# ------------------------------------------------
# Atlantic cod comparison .. adding environmental variation
# ------------------------------------------------
# load data common environment and parameter setting
# source( system.file( "scripts", "00_cod_comparisons_data_environment.R", package = "carstm") )
# --------------------------------
# construct basic parameter list defining the main characteristics of the study
# and some plotting parameters (bounding box, projection, bathymetry layout, coastline)
# NOTE: the data selection is the same as in (01_cod_comparisons_basic_stranal.R)
p = list(
speciesname = "Atlantic_cod",
groundfish_species_code = 10, # 10= cod
yrs = 1970:2017,
trawlable_units = "towdistance" # <<<<<<<<<<<<<<<<<<
# trawlable_units = "standardtow"
# trawlable_units = "sweptarea"
)
# --------------------------------
# parameter setting used to filter data via 'survey_db( DS="filter")'
# unlike stratanl, we do not need to remove strata until the last /aggregation step
# specific selection params required for survey_db(DS="filter") data selection mechanism
p = aegis.survey::survey_parameters(
p=p,
selection=list(
biologicals=list(
spec_bio = bio.taxonomy::taxonomy.recode( from="spec", to="parsimonious", tolookup=p$groundfish_species_code )
),
survey=list(
data.source="groundfish",
yr = p$yrs, # time frame for comparison specified above
months=6:8, # "summer"
# dyear = c(150,250)/365, # alternate way of specifying season: summer = which( (x>150) & (x<250) ) , spring = which( x<149 ), winter = which( x>251 )
settype = 1, # same as geartype in groundfish_survey_db
gear = c("Western IIA trawl", "Yankee #36 otter trawl"),
polygon_enforce=TRUE, # make sure mis-classified stations or incorrectly entered positions get filtered out
ranged_data = c("dyear") # not used .. just to show how to use range_data
)
)
)
# ------------------------------------------------
## using the "standard" polygon definitions .. see https://cran.r-project.org/web/packages/spdep/vignettes/nb.pdf
# Here we compute surface area of each polygon via projection to utm or some other appropriate planar projection.
# This adds some variabilty relative to "statanal" (which uses sa in sq nautical miles, btw)
sppoly = areal_units(
areal_units_type="stratanal_polygons_pre2014",
areal_units_proj4string_planar_km=p$areal_units_proj4string_planar_km
)
sppoly$strata_to_keep = ifelse( as.character(sppoly$AUID) %in% strata_definitions( c("Gulf", "Georges_Bank", "Spring", "Deep_Water") ), FALSE, TRUE )
# ------------------------------------------------
# neighbourhood structure --- required to do areal unit spatial modelling
# sppoly = neighbourhood_structure( sppoly=sppoly, areal_units_type="stratanal_polygons_pre2014" ) # not used here
# --------------------------------
# Get the data
p$selection$survey$strata_toremove = NULL # emphasize that all data enters analysis initially ..
set = survey_db( p=p, DS="filter" )
# categorize Strata
crs_lonlat = st_crs(projection_proj4string("lonlat_wgs84"))
sppoly = st_transform(sppoly, crs=crs_lonlat )
set$AUID = st_points_in_polygons(
pts = st_as_sf( set, coords=c("lon","lat"), crs=crs_lonlat ),
polys = sppoly[, "AUID"],
varname="AUID"
)
set = set[ which(!is.na(set$AUID)),]
set$totno[which(!is.finite(set$totno))] = NA
# --------------------------------
# ensure we have some estimate of sweptarea and choose the appropriate
# one based upon which trawlable units we are using
ft2m = 0.3048
m2km = 1/1000
nmi2mi = 1.1507794
mi2ft = 5280
standardtow_sakm2 = (41 * ft2m * m2km ) * ( 1.75 * nmi2mi * mi2ft * ft2m * m2km ) # surface area sampled by a standard tow in km^2 1.75 nm
set$data_offset = switch( p$trawlable_units,
standardtow = rep(standardtow_sakm2, nrow(set)) , # "standard tow"
