2_baselineModel_v1_3.Rmd

Introduction

modelVersion <- "v1_3"

Model: r modelVersion

This version of the baseline microsimulation model uses a household dataset (n = 1800) with household size (occupancy) set to match the UK 2011 Census distributions together with regression coefficients from [@parkerThesis2014] to construct a synthetic household level micro-component water consumption dataset.

# set model v1 paths for BA laptop ----
if(startsWith(userName, "ben")) # => BA laptop
  m1iPath <- paste0(iPath, "model_v1/")
if(startsWith(userName, "ben")) # => BA laptop
  m1oPath <- paste0(dPath, "model_outputs/model_v1/")

# set paths for DM laptop ----
if(startsWith(userName, "despina")) # => DM laptop
  m1iPath <- iPath
if(startsWith(userName, "despina")) # => DM laptop
  m1oPath <- oPath

print(paste0("User: ",userName))
print(paste0("Platform: ",sysName))
print(paste0("Model version: ",modelVersion))
print(paste0("Input data path: ",m1iPath))
print(paste0("Output data path: ",m1oPath))

How it works

• Import data:
  • Household level data (hhid, occupancy & baseline microcomponent median consumption values from [@parkerThesis2014])
  • UK Met office climate data per month for period
  • Microcomponent regression coefficients that link occupancy & climate with demand:
    • Metered household co-efficients for climate factors, occupancy, month and yearly change
    • Un-metered household coefficients for climate factors, occupancy, month and yearly change
• Set baseline (Year 1) consumption per micro-component
  • Set baseline consumption for each microcomponent for each household using summary statistics from [@parkerThesis2014]
  • Randomise the distributions using a series of skewed normal distribution models
  • Adjust modeled baseline microcomponent consumption using occupancy co-efficients from [@parkerThesis2014]
• Write output:
  • Adjusted micro-component consumption (litres/day) for each household for each month & year
  • Summary statistics for comparison purposes

Import household level attributes

This is a synthetic household sample of n = 1800 constructed to match the occupancy distributions of the UK.

hhPropsDT <- data.table(read.csv(paste(m1iPath,"rdata/hhpropstemp_v1_2.csv", sep = "")))
kable(caption = "First 10 rows)",
      head(hhPropsDT,10)
)

kable(caption = "Summary of baseline hh properties",
      summary(hhPropsDT)
)

n_hh <- nrow(hhPropsDT) #n_hh <- nrow(hhpropsDT) # Number of housesholds in the modelled area.

We now randomly set 40% of the households to be metered to reflect overall metering rates in England (XX https://link.springer.com/article/10.1007%2Fs11269-012-0210-2 any update? XX).

hhPropsDT <- ba_IMPETUSsetMeteredHH(hhPropsDT)

# check
kable(caption = "Check % metered distribution", 
      100*round(prop.table(table(hhPropsDT$metered, hhPropsDT$occupancy),2),3)
)

Baseline consumption estimation

Component litres/day

The first step in estimating the baseline per-component litres consumed is to extract sensible `daily l' values from Parker's thesis [@parkerThesis2014] for both metered and un-metered households.

mlmDataSummaryDT <- data.table(read.csv(paste0(iPath,"Parker-2014-Thesis-Micro-Component-Golden-100_mlm_summary.csv")))

kable(caption = "Summary of mean/median values per micro-component (Parker, 2014, all households)",
  mlmDataSummaryDT
)

As the table above shows, this causes an immediate problem as the overall mean values are highly skewed (compare mean & median) and the overall ('all days') median values are 0 for some microcomponents (Bath, Dishwasher, External) indicating relatively rare usage. However it is noticeable that where the median is non-zero for all days, the ratio of mean:median is similar for 'all days' vs 'usage days'. For model modelVersion we have therefore chosen to use:

• the median values for 'all days' for metered and unmetered households (Parker, 2014, Table 4.4);
• the median values for 'all days' for Bath, Dishwasher, External for metered and unmetered households (Parker, 2014, Table 4.4) corrected by the ratio between mean 'all days' basin use and mean 'all days' consumption for the component in question.

