene Production Operators Location E, CRU Platformer Head Operator, Controlman Location E, CRU Platformer Misc. Operators Location E, Petrochemical Plant Head Operator, Controlman Location E, petrochemical Plant Misc. Operators Location F, Benzene Extraction Unit Operator Location F, Aromatics-East ACU Operator Location F, Aromatics Gauger/Operator Location F, Aromatics Column Operator Location F, Aromatics-West ACU Operator Location F, Aromatics Foreman Location F, Cumene-PhenoVAcetone Operator Location A, Cracking Department Still-Platformer, Refining Oper. Location A, Petrochemical Oept., CurnenelBenzene, Benz. Prod. Oper. Location S, Benzene TransfedMovement Operators Location T, Lab Unit Technical Location G, Main Deck, Transportation-Marine Location B, Bulk Oil Pumphouse Refining Operators Location B, Bulk Oil Pump Station Refining Operators Location B, Laboratory Staff Location E, Petrochemical Unit Location M, Lab QC Unit All Staff Location F, Dispatching Dock Dockman Location F, PhenoWAcetone Foreman Location F, Oil Recovery-Environmental, Operator Location 0,Platformer, Hydro-desulfurizer, Stillman, Refining Location Q, Unit D Stillman, Refining Location Q,Control Room Laborer, Yardman, Roustabout Location R, Compound Plant Yardman Location U, Maintenance Location U, Lab, Quality Control Location W, Receipt. Storage, and Movement Location X, Lab Unit, Technical Location Y, Catalytic Cracking Unit Outside Journeyman Location Y, Aromatics Recovery Unit, Outside Journeyman Location AA, Benz. Unit Operators Location AA, Ethylene Unit Feedstock Operators Location BB. Lab Tech Location CC, Lab Tech Location DD, Lab Tech Loca-tion II. Lab Tech \ MediunP ov ov ov ov CT CT CT CT ov ov ov ov ov ov ov ov ov CT CT CT ov ov ov CT CT ov ov ov CT CT CT CT CT CT CT CT CT CT CT CT CT CT CT CT Samples 202 170 83 58 55 66 80 161 124 97 113 98 79 89 177 90 59 48 55 49 33 45 37 96 38 45 116 35 56 42 36 36 37 52 55 35 34 34 58 30 46 48 106 44 Persons 39 44 22 24 16 22 20 38 23 25 33 25 19 17 25 18 20 24 18 n/ac 14 16 34 56 32 23 19 22 10 17 a 8 34 47 36 17 22 18 39 24 32 12 63 20 P(X>0.5) 0.01 0.09 0.37 0.41 0.13 0.20 0.28 0.35 0.26 0.20 0.16 0.24 0.11 0.24 0.24 0.10 0.49 0.33 0.23 0.61 0.36 0.18 0.35 0.11 0.37 0.36 0.14 0.11 0.14 0.14 0.1 1 0.11 0.14 0.19 0.49 0.11 0.12 0.50 0.31 0.10 0.11 0.60 0.21 0.11 WX>l.O) 0.00 0.01 0.13 0.14 0.09 0.17 0.20 0.25 0.12 0.11 0.04 0.07 0.06 0.11 0.12 0.02 0.19 0.25 0.14 0.55 0.12 0.13 0.08 0.08 0.21 0.22 0.08 0.ffi 0.00 0.05 0.00 0.w 0.03 0.10 0.25 0.03 0.09 0.26 0.16 0.10 0.00 0.50 0.07 0.02 P(X%?.Ol 0.00 0.01 0.06 0.03 0.05 0.14 0.10 0.17 0.03 0.07 0.00 0.01 0.01 0.03 0.08 0.00 0.05 0.19 0.02 0.4 1 0.00 0.07 0.03 0.05 0.13 0.13 0.04 0.03 0.00 0.m 0.m 0.00 0.03 0.03 0.20 o.m 0.09 0.09 0.03 0.00 0.m 0.27 0.03 0.00 K BUT 27.7 8.2 15.7 0.0 3.6 7.6 2.5 3.1 10.5 18.6 15.9 14.3 25.3 14.6 24.3 2.2 0.0 6.2 9.1 0.0 15.2 26.1 0.0 14.6 2.6 26.7 49.1 42.9 73.2 64.3 0.0 0.0 16.2 19.2 10.9 11.4 11.8 0.0 6.9 133 41.3 20.8 8.5 52.3 %dam, OV = Passive Organic Vapor Monitor. CT = Charcoal Tube gm = Below the limn of detectton %h= not applicable since there are no ID numbers I .{NO.HYG. VOL 2, NO. I JULY 1987 157 9 z 3 888 c 000 3 asla 000 a 888 000 roo-r=oz2oE 888888888 000000000 888Z88888 000000000 88RiZ85888 000000000 IO- P a 5" f f f E 5 158 -APPL /NO. HYG. VOL 2, NO. 4 JULY 1987 .APPL /NO. H E . VOL 2 NO. 4 JULY 1987 159 -TOTAL h (a) WITHIN tical expectations, estimated by the arithmetic means, of the lognormaldistributions; they are not the geometric means or the means of the logarithms. This i s a subtle, but important, point which can easily lead to confusion.) Consequently, the exposure distribution obtained by randomly selecting workers and randomly selecting days on which measurements were to be taken is also l~gnormal.