towdistance = set$sa_towdistance, # "sa"=computed from tow distance and standard width, 0.011801==),
sweptarea = set$sa # swept area based upon stand tow width and variable lenths based upon start-end locations wherever possible
)
set$data_offset[which(!is.finite(set$data_offset))] = median(set$data_offset, na.rm=TRUE ) # just in case missing data
# ------------------------------------------------
# update set with AUID factor variables and a few other repeatedly used variables
# set$AUID = factor(set$AUID, levels=levels(sppoly$AUID))
space.id = slot( sppoly, "space.id")
set$space = set$space_time = match( set$AUID, space.id )
set$yr_factor = factor(set$yr)
set$time = set$time_space = as.numeric( set$yr_factor)
set$iid_error = 1:nrow(set) # for inla indexing
set$tag = "observations"
## --------------------------------
# construct meanweights matrix
weight_year = meanweights_by_arealunit( set=set, AUID=as.character( sppoly$AUID ), yrs=p$yrs, fillall=TRUE, annual_breakdown=TRUE )
# weight_year = meanweights_by_arealunit_modelled( p=p, redo=TRUE ) -- note: data passing of M needs to be modularized
# weight_year = weight_year[, match(as.character(p$yrs), colnames(weight_year) )]
# weight_year = weight_year[ match(as.character(sppoly$AUID), rownames(weight_year) )]
# adjust based upon RAM requirements and ncores
ncores = floor( ram_local( "ncores", ram_main=4, ram_process=6 ) / 2 )
inla.setOption(num.threads=ncores)
inla.setOption(blas.num.threads=ncores)
# RES = data.frame(yr=p$selection$survey[["yr"]]) # collect model comparisons
if (0) {
fn = file.path( getwd(), "RES.rdata" )
# save(RES, file=fn)
# load(fn)
}
## ----------------------------------
# covariates of interest
covars = c("t", "tsd", "tmax", "tmin", "degreedays", "z", "dZ", "ddZ" )
# currently supported:
# z = depth (m)
# dZ = bottom slope (m/km)
# ddZ = bottom curvature (m/km^2)
# substrate.grainsize = mean grain size of bottom substrate (mm)
# t = temperature (C) – subannual
# tlb = temperature lower 95% bound (C) –subannual
# tub = temperature upper 95% bound (C) –subannual
# tmean = mean annual temperature
# tsd = standard deviation of the mean annual temperature
# tmin = minimum value of temperature in a given year – annual
# tmax = maximum value of temperature in a given year – annual
# tamplitude = amplitude of temperature swings in a year (tmax-tmin) – annual
# degreedays = number of degree days in a given year – annual
# extract covariate means by strata
res = aegis_db_extract_by_polygon(
sppoly=sppoly,
vars=covars,
spatial_domain=p$spatial_domain,
yrs=p$yrs,
dyear=0.6 # 0.6*12 months = 7.2 = early July
)
# extract covariates and supplent survey data via lookups
set = aegis_db_lookup(
X=set,
lookupvars=covars,
xy_vars=c("lon", "lat"),
time_var="timestamp"
)
# good data
ok = which(
is.finite(set$totno) &
is.finite(set$t) &
is.finite(set$z) &
is.finite(set$data_offset) &
set$AUID %in% sppoly$AUID[sppoly$strata_to_keep]
)
# NOTE: R-INLA uses NA differently than other packages
# — NA in the response means no likelihood contribution, i.e. response is unobserved
# — NA in a fixed effect means no contribution to the linear predictor, i.e. the covariate is set equal to zero
# — NA in a random effect f(...) means no contribution to the linear predictor
APS = aegis_prediction_surface( aegis_data=res$means )
APS$yr = as.numeric( APS$year)
APS$totno = NA
APS$data_offset = 1 # force to be density n/km^2
APS$tag = "predictions"
varstokeep = c( "totno", "AUID", "yr", "t", "z", "data_offset", "tag" )
M = rbind( set[ok, varstokeep], APS[,varstokeep] )
M$t[!is.finite(M$t)] = median(M$t, na.rm=TRUE ) # missing data .. quick fix .. do something better
M$z[!is.finite(M$z)] = median(M$z, na.rm=TRUE ) # missing data .. quick fix .. do something better
