United States Environmental Protection Agency Air and Energy Engineering Research Laboratory Research Triangle Park, NC 27711 Research and Development EPA/600/S8-91/210 Dec 1991 w EPA Project Summary Development of Alternate Performance Standard for Radon Resistant Construction Based on Short-term/Long-term Indoor Radon Concentrations Ashley D. Williamson, Susan E. McDonough, and Charles S. Fowler This report gives results of a study of short- and long-term variations In radon concentration In approximately 80 houses In Florida. The study In- volves year-long comparative sampling using the most common radon mea- surement technologies. This study, pro- viding the most detailed database of which we are aware, addresses the time variation of Indoor radon concentra- tions In a significant number of occu- pied houses having moderately el- evated radon concentrations. In these study houses, the degree of variation of radon varies roughly in proportion to the long-term mean concentration, with a coefficient of variation within a calendar quarter of approximately 25% of the quarterly mean, and a coefficient of variation within a year of approxi- mately 35% of the annual mean. This pattern of variability supports the use of multiplicative models to fit the varia- tion and to predict intervals of;confi- dence for long-term averages based on short-term measurements. This study Indicates a distinct seasonal effect on the average radon, with quarterly aver- ages relative to the annual average In- creasing in the order of spring (82%) < summer (93%) < fall (97%) < winter (123%). These models have been used to develop threshold values for the per- formance criteria of the proposed Build- Ing Standard for Radon-Resistant Con- struction for the State of Florida. This Project Summary was developed by EPA's Air and Energy Engineering Research Laboratory, Research Tri- angle Park,, NC, to announce key find- Ings of the research project that Is fully documented In a separate report of the same title (see Project Report ordering Information at back). Introduction Many studies have been conducted na- tionwide to determine the extent of el- evated indoor radon concentrations in the U.S. Most of these studies have employed short-term screening techniques, ranging from 1 to 90 days, using either open- faced or diffusion barrier charcoal canis- ters or alpha track detectors according to EPA protocols. Several factors complicate the relation between short-term measure- ments and long-term indoor radon con- centrations. Primarily, radon concentrations have been shown to vary considerably with time; diurnal and seasonal variations are prominent in many houses and sug- gestions of weekly or other periods have been made. Some of these variations clearly correlate with house construction or occupant behavior patterns, such as heating and air-conditioning equipment and usage patterns, and the use of natural or mechanical ventilation during mild weather. The physical factors affecting the entry of radon into buildings are understood, at least in principle. However, no general means of computing the effect of these factors on resulting levels of indoor radon has been demonstrated. Added to this uncertainty due to fluctuations in actual radon concentrations is a smaller mea- surement of uncertainty due to the radon measurement devices themselves. Each major measurement technique has techni- cal shortcomings and limitations in the Printed on Recycled Paper ------- possible sampling periods. This report gives results of a study of short- and long- term variations in radon concentration in about 80 houses in Florida. The study involves comparative sampling using the most common radon measurement tech- nologies during the past year. To our knowledge, it is the most extensive study of "its kind. Conclusions and Recommendations This study, providing the most detailed database of which we are aware, ad- dresses the time variation of indoor radon concentrations in a significant number of occupied houses having moderately el- evated radon concentrations. In these study houses, the degree of variation of radon varies roughly in proportion to the long-term mean concentration, with a co- efficient of variation within a calendar quarter of approximately 25% of the quar- terly mean, and coefficient of variation within a year of about 35% of the annual mean. This pattern of variability supports the use of multiplicative models to fit the variation and to predict intervals of confi- dence for long-term averages based on short-term measurements. These models have been used to develop threshold val- ues for the performance criteria of the proposed Building Standard for Radon- Resistant Construction for the State of Florida. This study indicates a distinct seasonal effect on the average radon, with quar- terly averages relative to the annual aver- age increasing in the order of spring (82%) < summer (93%) < fall (97%) < winter (123%). One recommendation for further study includes follow-up studies on the seasonal effect. The seasonal trends seen in these data are clearly beyond experi- mental uncertainty; they only reflect the trends within a single year. Follow-on stud- ies (using