harrison.priors {agridat}

Ranges of analytes in soybean from other authors

Description

Ranges of analytes in soybean from other authors

Format

A data frame with 80 observations on the following 5 variables.

source

Source document

substance

Analyte substance

min

minimum amount (numeric)

max

maximum analyte amount (numeric)

number

number of substances

Details

Harrison et al. show how to construct an informative Bayesian prior from previously-published ranges of concentration for several analytes.

The units for daidzein, genistein, and glycitein are micrograms per gram.

The raffinose and stachyose units were converted to a common 'percent' scale.

The author names in the 'source' variable are shortened forms of the citations in the supplemental information of Harrison et al.

Source

Jay M. Harrison, Matthew L. Breeze, Kristina H. Berman, George G. Harrigan. 2013. Bayesian statistical approaches to compositional analyses of transgenic crops 2. Application and validation of informative prior distributions. Regulatory Toxicology and Pharmacology, 65, 251-258. https://doi.org/10.1016/j.yrtph.2012.12.002

Data retrieved from the Supplemental Information of this source.

References

Jay M. Harrison, Derek Culp, George G. Harrigan. 2013. Bayesian MCMC analyses for regulatory assessments of safety in food composition Proceedings of the 24th Conference on Applied Statistics in Agriculture (2012).

Examples

## Not run: 

library(agridat)
data(harrison.priors)
dat <- harrison.priors

d1 <- subset(dat, substance=="daidzein")

# Stack the data to 'tall' format and calculate empirical cdf
d1t <- with(d1, data.frame(xx = c(min, max), yy=c(1/(number+1), number/(number+1))))

# Harrison 2012 Example 4: Common prior distribution
# Harrison uses the minimum and maximum levels of daidzein from previous
# studies as the first and last order statistics of a lognormal
# distribution, and finds the best-fit lognormal distribution.

m0 <- mean(log(d1t$xx)) # 6.37
s0 <- sd(log(d1t$xx)) # .833
mod <- nls(yy ~ plnorm(xx, meanlog, sdlog), data=d1t,
           start=list(meanlog=m0, sdlog=s0))
coef(mod) # Matches Harrison 2012
##   meanlog     sdlog
## 6.4187829 0.6081558

plot(yy~xx, data=d1t, xlim=c(0,2000), ylim=c(0,1),
     main="harrison.priors - Common prior", xlab="daidzein level", ylab="CDF")
mlog <- coef(mod)[1] # 6.4
slog <- coef(mod)[2] # .61
xvals <- seq(0, 2000, length=100)
lines(xvals, plnorm(xvals, meanlog=mlog, sdlog=slog))


d1a <- d1
d1a$source <- as.character(d1a$source)
d1a[19,'source'] <- "(All)" # Add a blank row for the densitystrip
d1
libs(latticeExtra)
# Plot the range for each source, a density curve (with arbitary
# vertical scale) for the common prior distribution, and a density
# strip by stacking the individual bands and using transparency
segplot(factor(source) ~ min+max, d1a,
        main="harrison.priors",xlab="daidzein level",ylab="source") +
  xyplot(5000*dlnorm(xvals, mlog, slog)~xvals, type='l') +
  segplot(factor(rep(1,18)) ~ min+max, d1, 4, level=d1$number,
          col.regions="gray20", alpha=.1)


## End(Not run)