We acknowledge that this approach may mis-estimate water consumption for baths in particular as the model does not contain indicators or the presence/use of baths in a particular household.

# Combine the base household DT with the selected l/day for each microcomponent

#hhMcDataDT <- ba_IMPETUSsetBaseline(mlmDataSummaryDT, hhPropsDT)

# function to set basic consumption from consDT & merge to hhDT
# consDT is the data from Parker's thesis
# we start from the medians but select the specific rows relevant to the model we
# are running (see text that calls this function for more details)
# requires that modelVersion is set
mcMedianDT <- mlmDataSummaryDT[type == paste0("Final values ",modelVersion),
                     .(metered = var, Basin.parkerMedian	, Bath.parkerMedian	, Dishwasher.parkerMedian	, 
                       External.parkerMedian	, KitchenSink.parkerMedian	, Shower.parkerMedian	, WC.parkerMedian	, WashingMachine.parkerMedian)
                     ]

# Calculate the modelled consumption using the median data for all households from Parker.
# ideally we'd want to match occupancy levels but Parker has no tables that allow us to do this
# match on meter/not metered
setkey(mcMedianDT, metered)
setkey(hhPropsDT, metered)
hhMcDataDT <- merge(hhPropsDT, mcMedianDT, all.x = TRUE) # This will recycle the single median rows onto the relevant metered /unmetered rows



# This will have produced a table with medians which are not corrected for occupancy (or anything else) and so will be constant across occupancy groups thus:

sumt <- hhMcDataDT[, .(meanBasin = mean(Basin.parkerMedian),
                       sdBasin = sd(Basin.parkerMedian),
                       meanBath = mean(Bath.parkerMedian),
                       sdBath = sd(Bath.parkerMedian)
                       ), keyby = .(metered, occupancy)]

kable(caption = "Test l/day", sumt)

As we can see this produces constant values across occupancy levels and metered vs non-metered.

Distributional adjustment

As the table above shows, this procedure will have produced a table with medians which are not corrected for occupancy (or anything else) and so will be constant across occupancy groups within metered/un-metered households. To introduce random variation into the distributions we next generate a skewed normal distribution for each micro-component using the original median value taken from Parker as the distribution mean. We have no information on the correct standard deviation nor skewness to use but through experimentation we have idenitifed a range of s.d values and xi parameters that, when used with the R function rsnorm [@fgarch], produce results that are similar to Parker's per capita/day distributions (p110).

bathCut <- 30
showerCut <- 5
externalCut <- 10

In addition the following adjustments are made:

• r bathCut l/day to zero
• r showerCut l/day to zero
• r externalCut l/day to zero

The following density plots show the results of these distributional adjustments. All show litres per household per day.

# To fix this problem we calculate a distirbution for each micro-component using a normal distirbution with mean = the Parker median

# use rsnorm to re-distribute the input consumption values
# we have chosen parameters that produce distributions to match Parker's graphs
# We have no other information on the s.d. to use but Figure 5.3 (p110) of Parker's thesis 
# suggests somthing close to:
# rsnorm(n, mean = 15, sd = 5, xi = 4)) - Skew Normal Distribution
# https://stackoverflow.com/questions/20254084/plot-normal-left-and-right-skewed-distribution-in-r

nObs <- nrow(hhMcDataDT)
hhMcDataDT <- hhMcDataDT[, Basin.rsnorm := rsnorm(nObs, mean = Basin.parkerMedian, sd = 3, xi = 3)]
hhMcDataDT <- hhMcDataDT[, Bath.rsnorm := rsnorm(nObs, mean = Bath.parkerMedian, sd = 10, xi = 2)]
hhMcDataDT <- hhMcDataDT[, Dishwasher.rsnorm := rsnorm(nObs, mean = Dishwasher.parkerMedian, sd = 2, xi = 3)]
hhMcDataDT <- hhMcDataDT[, External.rsnorm := rsnorm(nObs, mean = External.parkerMedian, sd = 8, xi = 3)]
hhMcDataDT <- hhMcDataDT[, KitchenSink.rsnorm := rsnorm(nObs, mean = KitchenSink.parkerMedian, sd = 3, xi = 3)]
hhMcDataDT <- hhMcDataDT[, Shower.rsnorm := rsnorm(nObs, mean = Shower.parkerMedian, sd = 5, xi = 2)]
hhMcDataDT <- hhMcDataDT[, WC.rsnorm := rsnorm(nObs, mean = WC.parkerMedian, sd = 15, xi = 2)]
hhMcDataDT <- hhMcDataDT[, WashingMachine.rsnorm := rsnorm(nObs, mean = WashingMachine.parkerMedian, sd = 14, xi = 3)]