only quarterly alpha track de- tectors) over 1 or more additional years would allow'replication of the seasonal data and evaluation of the reproducibility of the trend in different years. Background and Approach The only known significant prior study of time variability of concentrations in Florida houses covered the period 1987- 88. This study included a year-long mea- surement program in 37 houses in the Gainesville, FL area. Short-term average radon was measured by charcoal cannis- ters deployed once a month in each house, and long-term average concentrations were measured using alpha-track detec- tors deployed for the 1- year study. Con- siderable variability was noted from month to month, and there were clear sugges- tions of a seasonal effect, with November- March elevated, April and May depressed, and the remaining months intermediate. On a quarterly basis, these results trans- lated into the relationship summer < spring < fall < winter. The data reduction meth- ods used in this study were developed from those of the Gainesville study. Since the goal of this project was to support a statewide building standard, the scope of this project was expanded in several ways beyond that of the Gainesville study: (1) this study was to be statewide in scope, so it included four regions ex- pected to span the climate and geological variations in the radon-prone portions of the state; (2) the study was more compre- hensive in the number of devices em- ployed, including candidate samples not in widespread use at the time of most of the earlier studies; and (3) the study was structured to provide exploration or con- trol for house structural and operational variables which can potentially affect ra- don entry. The short-/long-term study was initiated in November 1989. Originally, the project work plan called for 40 houses to be se- lected for the project. The houses were selected based on the characteristics iden- tified as common to Florida housing stock 'such as: - Single-family, single-level, slab-on- grade housing with forced air heating and cooling; - Low to moderate radon levels (2 to 20 pCi/l); - Unmitigated (although two previously mitigated houses were selected for comparison in Polk County); - Air handler characteristics: split be- tween houses with air handler inside building shell (closet) and outside shell (garage, attic); and - Natural ventilation: attempt to select about half of the houses which never use natural ventilation for cooling. Candidate houses were screened and 10 study houses selected in each of four regions in the state, including, Alachua, Dade, Leon, and Polk Counties. In Febru- ary 1990 the project increased in scope to include up to 20 more houses in each county. The same selection criteria were employed in identifying the additional houses for the study. All houses are single- story, single-family, slab-on-grade houses. Regional data were collected by the fol- lowing investigators: - Alachua County: C. E. Roessler, Uni- versity of Florida; - Dade County: Howard Moore, Florida International University; - Leon County: James Cowart, Florida State University; and - Polk County: Susan McDonough, Southern Research Institute. In order to develop a predictive relation- ship between short-term measurements and long-term (annual) average concen- trations, a variety of short- and long-term sampling devices were deployed in each study house. The devices selected and their deployment periods are: - Alpha-track detector (ATD) (deployed for 1 year); - Alpha-track detector (deployed for 3 months each; four per house); - Low-sensitivity Electret Passive Envi- ronmental Radon Monitor (EPL) (read on about 4-week intervals); - High-sensitivity Electret Passive En- vironmental Radon Monitor (EPS) (read on about 1- and 2- week inter- vals); - Seven-day Charcoal Canisters (CC7) (1 week per month per house); - Two-day Charcoal Canisters (CC2) (one 2-day deployment per month); and - Pylon AB-5 with Passive Radon De- tector (rotated between houses about 4 weeks per house). Results and Discussion Sampling was conducted in the study house set between November 1989 and early March 1991. The median quarterly radon concentration in the study houses is 3.7 pCi/l, with 35% between 2 and 4 pCi/l, 19% between 4 and 6 pCi/l, and 26% above 6 pCi/l. This distribution is desirable for the goals of the study in several ways. First, by minimizing the num- ber of measurements below 2 pCi/l (about 19% here), all devices were generally able to operate above their detection limits and avoid the complications of censored data. More significantly, most of the houses fall in the zone near 4 pCi/l in which greatest uncertainty exists in predicting from a short-term measurement whether the long- term average radon will be above or be- low the 4 pCi/l Department of Health and Rehabilitation Services (DHRS) standard. ft is instructive to observe the relation- ship between the quarterly mean radon concentrations and the standard deviation of the corresponding set of EPS measure- ments in that house and quarter. As is apparent from plots of these data, the standard deviation has a clear positive correlation with the mean and can be fit (r2 = 0.59) to the linear