(~'Thlaistter distribution is shown at the top of the figure and includes both the day-to-day and worker-to-worker components of variability in exposure. There is empirical evidence that these distributional assumptions are valid in many exposure situations, but the goodness of fit issue will be subsequently addressed in the context of the benzene data. In order to apply this model to the analysis of the benzene data, it is necessary to give its analytical formulation and specify the parameters that will be estimated from the data. To this end, let h(x) be the overall distribution of exposure levels (the distribution at the top of the figure), f(x(p) be the distribution of the exposure of an individual worker whose mean exposure is p (the distributions in the middle of the figure illustratingseveral values of p), and g(p) the distribution of means across the worker population (the distribution at the bottom of the figure). These three distributions are related by the expression: Figure l-bgnonnal exposure model. techniques to exposure data to investigate worker-to-worker versus day-today variability in exposure.(z'The concept is shown schematically in Figure 1. The idea is that each worker's exposure differs from day to day and that, over time, the exposure is best described by a distribution of eight-hour TWA measurements which will generally be different for each worker. These are the distributions shown in the middle of the figure. If these distributionsare not the same for each worker, then one`s view of the exposure of the group, based on a sample of exposure data of the sort presented above, obviously depends upon which workers were selected for measurement and how many times each was measured, hence the difficulty in interpreting the benzene data without directly addressing this issue. An analysis of variance model was adopted based on the proposition that the distribution of individual exposures was lognormal and that the variance of each of these individualdistributions is the same for each worker, but with differing mean levels of exposure. Because the individual variances are assumed identical, the only remaining element of the model to be specified concerns the distribution of mean values among workers. It i s not unreasonable to postulate that this distribution is also lognormal. This i s the distribution shown at the bottom of the figure. That is, if one were to determine the mean value of each worker's exposure distribution and if there were a large number of workers, the distribution of means would also be lognormal. (The means referred to here are the statis- To explore the properties of h(x) it is necessary to specify the parameters associated with two basic lognormal distributions. Specifically, g(p) is lognormal: geometric mean = GMB geometric standard deviation = GSDB f(x(p) is lognormal: geometric mean = GMW geometric standard deviation = GSDW Further, represent the means and variances of the normal distributions of log(F) as pa = log(GMB), uB = log(GSDB), and log(x) as pW = log(GMW) and uw= log(GSDW). Note that B refers to the distribution between individuals which is the distribution at the bottom of Figure 1. Sim- .160 APPL /NO.HYG. VOL 2, NO. 4 JULY 1387 ilady, W refers to within individual distributions and T to the total distribution, i.e., the distributions in the middle and at the top of Figure 1 , respectively. Under these conditions (g and f both lognormal), h(x)is also a lognormaldistribution with geometric mean equal to e)exp(b - % and geometric standard deviation exp(dut, + 4)T.his is not a particularly well known result, but one that is to be found in texts on the ~ubject.'