# implement the 'cuts' specified above
hhMcDataDT <- hhMcDataDT[, Bath.rsnorm := ifelse(Bath.rsnorm < bathCut, 0, Bath.rsnorm)] # set < bathCut l/day to zero
hhMcDataDT <- hhMcDataDT[, External.rsnorm := ifelse(External.rsnorm < externalCut, 0, External.rsnorm)] # set < externalCut l/day to zero
hhMcDataDT <- hhMcDataDT[, Shower.rsnorm := ifelse(Shower.rsnorm < showerCut, 0, Shower.rsnorm)] # set < showerCut l/day to zero

# test them
myCaption <- "Baseline consumption (litres/day/household), rsnorm distribution"

myTitle <- "Basin use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "Basin.rsnorm")
myPlot

myTitle <- "Bath use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "Bath.rsnorm")
myPlot

myTitle <- "Dishwasher use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "Dishwasher.rsnorm")
myPlot

myTitle <- "External use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "External.rsnorm")
myPlot

myTitle <- "Kitchen sink use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "KitchenSink.rsnorm")
myPlot

myTitle <- "Shower use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "Shower.rsnorm")
myPlot

myTitle <- "WC use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "WC.rsnorm")
myPlot

myTitle <- "Washing machine use"
myPlot <- ba_IMPETUSmakeDensityPlot(hhMcDataDT, "WashingMachine.rsnorm")
myPlot

testDT <- hhMcDataDT[, .(Basin.rsnorm,
                         Bath.rsnorm,
                         Dishwasher.rsnorm,
                         External.rsnorm,
                         KitchenSink.rsnorm,
                         Shower.rsnorm,
                         WC.rsnorm,
                         WashingMachine.rsnorm)
                       ]

kable(caption = "Test Basin & Bath l/day as modelled using rsnorm()", summary(testDT))

The table above shows the effect on the mean consumption values for Basin and Bath by occupancy group and metered/un-metered. Note that there is still no correction for occupancy levels.

Test metered vs non-metered differences:

t <- hhMcDataDT[, .(Basin = mean(Basin.rsnorm),
                    Bath = mean(Bath.rsnorm),
                    Dishwasher = mean(Dishwasher.rsnorm),
                    External = mean(External.rsnorm),
                    KitchenSink = mean(KitchenSink.rsnorm),
                    Shower = mean(Shower.rsnorm),
                    WC = mean(WC.rsnorm),
                    WashingMachine = mean(WashingMachine.rsnorm)
                    ),
                by = .(metered)
                ]

kable(caption = "Metered vs un-metered estimates (mean)",
      t
      )

wmMetered <- t[metered == "Metered", WC]
wmNonMetered <- t[metered == "Not metered", WC]

% difference between means for washing machine: r round(100*(wmNonMetered - wmMetered)/wmNonMetered,2).

Occupancy based adjustment

To correct for the household occupancy, we start with the initial household attributes (occupancy) and modelled daily microcomponent usage volumes and adjust them according to the occupancy regression coefficiencts given by the table below.

kable(caption = "Table of occupancy adjustments (from [@parkerThesis2014] Table A3/A4, p221-222, zeros indicate n/s coefficients)",
      occupancyAdjCoeffDT
      )

Thus no adjustments are made where occupancy = 1. Note the presence of -ve values - these reflect negative values in Parker's regression model results which are potentially a side effect of the small sample sizes, especially for metered households, in her data. it should be noted that these coefficients are part of an overall model of each microcomponent's litres/day and includes a range of co-variates that are not in our model such as day of the week, ACORN code, Temperature range, rainfall over previous seven days and an estimate of soil moisture deficit. This means that it may not be entirely appropriate to apply just the occupancy, climatic and monthly coeficients in the baseline estimation but without the ability to re-estimate Parker's regression models with the reduced variable set, we have little choice.