M$yr_factor = factor( as.character(M$yr) )
M$AUID = factor( M$AUID, levels=levels(sppoly$AUID ))
M$strata = as.numeric( M$AUID)
M$year = as.numeric( M$yr_factor)
M$ti = discretize_data( M$t, p$discretization$t )
M$zi = discretize_data( M$t, p$discretization$z )
M$iid_error = 1:nrow(M) # for inla indexing for set level variation
# generic PC priors
m = log( {set$totno / set$data_offset}[ok] )
m[!is.finite(m)] = min(m[is.finite(m)])
H = inla_hyperparameters( sd(m), alpha=0.5, median(m) )
# H$prec$prec.intercept = 1e-9
# ------------------------------------------------
# Model 6: as in Model 5 but with main effects .... this is a full factorial model
# NOTE this is rank-deficient ... some factor combinations have insufficient data
# missing covariates/data forces more fiddling with 'kk'
fit = glm(
formula = totno ~ offset(log(data_offset)) + 1 + AUID:yr_factor + AUID + yr_factor ,
family=poisson(link="log"),
data=set[ok, ]
)
s = summary(fit)
AIC(fit) # 359785
# prediction surface as a data frame
aps=APS
aps$yr_factor = factor( aps$year, levels=p$yrs )
kk = which(aps$AUID %in% unique(set$AUID[ok]))
preds = predict( fit, newdata=aps[kk,], type="response", na.action=na.omit, se.fit=TRUE )
aps$predictions = NA
aps$predictions[kk] = preds$fit
aps$predictions.sd = NA
aps$predictions.sd[kk] = preds$se.fit
# reformat predictions into matrix form
out = reformat_to_array(
input = aps$predictions,
matchfrom = list( AUID=aps$AUID, yr_factor=aps$yr_factor),
matchto = list( AUID=sppoly$AUID, yr_factor=levels(set$yr_factor))
)
# convert numbers/km to biomass/strata
RES$glm_poisson_totno_factorial = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep , ], na.rm=TRUE )
lines( glm_poisson_totno_factorial ~ yr, data=RES, lty=5, lwd=4, col="red", type="b")
# map means adjusted by temperature and depth . .. as depth does not change, time dynamics in maps due to temperature and
iy = match( as.character(sppoly$AUID), aps$AUID )
vn = "pred"
sppoly[,vn] = aps$predictions[iy]
brks = interval_break(X= sppoly[[vn]], n=length(p$mypalette), style="quantile")
spplot( sppoly, vn, col.regions=p$mypalette, main=vn, at=brks, sp.layout=p$coastLayout, col="transparent" )
# ------------------------------------------------
# Model 6:
# "INLA full factorial" fixed fixed - Poisson 6136 unstable -
# full factorial upon totno -- base model for comparison with GLM ..
# NOTE ::: iid_error is req to stabilize solution as
# there are additional sources of variation that the
# factorial model does not fully account for
fit = inla(
formula = totno ~ 1 + AUID:yr_factor + AUID + yr_factor + f(iid_error, model="iid", hyper=H$iid),
family = "poisson",
data=M,
control.compute=list(cpo=TRUE, waic=TRUE, dic=TRUE, config=TRUE),
control.results=list(return.marginals.random=TRUE, return.marginals.predictor=TRUE ),
control.predictor=list(compute=TRUE, link=1 ), # compute=TRUE on each data location
# control.inla=list(strategy="laplace", dz=0.25, diff.logdens=9, restart=3, npoints=11, cutoff=1e-5),
control.inla=list(correct=TRUE, correct.verbose=FALSE ), # adding this will make it 3.5hrs long
control.fixed= H$fixed,
verbose=FALSE
)
fit$dic$dic # 34897
fit$dic$p.eff # 6848
s = summary(fit)
# reformat predictions into matrix form
ii = which(M$tag=="predictions")
out = reformat_to_array(
input = fit$summary.fitted.values[ ii, "mean" ],
matchfrom = list( AUID=M$AUID[ii], yr_factor=M$yr_factor[ii]),
matchto = list( AUID=sppoly$AUID, yr_factor=factor(p$yrs) )
)
# out[ out>1e10] = NA
RES$INLA.full.factorial = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep, ], na.rm=TRUE ) / standardtow_sakm2
INLA.full.factorial
lines( INLA.full.factorial ~ yr, data=RES, lty=1, lwd=8, col="blue")