~'Theexpected value of the exposure of a random individual selected from the distribution h(x) is exp(FB+ 1/2 &), which i s identicalto the mean of the distribution g(F).This can be seen in Table IV which contains an illustration of these three distributions for specific values of the lognormal parameters. The value of GSDT, for example, is found at the intersection of the GSD column and the total row and is equal to 3.666. It is the estimation and manipulation of these parameters that allows separate assessment of exposure due to individual differences (GSDB) versus those due to common environmental factors (GSDW). Highly sampled groups In order to estimate the parameters of the model, it is clear that the exposure data have to include multiple measurements on each worker and multiple workers per group. In the benzene data, we required, somewhat arbitrarily, that each group analyzedshould haveat least 45 eight-hour measurementson at least 15 individualswith an average of at least two measurements per individual and not more than 30 percent of the measurements below the limit of detection. These criteria provide sufficient numbers of observations to produce stable estimates of the lognormal parameters. Standard statistical procedures were used to estimate the model parameters.") These procedures provide consistent estimates while taking account of the fraction of the measurements below the limits of detection. Before presenting these results, let us deal with the goodness of fit issue; that is, the degree to which the model adequately summarizes the data. The right tail of the top, h(x), and bottom, g(p), distributions of Figure 1were chosen as the focal points, clearly the most critical part of any exposure distribution. In particular, the method of choice was to contrast the proportion of measurements above 1.O ppm predicted by the model TABLE IV Example of the relationship between distributional parameters in the lognormal analysis of variance model Distribution Parametan GM GSD c U Expectation Between = g(pj Within = flx(p) Total = h(xJ 2.000 2.500 1.573 3.000 2.000 3.666 0.693 0.916 0.453 1.099 0.693 1.299 3.657 3.178 3.657 TABLE V Empiric and parametric estimates for the proportion of samples exceeding 1.0 ppm Group 1 2 3 4 5 6 7 8 9 10 P(X>l.O) (empiricj 0.005 0.012 0.133 0.138 0.091 0.167 0.200 0.255 0.121 0.113 P(X>l.O) (parametric) 0.007 0.042 0.192 0.154 0.046 0.122 0.184 0.260 0.102 0.108 Group 11 12 13' 14 15 16 17 18 19 P(X>l.O) (empiricj 0.035 0.071 0.063 0.112 0.124 0.022 0.186 0.250 0.145 P(X>l.O) (parametric) 0.023 0.066 0.045 0.090 0.123 0.012 0.193 0.274 0.112 TABLE VI Empiric and parametric estimates for the proportion of means exceeding 1.0 ppm Gmup 1 2 3 4 5 6 7 8 9 P(C>l.O) (ompiric) 0.000 0.023 0.182 0.125 0.000 0.136 0.200 0.342 0.087 P(P>llJ (parametric) 0.000 0.023 0.263 0.172 0.039 0.127 0.192 0.315 0.099 Group 11 12 13 14 15 16 17 18 19 P(p>l.Oj P(p>l.O) (empiricj (parametric) 0.000 0.OOO 0.053 0.059 0.160 0.OOO 0.200 0.333 0.056 0.015 0.012 0.059 0.095 0.144 0.001 0.198 0.380 0.098 with that observed and to contrast the predicted proportion of individual means about 1.0 ppm with those observed in the data set. That is, the means and variances of the model were estimated from the data, and these estimates were used to predict the fraction of measurements or the fraction of individual means above 1.0 ppm. These estimates were then contrasted with the observed exceedances. These contrasts are presentedin Tables V and VI. These values are calculated using the rela- tionships in the previous section. For example, the values in Table V come from h(x) where: -p = qz > flog(l.0) IO~(CMT)J~IO~(CSDT)} where Z has a standard normal distribution. As can be seen from Tables V and VI, the observed and estimated exceedances are generally in close agreement. Where there are differences, it is not clear whether they are due to inadequacies of the model or problems inherent in the data. That some datarelated differences should exist i s not surprisingsince, as noted earlier, these data were collected for many reasons, and it is unlikely that considerations of representativeness and independence were high priority criteria. In other words, some caution should be exer- cised in attempting to form useful conclusions based on statistical analysis of APPL IND. HYG. VOL 2, NO. 4 . JULY 1987 181 - data that were not originally collected number is near 100 percent, it indicates be associatedwith high values of GSDT, with such analyses in mind. that the group was heterogeneouslyex- indicating that the total variability was The benzene data are right-skewed