# adjust baseline rsnorm values according to Parker's occupancy coefficients
setkey(occupancyAdjCoeffDT, metered, occupancy) # appropriate adjustment table
setkey(hhMcDataDT, metered, occupancy) # hh data

occAdjValuesDT <- merge(hhMcDataDT, occupancyAdjCoeffDT, all.x = TRUE) # adds appropriate adjustment coefficients to the hh data, keeps only the original data

# Compute adjusted values for each household
# Add the relevant occupancy adjustment coefficient to the original value - note that the occupancy value is a factor in Parker's model (so compared to 1 person)
# This coefficient = 0 for single person households so the addition will have no effect
# for 1 person households
# Need to multiply by 30 to convert from daily to monthly if desired (we work in daily values)

hhDataFinalDaily_v1_3DT <- occAdjValuesDT[, ':=' (Basin.baseline = Basin.rsnorm + Basin,
                                             Bath.baseline = Bath.rsnorm + Bath,
                                             Dishwasher.baseline = Dishwasher.rsnorm + Dishwasher,
                                             External.baseline = External.rsnorm + External,
                                             KitchenSink.baseline = KitchenSink.rsnorm + KitchenSink,
                                             Shower.baseline = Shower.rsnorm + Shower,
                                             WC.baseline = WC.rsnorm + WC,
                                             WashingMachine.baseline = WashingMachine.rsnorm + WashingMachine
)
]

hhDataFinalDaily_v1_3DT <- hhDataFinalDaily_v1_3DT[, .(metered,occupancy,hhid,lsoaarea,dtype, 
                                             Basin.baseline, Bath.baseline, Dishwasher.baseline, External.baseline,
                                             KitchenSink.baseline, Shower.baseline, WC.baseline, WashingMachine.baseline)] # keep only final occupancy adjusted values to avoid confusion

hhDataFinalDaily_v1_3DT <- hhDataFinalDaily_v1_3DT[, sumDaily.baseline := Basin.baseline + Bath.baseline + Dishwasher.baseline + 
                                           External.baseline + KitchenSink.baseline + Shower.baseline + WC.baseline + WashingMachine.baseline]


sumt <- hhDataFinalDaily_v1_3DT[, .(meanBasin.baseline = round(mean(Basin.baseline),2),
                       sdBasin.baseline = round(sd(Basin.baseline),2),
                       meanBath.baseline = round(mean(Bath.baseline),2),
                       sdBath.baseline = round(sd(Bath.baseline),2)
                       ), keyby = .(metered, occupancy)]

kable(caption = "Test monthly fixed and occupancy adjusted l/day", sumt)

The table above shows the effect on the median consumption values for Basin and Bath by occupancy group and metered/un-metered.

Test baseline consumption

The following charts are a sense-check of the final adjusted baseline consumption values.

# some checks to see if it all makes sense
myCaption <- "Baseline household consumption, corrected for occupancy and rsnorm distributed"

myTitle <- "Basin use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "Basin.baseline")
myPlot

myTitle <- "Bath use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "Bath.baseline")
myPlot

myTitle <- "Dishwasher use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "Dishwasher.baseline")
myPlot

myTitle <- "Kitchen sink use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "KitchenSink.baseline")
myPlot

myTitle <- "External use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "External.baseline")
myPlot

myTitle <- "Shower use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "Shower.baseline")
myPlot

myTitle <- "WC use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "WC.baseline")
myPlot

myTitle <- "Washing machine use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "WashingMachine.baseline")
myPlot

myTitle <- "Total daily use"
myPlot <- ba_IMPETUSmakeOccupancyBoxPlot(hhDataFinalDaily_v1_3DT, "sumDaily.baseline")
myPlot

As we can see there are only small 'occupancy' effects for most usages, reflecting the relatively low values of the regression coefficients taken from [@parkerThesis2014]. Effects for metered households do not closely follow the direction expected and may reflect unusual characteristics of the un/metered households in Parker's sample. This may be particularly evident for Dishwashers and Showers where we are applying negative regression coefficients.