# ------------------------------------------------
# Model 9: as in Model 6 but full factorial with covariates
# note this is rank-deficient
# problem .. missing values in some strata/years (as with stratanal)
# .. but this is more likely with more covariates
# this means we cannot get a consistent estimates ( as with stratanal)
# NOTE: this is a full interaction model with strata and years ...
# missing covariates/data forces more fiddling
fit = glm(
formula = totno ~ offset(log(data_offset)) + 1 + AUID:yr_factor + AUID + yr_factor +t + z,
family=poisson(link="log"),
data=set[ok, ]
)
s = summary(fit)
AIC(fit) # 348154
aps = APS
aps$yr_factor = factor( aps$year, levels=p$yrs)
kk = which(aps$AUID %in% unique(set$AUID[ok]))
preds = predict( fit, newdata=aps[kk,], type="response", na.action=na.omit, se.fit=TRUE )
aps$predictions = NA
aps$predictions[kk] = preds$fit
aps$predictions.sd = NA
aps$predictions.sd[kk] = preds$se.fit
# reformat predictions into matrix form
out = reformat_to_array(
input = aps$predictions,
matchfrom = list( AUID=aps$AUID, yr_factor=aps$yr_factor),
matchto = list( AUID=sppoly$AUID, yr_factor=levels(set$yr_factor))
)
# convert numbers/km to biomass/strata
RES$glm_poisson_totno_env = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep , ], na.rm=TRUE )
lines( glm_poisson_totno_env ~ yr, data=RES, lty=5, lwd=4, col="green", type="b")
# map means adjusted by temperature and depth . .. as depth does not change, time dynamics in maps due to temperature and
iy = match( as.character(sppoly$AUID), aps$AUID )
vn = "pred"
sppoly[,vn] = aps$predictions[iy]
brks = interval_break(X= sppoly[[vn]], n=length(p$mypalette), style="quantile")
spplot( sppoly, vn, col.regions=p$mypalette, main=vn, at=brks, sp.layout=p$coastLayout, col="transparent" )
# Bottom line: an additive, linear, fixed effect model requires a lot of handwaving and fiddling
# esp when missing data and linear extrapolation that might not make sense at the extremes
# ------------------------------------------------
# Model 9b: as in Model 9 but GAM .. occasionally unstable .. optimizer is finding a flat area
# rank deficient
fit = mgcv::gam(
formula = totno ~ offset(log(data_offset)) + 1 + AUID:yr_factor + AUID + yr_factor + s(t, bs="tp", k=3) + s(z, bs="tp", k=3),
family=poisson(link="log"),
data=set[ok, ]
)
s = summary(fit)
AIC(fit) # 340300
aps = APS
aps$yr_factor = factor( aps$year, levels=levels(set$yr_factor))
kk = which(aps$AUID %in% unique(set$AUID[ok]))
preds = predict( fit, newdata=aps[kk,], type="response", na.action=na.omit, se.fit=TRUE )
aps$predictions = NA
aps$predictions[kk] = preds$fit
aps$predictions.sd = NA
aps$predictions.sd[kk] = preds$se.fit
# reformat predictions into matrix form
out = reformat_to_array(
input = aps$predictions,
matchfrom = list( AUID=aps$AUID, yr_factor=aps$yr_factor),
matchto = list( AUID=sppoly$AUID, yr_factor=levels(set$yr_factor))
)
# convert numbers/km to biomass/strata
RES$gam_poisson_totno_env = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep , ], na.rm=TRUE )
lines( gam_poisson_totno_env ~ yr, data=RES, lty=5, lwd=4, col="green", type="b")