trend: STD = mean* (0.2466) with a standard error of 0.0065 for the constant of proportionality, and the intercept not significantly different ------- from zero. A corresponding plot using an annual averaging period gives regression parameters of r2 = 0.78, slope = 0.357 ± 0.013, and intercept not significantly dif- ferent from zero. The roughly linear corre- lation of the standard deviation with the mean is significant. This assumption is used in multiplicative models such as that of the Gainesville study and other fitting techniques that use logarithmic transfor- mations for stabilization of variance. Comparison of Study Devices As might be expected, the correlations between the data from different devices in the same house are high. Simple linear regressions for each pair of data sets were performed. Standard linear regressions show r2 values above 0.95, intercepts not significantly different from zero, and con- stants of proportionality ranging from 7% lower (for CC2 measurements) to 8% higher (ATD measurements) than the EPS averages. Thus, while some degree of scatter remains, the comparability of dif- ferent devices is high and well within the accuracy objectives for each device indi- vidually. Seasonal Trends , A key issue in the variability of radon measurements is the seasonal component of this variability. To the extent that radon in a structure varies with a short period (hours, days, or weeks), multiple short period measurements (multi-day) or single medium period measurements (weeks) can average the fluctuations and give good predictions of the long-term average. How- ever, to the extent that a systematic sea- sonal trend is present, increasing the num- ber or duration of short-term measure- ments can reach a point of diminishing returns unless the general form of the seasonal effect can be predicted by other means. Without such a priori knowledge of the seasonal trend, this trend defines a minimum level of uncertainty for estimates of the annual average by any short-term measurement strategy. In order to assess the seasonal trends in this study, quarterly average radon con- centrations in each house were normal- ized to the annual average for that house. The spring quarter radon is tower in es- sentially all houses, with a mean quarterly concentration of 82% of the annual aver- age. The effect is fairly consistent be- tween houses, with half the houses show- ing quarterly ratios between 65 and 90% of the annual. On the other hand, winter quarter radon was elevated in most houses (mean concentration 1.28 times the an- nual average), but the degree of this ef- fect varied considerably between houses (for winter the central half of the popula- tion extended from 1.06 to 1.45 times the annual mean). Both the elevated radon and greater house-to-house variability are evident in two winter seasons a year apart. The seasonal trends in radon concentra- tion seen in these data are qualitatively in accord with the results of the Gainesville study wrth the same winter/fall elevation and spring/summer minimum. The only difference in these results is the reversal of the spring/summer trend in the present study. Some of the features of the seasonal variation can be noted by inspection of the seasonal variability in individual houses. While the average radon trend follows the pattern spring < summer < fall < winter, most individual houses do not. Of 65 houses with complete data for four full quarters only 18 fall into a class which has a winter maximum and spring mini- mum. The most common class (25 mem- bers) shows the winter maximum and sum- mer minimum which is typical of other regions of the country (ironically, this pat- tern is dominant in Dade County). The third most abundant pattern (eight houses) shows a summer maximum and spring minimum. The remaining houses do not appear to fall into groups of any signifi- cance. Somewhat surprisingly, the aver- age coefficient of variance remains es- sentially constant through the four sea- sons in the range of 24-28% relative to the quarterly average. Prediction of Long-term Average from Short-term Data Several slightly different approaches to the prediction of long-term averages from short-term concentrations were investi- gated in the course of this study. The first, which was adapted from the Gainesville study, relies on the assumption that the relative variability of radon concentration is on the same order in all houses in the state (at least for houses in the 2-8 pCi/l range). This assumption is inherent in the use of radon concentrations only as nor- malized to the long-term concentration in the fitting process. Other conventional re- gression approaches were considered which incorporate long-term radon explic- itly as a variable and typically use addi- tional parameters. These models make the slightly different assumption that the absolute variability of radon in the houses in the 2-8 pCi/1 range in this study is representative of houses in the state. Since our data lies in this range, the two as- sumptions are effectively indistinguishable. All approaches were found to give similar results in this case, so the simplest model was used. In order to describe the selected