For comparison, the EST 2013 report provides estimates of the daily mean litres of water per microcomponent. These are shown in the the table below and alongside the IMPETUS model results in the following chart.

kable(caption = "Summary of mean daily litres per household with different occupancy levels (EST, 2013, imputed data using assumptions and survey of unknown bias)",
      est2013dataFig2DT
)

modelDT <- hhDataFinalDaily_v1_3DT[, .(modelMean = mean(sumDaily.baseline),
                                  modelPcc = mean(sumDaily.baseline/occupancy)
), by = occupancy]
modelDT <- modelDT[, occupancy := ifelse(occupancy == 6, "6+", occupancy)]
setkey(modelDT, occupancy)

est2013dataFig2DT <- est2013dataFig2DT[, occupancy := N.people]
setkey(est2013dataFig2DT, occupancy)

# need to fix so colours constant for source
ggplot(est2013dataFig2DT[modelDT], aes(x = factor(occupancy))) +
  geom_point(aes(y = AverageLitresPerDayPerHousehold, colour = "Per household (EST, 2013)"), shape = 1) +
  geom_point(aes(y = AverageLitresPerDayPerCapita, colour = "Per capita (EST, 2013)"), shape = 1) +
  geom_point(aes(y = modelMean, colour = "Per household (IMPETUS Model)"), shape = 7) +
  geom_point(aes(y = modelPcc, colour = "Per capita (IMPETUS model)"), shape = 7) +
  theme(legend.title = element_blank()) +
  labs(title = "Total use (l/day, EST 2013)",
       y = "'Average' litres/day",
       x = "N people")

As we can see the EST litres per household estimates increase linearly with occupancy and the per capita consumtion estimates decline. The IMPETUS model on the other hand shows a small increase in daily litres per household and a quite rapid decline in PCC as occupancy increases. Whilst we would expect an increase in 'per household' values as occupancy increases it is not clear that we would expect a linear increase due to economies of scale and shared resources (such as washing machines) as Parker's results show. Nevertheless the IMPETUS estimates for single and double occupancy houdeholds do seem high compared to the same values for larger households. It is possible that either smaller household consumption is over-estimated or larger household consumption is under-estimated. Given the similarity of PCC values at higher occupancy levels it seems likely that the IMPETUS model over-estimates consumption for smaller households

Comments?

In terms of indiviudal components, according to EST the average household uses 349 litres of water each day (p8) and this is made up of the usages shown in the table below.

kable(caption = "Summary of mean daily litres per micro-component (EST, 2013, imputed dta using assumptions and survey of unknown bias)",
      est2013dataFig1DT
)

Comparing these values with the IMPETUS model is not straightforward as not all of the usages match to the microcomonents modelled. However, the following chart attempts to show all values on the sae graphs as far as possible. Wider bars indicate values which cannot be matched.

myCap <- "IMPETUS model: synthetic households (n = 1800)\n Model v1" # simple for
# change labels to match EST
modelDT <- hhDataFinalDaily_v1_3DT[, .(vol = "vol",
                             Basin = round(mean(Basin.baseline),2),
                             Dishwasher = round(mean(Dishwasher.baseline),2),
                             External = round(mean(External.baseline),2),
                             "Kitchen Sink" = round(mean(KitchenSink.baseline),2),
                             Bath = round(mean(Bath.baseline),2),
                             Shower = round(mean(Shower.baseline),2),
                             Toilet = round(mean(WC.baseline),2),
                             "Washing machine" = round(mean(WashingMachine.baseline),2),
                             Total = round(mean(sumDaily.baseline),2)
)
]
# recast
modelDT <- dcast(melt(modelDT, id.vars = "vol"), variable ~ vol)
modelDT <- modelDT[, Usage := variable]
modelDT <- modelDT[, source := paste0("IMPETUS Model")]
modelTot <- modelDT[Usage == "Total", vol]
modelDT <- modelDT[, pcTot := 100*(vol/modelTot)]

setkey(modelDT, Usage, variable, source) # simplest way to remove vars!