# ------------------------------------------------
# Model 11:
# "INLA Envir 0" -- base model using inla + envir --- mimicking the GAM model as much as possible
# simple factorial with totno and poisson
# improvement upon Model 6b .. INLA imputes missing data given proper data model .. less fiddling ..
# random effects.. `fewer params`
# NOTE: R-INLA uses NA differently than other packages
# — NA in the response means no likelihood contribution, i.e. response is unobserved
# — NA in a fixed effect means no contribution to the linear predictor, i.e. the covariate is set equal to zero
# — NA in a random effect f(...) means no contribution to the linear predictor
# interaction-only model does not really make sense here .. esp as there are other model components
fit = inla(
formula =
totno ~ 1 + offset( log( data_offset) )
+ AUID:yr_factor + AUID + yr_factor
+ f(iid_error, model="iid", hyper=H$iid)
+ f(ti, model="rw2", scale.model=TRUE, diagonal=1e-6, hyper=H$rw2)
+ f(zi, model="rw2", scale.model=TRUE, diagonal=1e-6, hyper=H$rw2),
family = "poisson", # "zeroinflatedpoisson0",
data= M,
control.compute=list(cpo=TRUE, waic=TRUE, dic=TRUE, config=TRUE),
control.results=list(return.marginals.random=TRUE, return.marginals.predictor=TRUE ),
control.predictor=list(compute=FALSE, link=1 ),
control.fixed= H$fixed,
verbose=TRUE
)
s = summary(fit)
s$dic$dic # 34988
s$dic$p.eff # 6894
plot(fit, plot.prior=TRUE, plot.hyperparameters=TRUE, plot.fixed.effects=FALSE )
# reformat predictions into matrix form
ii = which(M$tag=="predictions")
out = reformat_to_array(
input = fit$summary.fitted.values[ ii, "mean" ],
matchfrom = list( AUID=M$AUID[ii], yr_factor=M$yr_factor[ii]),
matchto = list( AUID=sppoly$AUID, yr_factor=factor(p$yrs) )
)
# out[ out>1e10] = NA
RES$INLA.Envir.0 = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep, ], na.rm=TRUE )
lines( INLA.Envir.0 ~ yr, data=RES, lty=1, lwd=2.5, col="blue", type="b")
# map it
vn = "pred"
yr = "2017"
sppoly[,vn] = out[,yr] * weight_year[,yr] # biomass density
brks = interval_break(X= sppoly[[vn]], n=length(p$mypalette), style="quantile")
spplot( sppoly, vn, col.regions=p$mypalette, main=vn, at=brks, sp.layout=p$coastLayout, col="transparent" )
# ------------------------------------------------
# Model 11a:
"INLA Envir 1" rw2: temp+depth Poisson 5960 33652 4
# simple factorial with totno and poisson
# improvement upon Model 6b .. INLA imputes missing data given proper data model .. less fiddling ..
# random effects.. `fewer params`
# NOTE: R-INLA uses NA differently than other packages
# — NA in the response means no likelihood contribution, i.e. response is unobserved
# — NA in a fixed effect means no contribution to the linear predictor, i.e. the covariate is set equal to zero
# — NA in a random effect f(...) means no contribution to the linear predictor
if(0) {
# alter priors for fixed effects
H$fixed$prec = list( prior="pc.prec", param=c(1, 0.5) ) # NOTE: pc.priors are on sd scale ..
H$fixed$prec.intercept = 1
}
# interaction-only model does not make sense here .. esp as there are other model components
fit = inla(
formula =
totno ~ 1 + offset( log( data_offset) )
+ f(strata, model="iid", group=year, hyper=H$iid)
+ f(iid_error, model="iid", hyper=H$iid)
+ f(year, model="iid", hyper=H$iid)
+ f(ti, model="rw2", scale.model=TRUE, diagonal=1e-5, hyper=H$rw2)
+ f(zi, model="rw2", scale.model=TRUE, diagonal=1e-5, hyper=H$rw2),
family = "poisson", # "zeroinflatedpoisson0",
data= M,
control.compute=list(cpo=TRUE, waic=TRUE, dic=TRUE, config=TRUE),
control.results=list(return.marginals.random=TRUE, return.marginals.predictor=TRUE ),
