Instructions for Using R-project Software for Analyzing Industrial Hygiene Data

(Including Data with Non-Detects and Non-parametric Data)

 

 

1.      Introduction:  These are instructions for generating industrial hygiene metrics using R routines and MS Excel for Windows.   MS Excel can be used to clean-up data, group the data, create text files for analysis and create tables and charts for final reports.  Any word processing, spreadsheet, or database software could be used for creating the text file.  The Statistical Analysis of Non-Detects (sand) package contains R routines that read text files and generate an output file containing metrics.  The metrics generated are those recommended by the AIHA[1] and are interpreted and used as described in the book. 

2.      You should have already gone to http://www.csm.ornl.gov/esh/statoed/ and followed the instruction for installing R and the “sand” package.  Now you can use R to analyze data in a text file.  The text file must have at least two columns.  One column contains a value and the second a 0 or 1 for non-detect or detect.  Each column must have a one-word heading.  The following describes one way to create the text file using MS Excel. 


3.      This example uses data from Table IV.3 of the AIHA book and is also the example data in the file aihand.txt included in the download from the “statoed” web site.  The first column “Monitoring Data (mg/m3)” contains a mix of values and the text symbol “<”.  The “Value” and “Detected 0=No 1=Yes” columns were created using a variety of Excel text editing and logic functions.  For example the Excel logic function  =IF(LEFT(A2,1)="<",VALUE(REPLACE(A2,1,1,0)),A2)  removes the “<” symbol when it appears. While not important for a small file like this, these editing functions are very helpful when cleaning-up large data sets. 





4.      Columns F and G contain the values and flags that you need to convert to a text file.  Copy the two columns; open a new file, select “paste special”, and “values”.

 

 

 

 

 


5.      In this case Column B has a 3 word heading “Detected 0=No 1=Yes” and this has to be changed to a one word heading such as “Detected” or “Flag”.    The next step is to close the file and save in the rmain folder as a tab delimited text file. Click through the screens warning of the loss of formatting etc. 

 

 

 


6.      Open the R console by double clicking on the icon in the “rmain” folder you created.  Type in the command “aihand<-readss("aihand",L=5)”.  The L=5 is the value of the OEL being used to interpret data.  Hit “Enter” and the file is read and analyzed.  Once you have typed in commands, the up and down arrows toggle through the commands you have used.  If you “Save Workspace Image” at the end of the R session, the commands will be saved.  With a large dataset, one often will group data into subsets based on some variable (location, time, individual, etc) and create several text files for analysis.  Use the up arrow, edit the file name command line (i.e. Bldg1<-readss("Bldg1",L=5), Bldg2<-readss("Bldg2",L=5), etc.)  When the analyses are complete the prompt returns. 

 


7.      R creates a new comma delimited text file “aihandout.csv” that contains the metrics.  This can be opened with MS Excel and the two columns can be copied and pasted into spread sheet you will be using for further analysis and report writing.

 

8.      The “readss” command generates the following metrics.  The industrial hygienist chooses those that help interpret the data.  Mean and confidence intervals are useful for decisions on exposure groups and constructing job and exposure matrices.  Upper tolerance limits and percent exceedance are useful for determining compliance and other day-to-day risk management decisions.  Parametric and non-parametric versions of each are included.

Label                      Metric              Glossary                                              
mu                          0.925               Maximum likelihood estimate (MLE) of mean of the log
                                                      transformed data  (log of GM)
se.mu                     0.099               Estimate of the standard error of mu

sigma                      0.37                 MLE of standard deviation of log transformed data
                                                      (log of GSD)

se.sigma                 0.079               Estimate of standard error of sigma

GM                        2.522               MLE of geometric mean

GSD                      1.447               MLE of geometric standard deviation

EX                         2.7                   MLE of the EX the (arithmetic) mean

LCLa_95               2.26                 95% Lower Confidence Limit (LCL) for EX

UCLa_95               3.226               95% Upper Confidence Limit (UCL) for EX

KMmean                2.773               Kaplan-Meier (KM) Estimate of EX

KM.LCL               2.29                 95% LCL for KM EX

KM.UCL               3.257               95% UCL for KM EX

KM.se                   0.269               Standard Error of KMmean

Xp.obs                   4.75                 Observed 95th Percentile of data

Xp                         4.633               MLE of 95th Percentile

Xp.LCL                 3.521               MLE of 95% LCL for Xp

Xp.UCL                 6.096               MLE of the 95% Upper Tolerance Limit (UTL) of Xp

NpUTL                  NA                  Nonparmetric estimate of the 95% UTL of Xp. 

Maximum               5.5                   Largest value in the data set

NonDet%               20                    The percent of Xs that are left censored

n                            15                    The number of observations in the data set

Rsq                        0.969               Square of correlation for the data and standard log normal

m                           12                    The number of detected Xs

f                             3.208               MLE of the percent exceeding the specified limit L

f.LCL                     0.396               MLE of 95% LCL for f

f.UCL                    14.767             MLE of 95% UCL for f

fnp                         6.667               Nonparametric estimate of f for limit L

fnp.LCL                 0.341               Nonparametric estimate of 95% LCL for f

FnUCL_95            27.94               Nonparametric estimate of 95% UCL for f
m2logL                   41.3044           -2 times the log-likelihood function
L                            5                      L is specified limit for the percent exceeding; e.g., the OEL
        p                           0.95                 percentile for UTL p-gamma 
gam                        0.95                 one-sided confidence level gamma. Default is 0.95

9.      R will generate a log probability plot (also called a Q-Q Plot) that provides a visual check of whether the data fits the lognormal model (see the AIHA book.)  Creating a log probability plot requires two commands.  The first command, pnd<-plend(aihand), creates a data frame.  The second, qq.lnorm(pnd), generates a probability plot as a visual check of log-normality.   This plot displays only detected values and displays replicates as a single data point, which aids the visual check when the data set is large.  Clicking the camera button copies the image so that it can be pasted into Excel or another document. 


 


10.  The metrics calculated by the “readss” command can also be calculated separately.  The commands for these are described in the help menu, which is shown when you type “help(sand).” The “readss” routine requires at least 3 detected results to run.  One function that can be used with all non-detect is “nptl(n , p = 0.95, gam = 0.95)”, which provides the order of the value in a data set with n values that corresponds to non-parametric upper tolerance limit for specified percentile and upper tolerance. 

 



[1] Ignacio, J.S., and W.H. Bullock: A Strategy for Assessing and Managing Occupational Exposures, 3rd ed. Fairfax, Va.: AIHA Press, 2006.