estDT <- est2013dataFig1DT
estDT <- estDT[, pcTot := percentOfTotalDaily]
estDT <- estDT[, vol := imputed.l.hh.day]
estDT <- estDT[, source := "EST (2013)"]

modelDT$variable <- NULL # remove
estDT$percentOfTotalDaily <- NULL
estDT$imputed.l.hh.day <- NULL

plotDT <- rbind(estDT, modelDT)

plotDT <- plotDT[, Usage := as.factor(Usage)]
plotDT <- plotDT[, UsageRo := relevel(Usage, "Total")] # put usage at the end

ggplot(plotDT, aes(x=UsageRo, fill = source)) +
  geom_col(aes(y = vol), position = "dodge") +
  labs(y = "Mean l/day",
       x = "Usage",
       caption = myCap) +
  coord_flip() # rotate for legibility

estPlot <- ggplot(plotDT[Usage != "Total"], aes(x=UsageRo, fill = source)) +
  geom_col(aes(y = pcTot), position = "dodge") +
  labs(y = "% household total",
       x = "Usage",
       caption = myCap) +
  coord_flip() # rotate for legibility

estPlot

# Grey scale version if required
estPlot <- estPlot + theme_bw()

ggsave(paste0("plots_v1/Fig2_CompareModelv1_3withEST2013.png"), dpi = 400)

These charts suggest that compared to the EST (2013) estimates our model underestimates Shower use and over-estimates Bath use. However, given that the EST estimates used a self-selecting sample who may have been more likely to be 'careful' water users, it may be that respondents to their survey were more likely to use showers than baths. Of the other usages that can be directly compared, our model also slighty overestimates external, dishwasher, toilet and (especially) washing machine use, perhaps for similar reasons. Overall, accounting for the usages that are not directly comparable (basin, taps, kitchen sink etc) the mean 'Total' usage figures were broadly comparable.

They are both models of course... How could we validate with measured data? Artesia?

Output final baseline adjusted microdata

Now write out the household level values for safety and for re-use.

# save out summary stats by metered/unmetered & occpuancy for comparison
summaryDT <- hhDataFinalDaily_v1_3DT[, .(meanBasin = mean(Basin.baseline),
                                    meanBath = mean(Bath.baseline),
                                    meanDishwasher = mean(Dishwasher.baseline),
                                    meanExternal = mean(External.baseline),
                                    meanKitchenSink = mean(KitchenSink.baseline),
                                    meanShower = mean(Shower.baseline),
                                    meanWC = mean(WC.baseline),
                                    meanWashingMachine = mean(WashingMachine.baseline)
), by = .(occupancy, metered)]
oFile <- paste0(m1oPath,"summary_output-hh-baseline-mcm-consumption_",format(Sys.time(), "%Y-%m-%d"), "_", modelVersion, ".csv")

write.csv(summaryDT, oFile)

# final household data
# save out final baseline hh mcm data
oFile <- paste0(m1oPath,"output-hh-baseline-mcm-consumption_",format(Sys.time(), "%Y-%m-%d"), "_", modelVersion, ".csv")

keepDT <- hhDataFinalDaily_v1_3DT[, .(hhid, occupancy, metered, 
                                 Basin.baseline, Bath.baseline , Dishwasher.baseline , 
                                 KitchenSink.baseline , Shower.baseline , WC.baseline , 
                                 WashingMachine.baseline , External.baseline, sumDaily.baseline)]
write.csv(keepDT, oFile)

# compress it
# now gzip new one
# print(paste0("gzipping file to: ", oFile, ".gz"))
# cmd <- paste0("gzip -f ", oFile) # forces over-write
# try(system(cmd)) # in case it fails (it will on windows - you will be left with a .csv file)
# path issue - fails

Data written to:

r oPath and gzipped