ap- proach, we will use a simplified form of the linear effects model described above. First, we assume that we can apply a log- normal effects model; that is, that all ef- fects are multiplicative and that short-term measurements of radon concentration in each house vary about the long-term av- erage with a standard deviation propor- tional to this mean. We define the quanti- ties C°,= STi/LT,and A,j = In (C°|j), where in effect C° be- comes a dimensionless relative radon concentration and A is its logarithm. Our model becomes An = U + 3j+ 6|j where u = an overall mean of A,, a, a> a group mean of effect of any subgroups found to be signifi- cant, and e,j = random error (assumed normally distributed in the log-transformed variable system). In terms of measured variables, where » In(STj), and • In (LT,), as described previously. Thus, our model can also be written in the form used for the Gainesville study, X|, - Y, = u + a, + ey In the event that other groupings are not treated as significant (which seems justified except for the possibility of sea- sonal corrections), the a, term disappears. The simplest predictive assumption is that and that the residuals are normally distrib- uted. This is in essence the approach used for the Gainesville study. It can also be viewed as a very simple regressional ap- proach where only the intercept is fit. Using the methodology described above, the data in the present study were used for estimation of probability ranges for long- term average radon, given single short- term radon measurements. For a given pool of data of short- and long-term aver- age radon concentrations, the quantities u, VAR(u), and VAR(A) are calculated, where X^ and Y, are defined by house for each combination of sampler and sample period, A, is calculated as above, u is the mean of the quantity (A*) over all measurements (i and j), VAR (u) is the square of the standard error of u given by the variance (s2) of the sample A, divided by the number of samples N, and ------- VAR(A) is the within-house sample variance of A, as determined from standard ANOVA methods. These quantities were then used in a pre- dictive sense as follows. For any postu- lated long-term reference value LTR, the probability p that the long-term average radon will exceed LTR will fit the relation- ship Zp- [IntST/LTn) - U]/[VAR(0)+VAR(A)]1/2 where Zp« is the p quantile of the stan- dard normal distribution. Rearranging and redefining the probability, if we wish to find the short-term average corresponding to a given probability that the long-term average will not exceed a given reference long-term average (that is, an upper confi- dence limit), we compute the relation ST- LTn exp[0 - ^ [VARCOH-VARfA)]"2] Plots analogous to these relationships are used for the Gainesville data. These relationships can also be applied to any homogeneous subsets of the study pool. If quarterly data are evaluated, in principle the seasonal factor would be ab- sorbed into the bias factor Q and the VAR (A) for the quarterly population would ap- ply (recall that this value is smaller for our quarterly averages). These analyses are not included here, partly due to the arbi- trariness in defining quarterly boundaries. In order to compare the difference in the predictive strength of the different short-term sampling techniques used in this study, these calculations were ap- plied to the data for all short-term sam- plers. The EPS data were further subdi- vided, since these samplers were oper- ated over different time periods. Likewise, averages of continuous radon monitor data from a subset of the houses over three different data averaging periods were com- puted as a comparison. The data from the non-continuous samplers were further sub- divided into three sets based on house ventilation characteristics. The first analy- sis was performed on all 65 unmitigated houses which had complete data over the period from February 1990 to February 1991. A second calculation was run on the subset of 26 houses which never use natural ventilation (open windows) for cool- ing. A third calculation was performed with the closed houses and the eight houses which "rarely" opened their windows (nominally < 5% of the time). Table 1 contains upper confidence limit calcula- tions for these data sets for several prob- ability values. Comparison of the data shows very little difference between the three groups of houses. This suggests that the variability due to the use of natural ventilation status is relatively minor compared to the vari- Table 1. Threshold Short-term Radon Concentrations (in pCi/1) Corresponding to Differing Levels of Confidence That Long-term Average Does Not Exceed 4 pCi/l Device/days* 0.5 0.6 All Houses (70), Last 4 Qtrs Confidence 0.7 0.75 0.8 0.85 0.9 Device/days* 0.5 Closed Houses (26), Last 4 Qrtrs Confidence 0.6 0.7 0.75 0.8 0.85 0.9 Device/days* 0.5 Mostly Closed Houses (39), Last 4 Qrtrs Confidence 0.6 0.7 0.75 0.8 0.85 0.9 (*) crm = continuous radon monitor eps = short-term EPerm epl = long-term EPerm cc2 s open face charcoal canister (2 day) cc7 = diffusion barrier charcoal canister (7 day) 0.95 crm-1 crm-7 crm-1 4 eps-7 eps-14 epl-14 epl-28 cc2 cc7 3.82 3.98 4.00 3.81 3.46 