posed and that there were differences inflated in some of these groups by re- and bounded below by zero, both car- in the exposure of individual workers peated sampling of workers with the acteristics of the lognormal distribu- due to different work practices or to most variable exposures. tion. These facts, together with the data different tasks. A percent value of 100 For some of these groups, this issue presented in these tables, are the prin- corresponds to a GSDW of 1.00which, was pursued further and the possibility cipal evidence that can be offered on in turn, corresponds to zero variability was explored that these high variances the distributional questions in any rig- associatedwith the environment.As can were associated with nonroutine ex- orous sense. The quantity of indepen- be seen from Table VII, there are groups posure conditions. For Groups 5,6,7, dent data was not sufficient to allow a where the variability i s principally en- and 8, all associatedwith one company, more thorough evaluation of the dis- vironmental, e.g., Group 1 , and others it was possible to determine that mea- tributional properties of the observed where variability is totally between surements taken during turnaround and data. Nevertheless, the authors felt workers. In general, however, a some- inspection conditions were over- comfortable in adopting the lognormal what greater proportion of the total var- represented in the sample and contrib- model as a basis for subsequent anal- iability seems to be associated with dif- uted disproportionately to the variance yses. ferences between workers rather than estimates. Similarly, the most extreme Table VI1 contains the results of an with the common environment. value of GSDT, 9.62, comes from Group analysis of variance for the 19 groups. Three out of four cases in which the 18 and results from the dominance of The mean value reported in Table VI1 variability (GSDT) i s almost totally as- the datafrom only three individualswho is the estimatedarithmetic mean of both sociated with differences between have highly variable levels of exposure the total distribution and of the distri- workers are associatedwith groups from and were sampled repeatedly (r = bution of individual means. This value a single company. It was suspectedthat 0.402).As mentioned previously, these and the appropriate GSD allow one to this might be due, at least in part, to results underscore the sensitivityof the calculatethe geometric meanfor either the sampling strategy usedby this com- analysis to the manner in which data the total distribution, h(x), or the dis- pany. For example, workers might have are collected. If these data had resulted tribution of means, g(p). been preferentially selected for moni- from a random and independent sam- The percent column of Table VI1 gives toring becausetheir exposurewas highly plingprocess, one might conclude that the percent of the total variability that variable. To explore this issue, the cor- these high variance groups were com- is attributable to differences between relation between the number of mea- prised of workers who had very differ- workers as opposed to that associated surements per individual and the vari- ent exposure experience and that con- with the common environment. (This ance of those values for each group were trol of exposuresshould be focused at \ value is based on the variance of the calculated. Correlation coefficients are the individual level. As it stands, these logs of the measurements rather than given in Table VI1 under the column high variances seem to result largely the GSDs or the second moments of labeled r. As can be seen, there is a from the way in which data were col- the lognormal distributions.) If this clear tendency for high values of r to lected. TABLE VI1 Parameters of the lognormal model for benzene exposure and resutts from the oneway analysis applied to 19 specific job groups GSDB Grwp dcl 1 1.87 2 2.26 3 3.36 4 1.91 5 3.62 6 6.52 7 5.52 8 6.31 9 2.28 10 2.67 11 2.1 1 12 2.02 13 3.11 14 