control.predictor=list(compute=FALSE, link=1 ),
control.fixed= H$fixed,
verbose=TRUE
)
s = summary(fit)
s$dic$dic # 33687
s$dic$p.eff # 5969
plot(fit, plot.prior=TRUE, plot.hyperparameters=TRUE, plot.fixed.effects=FALSE )
# reformat predictions into matrix form
ii = which(M$tag=="predictions")
out = reformat_to_array(
input = fit$summary.fitted.values[ ii, "mean" ],
matchfrom = list( AUID=M$AUID[ii], yr_factor=M$yr_factor[ii]),
matchto = list( AUID=sppoly$AUID, yr_factor=factor(p$yrs) )
)
# out[ out>1e10] = NA
RES$INLA.Envir.1 = colSums( {out * weight_year * sppoly$au_sa_km2}[sppoly$strata_to_keep, ], na.rm=TRUE )
lines( INLA.Envir.1 ~ yr, data=RES, lty=1, lwd=2.5, col="blue", type="b")
dev.new(width=11, height=7)
col = c("slategray", "turquoise", "darkorange", "blue", "green", "darkred", "cyan", "darkgreen", "slateblue" )
pch = c(20, 21, 22, 23, 24, 25, 26, 27, 20)
lty = c(1, 3, 4, 5, 6, 7, 1, 3, 4 )
lwd = c(4, 4, 4, 4, 4, 4, 4, 4, 4 )
type =c("l", "l", "l", "l", "l", "l", "l", "l", "l")
legend=c("Standard tow stratanal", "GLM factorial", "GLM Envir", "GAM Envir")
plot( stratanal_towdistance ~ yr, data=RES, lty=lty[1], lwd=lwd[1], col=col[1], pch=pch[1], type=type[1], ylim=c(0,0.3e9), xlab="Year", ylab="kg")
lines( glm_poisson_totno_factorial ~ yr, data=RES, lty=lty[2], lwd=lwd[2], col=col[2], pch=pch[2], type=type[2])
lines( glm_poisson_totno_env ~ yr, data=RES, lty=lty[3], lwd=lwd[3], col=col[3], pch=pch[3], type=type[3])
lines( gam_poisson_totno_env ~ yr, data=RES, lty=lty[4], lwd=lwd[4], col=col[4], pch=pch[4], type=type[4])
ii = 1:4
legend("topright", legend=legend[ii], lty=lty[ii], col=col[ii], lwd=lwd[ii] )
dev.new(width=6, height=4)
hist( RES$INLA.Envir.1 / RES$stratanal_towdistance, breaks=20 )
cor( RES[, c("stratanal_towdistance", "glm_poisson_totno_factorial", "INLA.Envir.1")])
plot( RES[, c("stratanal_towdistance", "glm_poisson_totno_factorial", "INLA.Envir.1")])
# ---- bias in station selection:
set$strata_year = paste( set$AUID, set$yr, sep=".")
zz = applyMean( set[, c("strata_year", "z")] )
tt = applyMean( set[, c("strata_year", "t")] )
APS$strata_year = paste( APS$AUID, APS$yr, sep=".")
APS = merge( APS, zz, by="strata_year", all.x=TRUE, all.y=FALSE, suffixes=c("", ".set") )
APS = merge( APS, tt, by="strata_year", all.x=TRUE, all.y=FALSE, suffixes=c("", ".set") )
APS$z_diff = APS$z - APS$z.set
APS$t_diff = APS$t - APS$t.set
# sampling bias between survey and strata
dev.new(); plot( z ~ z.set, APS ); abline(0,1)
dev.new(); plot( t ~ t.set, APS ); abline(0,1)
dev.new(); plot( z_diff ~ yr, APS, pch=20, cex=0.75, col="slategray"); abline(h=0); out=data.frame(yr=p$yrs); out$zz=predict( loess(z_diff ~ yr, APS, span=0.05 ), newdata=out, se=FALSE); lines(zz~yr, out, lwd=4, col="red")
dev.new(); plot( t_diff ~ yr, APS, pch=20, cex=0.75, col="slategray"); abline(h=0); out=data.frame(yr=p$yrs); out$zz=predict( loess(t_diff ~ yr, APS, span=0.05 ), newdata=out, se=FALSE); lines(zz~yr, out, lwd=4, col="red")
APS$strata = as.numeric( APS$AUID )
dev.new(); plot( z_diff ~ strata, APS, pch=20, cex=0.75, col="slategray"); abline(h=0); out=data.frame(strata=sort(unique(APS$strata))); out$zz=predict( loess(z_diff ~ strata, APS, span=0.05 ), newdata=out, se=FALSE); lines(zz~strata, out, lwd=4, col="red")
dev.new(); plot( t_diff ~ strata, APS, pch=20, cex=0.75, col="slategray"); abline(h=0); out=data.frame(strata=sort(unique(APS$strata))); out$zz=predict( loess(t_diff ~ strata, APS, span=0.05 ), newdata=out, se=FALSE); lines(zz~strata, out, lwd=4, col="red")
# ### end
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.