3.75 3.62 3.66 3.77 3.54 3.79 3.83 3.44 3.12 3.34 3.32 3.27 3.42 3.25 3.58 3.66 3.08 2.78 2.95 3.02 2.90 3.08 3.10 3.48 3.57 2.90 2.61 2.75 2.87 2.72 2.91 2.95 3.36 3.46 2.71 2.44 2.55 2.71 2.52 2.73 2.77 3.23 3.35 2.51 2.25 2.33 2.53 2.32 2.53 2.57 3.07 3.21 2.27 2.03 2.08 2.32 2.08 2.30 2.30 2.86 3.02 1.96 1.75 1.77 2.05 1.77 2.01 0.95 eps-7 eps-14 epl-14 epl-28 cc2 cc7 3.81 3.58 3.67 3.48 3.67 3.84 3.43 3.26 3.29 3.18 3.30 3.52 3.08 2.94 2.92 2.89 2.95 3.20 2.90 2.78 2.74 2.75 2.77 3.04 2.71 2.61 2.55 2.59 2.58 2.86 2.50 2.43 2.34 2.42 2.38 2.68 2.26 2.21 2.11 2.21 2.15 2.46 I .96 .93 .80 .95 .85 2.17 0.95 eps-7 eps-14 epl-14 epl-28 cc2 cc7 3.81 3.55 3.67 3.64 3.68 3.83 3.46 3.22 3.29 3.36 3.31 3.52 3.13 2.90 2.92 3.08 2.95 3.21 2.95 2.74 2.74 2.94 2.77 3.05 2.77 2.56 2.55 2.78 2.58 2.89 2.58 2.38 2.34 2.62 2.38 2.71 2.35 2.16 2.11 2.42 2.14 2.49 2.05 1.88 1.80 2.16 1.84 2.21 ability from other causes. If this is gener- ally true, these results may be generally applicable to houses with a wide range of ventilation practices. The data collected and analyzed to date in the FRRP Alternate Performance Stan- dard project have been incorporated into thresholds in the recommended code cur- rently in the rule making process. In sum- mary, the assumptions and philosophy that have been used to develop the standard are: (1) The goal of a building standard is to reduce the long-term average (annual or longer) radon concentration in the build- ing to be occupied. (2) Short-term measurements in the building will have uncertainty due to (a) measurement accuracy of the device used and (b) variability of the indoor radon con- centration with time. Uncertainty due to the second effect can be reduced by in- creasing the measurement time. (3) A performance test must be com- pleted and the results known prior to oc- cupancy for practical enforcement of a construction performance standard. In view of the time pressures on the construction industry, the measurement period in a workable performance standard will prob- ably be a compromise between the sched- ule needs of the builder and the uncer- tainty of the radon measurement. (4) The radon standard set by DHRS is assumed to remain at 4 pCi/l. (5) The threshold for passing a short- term performance test should be conser- vative; i.e., low enough to ensure that ------- (within a confidence level to be deter- mined by the state) the building will not have a long-term average radon concen- tration in excess of the MRS standard if a short-term performance test gives results less than the threshold. (6) Thresholds of this type are being developed for several device/measurement period combinations, so that the builder may elect to use a shorter duration test with a lower pass/fail threshold in order to achieve the same confidence that the building will comply with the standard. (7) Similarly, the project data have been analyzed to allow the state to choose thresholds based on different levels of con- fidence according to its regulatory priori- ties and the standard ultimately to be set byHRS. (8) If the effects of the time of year on indoor radon concentration can be quanti- fied, an algorithm to account for seasonal effects can be built into the threshold cri- teria. If such an algorithm cannot be de- veloped, the variability due to season must be included in the total variability of radon measurements in determining the thresh- olds for all times of the year. (This ap- proach has been taken in the current rec- ommendations.) The code language incorporates the possibility of several combinations of de- vice and measurement period. No provi- sion for incorporating average seasonal variations in radon data is included, due to lack of sufficient data on the form of systematic seasonal differences. A. Williamson, S. McDonough, and C. Fowler are with Southern Research Institute, Birmingham, AL 35255-5305 David C. Sanchez is the EPA Project Officer (see below). The complete report consists of two volumes, entitled "Development of Alternate Performance Standard for Radon Resistant Construction Based on Short-term/Long-term Indoor Radon Concentrations": "Volume 1" (Order No. PB92-115 211/AS; Cost: $19.00, subject to change) is the technical report. "Volume 2" (Order No. PB92-115 229/AS; Cost: $19.00, subject to change) contains the appendices. Both volumes of this report will be available only from: National Technical Information Service 5285 Port Royal Road Springfield, VA 22161 Telephone: 703-487-4650 The EPA Project Officer can be contacted at: Air and Energy Engineering Research Laboratory U.S. Environmental Protection Agency Research Triangle Park, NC 27711 &U.S. GOVERNMENT PRINTING OFFICE: 1992 - 648-080/40118 ------- ------- ------- II Q) O g-? en 75' p- CD c (n CD CO O O m PA/600/S8-91 !\5 *T 0 If you do not wis detach, or copy t upper left-hand 8 -*" 2 *"' 0 •r O -i 2 o to . < o CD (D """ S' Q) (D It 2S C -i ^ "S o ^ CD O 01 X Q. m I10 3 *** w X ,-. m Please make all detach or copy, - left-hand corner • :;; D Q-S -t O CD O •£ W C to 3 •< ss- ~* 0) 3" 3 CD (O Q> » Q. W CL O 3 ^ W 31 _. CD 3 0) CD c _ • 2 ^ T3 n- « g > m c: 3 -i CD 41 IS CD 0> 2 * T3 -1 O CD O L> 3 n CD 00 3 CD l-¥ a_ CB % ffi O ------- |