2.41 15 4.15 16 1.56 17 1.75 18 9.29 19 3.01 GSDW Yxlcl 2.83 2.52 3.42 1.54 1.98 1.00 1.00 1.26 2.01 2.44 1.69 1.97 1.66 2.08 2.22 1.73 1.54 -1.49 2.72 GSM h(xl 3.36 3.43 5.62 2.18 4.30 6.52 5.52 6.41 2.94 3.76 2.49 2.66 3.46 3.15 5.12 2.02 2.03 9.62 4.43 Percent 26.6 43.7 49.2 68.9 77.9 100.0 100.0 98.5 58.0 54.7 67.0 51.5 83.5 59.0 76.0 40.1 62.3 96.9 54.9 c: 0.09 0.28 0.96 0.67 0.24 0.68 0.23 2.24 0.49 0.48 0.26 0.26 0.32 0.46 0.61 0.26 0.73 6.07 0.44 1 0.041 0.106 0.065 0.101 0.275 0.327 0.316 0.311 0.030 0.182 0.051 0.088 0.145 0.103 0.130 0.031 0.016 0.402 0.281 Wit column is included so the entire parametric model may be reconstructed from the given data. 162 Conclusions From a descriptive point of view, it was found that average eight-hour TWA exposures in most job groups that were analyzed were below 1.0 ppm. However, in a number of cases, the variability was sufficiently high to result in 1.O ppm exceedance fractions of 0.1 or above. This finding is true for both the distribution of means, g(p), and the distribution of measurements,h(x), inthose cases where adequate data were available to partition the variability. Inasubsequent paper, the implications of this finding to health risk assessment and to compliance monitoring will be explored. Although the data were not adequate to allow a conclusive investigation of the goodness of fit of the lognormal model, this limited exploration of the issue gave no indication that the assumption was misleading in this application. In fact, the conceptual clarification arising out of the analysis of variance model is a principal result of APPL IND. HYG. VOL Z NO. 4 * JULY 1987 t k benzene study. If the data have been collected so as to be representativeof the exposure of the group, the lognormal model offers the hygienist an opportunity to identify the most promising strategy for exposure reduction. If, for example, it is found that there is little difference in the exposure of the workers in the group, then environmental controls are indicated. Conversely, large differences in the exposures of the individual workers suggests that an analysis of job tasks or of work practices may be more useful. Recommendations Incharacterizingthe exposureof agroup of workers to an airborne contaminant, sampling resources are most efficiently utilized if measurements are collected such that the tools of statistical infer- ence can be used in the analysis of the data. This implies that exposure measurements should be made with due regard for randomness and independence. If, on the other hand, sampling strategies are used which focus on workers with particularly high or variableexposures, then analyzingsuchdata as representative produces a distorted picture of the exposure of the group as awhole. This experiencehas shownthat by assuming fray the outset that each worker's exposure has both an individual and a common environmental component, one is led to an exposure model that appears to summarize the exposure data adequately and in a format that provides guidance in identifying strategies for diminishing exposures. While it cannot be recommended that this approach be applied to all expo- sure situations, it does provide a conceptual point of departure for a more detailed understanding of the factors underlying the variability in workplace exposures to toxic air contaminants. References 1. Spear, R.C.; Selvin, S.; Francis, M.: The Influence of Averaging Time on the Distribution of Exposures.Am. Ind. Hyg. AsSOC. J. 47:365 (1986). 2. Oldham, P.D.; Roach, S.A.: A Sampling Procedurefor MeasuringIndustrialDust Exposure. Br. J. Ind. Med. 9:112 (1952). 3. Aitchison, J.; Brown, J.A.C.: The Lognormal Distribution, pp. 110-111. Cambridge University Press, Cambridge (1976). Received 10122/86; review/decision 12/061(16; revision 2/1W7; approved VlW7 Answers to the "Action Level" questions 1. c 2. a 3. c 4. omission of water vapor, use of incorrect values for alveolar ventilation 5. c 6. a 7. "how safe" situations .APPL /NO. H E . WL 2 NO. 4 JULY 19111 163