sn (version 2.1.1)

fitdistr.grouped-class: Methods for objects of class created by fitdistr.grouped

Description

A successful call to function fitdistr.grouped creates an object of class, also named fitdistr.grouped, for which a set of methods exist. The structure of an object of this class is described in section ‘Object components’.

Usage

# S3 method for fitdistr.grouped
logLik(object, ...)
  # S3 method for fitdistr.grouped
coef(object, ...)
  # S3 method for fitdistr.grouped
vcov(object, ...)
  # S3 method for fitdistr.grouped
print(x, ...)
  # S3 method for fitdistr.grouped
summary(object, cor=FALSE, ...)
  # S3 method for fitdistr.grouped
fitted(object, full=FALSE, ...)

Arguments

x, object

an object of class fitdistr.grouped as created by a call to the function with this name.

cor

logical (default=FALSE); is the correlation matrix required?

full

logical (default=FALSE); must the vector of fitted frequencies include the boundary classes, when they are added to cover the full support of the fitted distribution?

...

further arguments passed to or from other methods.

Object components

Components of an object of class fitdistr.grouped:

call

the matched call

family

the selected family of distributions

logL

the achieved maximum log-likelihood

param

a vector of estimated parameters

vcov

the approximate variance-covariance matrix of the estimates

input

a list with the input quantities and some derived ones

opt

a list as returned by optim

Author

Adelchi Azzalini

References

Possolo, A., Merkatas, C. and Bodnar, O. (2019). Asymmetrical uncertainties. Metrologia 56, 045009.

See Also

the function fitdistr.grouped, the plotting method plot.fitdistr.grouped

Examples

data(barolo)
attach(barolo)
A75 <- (reseller=="A" & volume==75)
logPrice <- log(price[A75], 10) # as used in selm documentation; see its fitting
detach(barolo)
breaks <- seq(1, 3, by=0.25)
f <- cut(logPrice, breaks = breaks)
counts <- tabulate(f, length(levels(f))) 
fit.logPrice.gr <- fitdistr.grouped(breaks, counts, family='ST')
summary(fit.logPrice.gr) # compare this fit with the ungrouped data fitting 
print(fit.logPrice.gr)
coef(fit.logPrice.gr)
vcov(fit.logPrice.gr)
data.frame(intervals=levels(f), counts, fitted=format(fitted(fit.logPrice.gr)))
full.intervals <- c("(-Inf, 1]", levels(f), "(3, Inf)")
data.frame("full-range intervals" = full.intervals,
       "full-range counts" = c(0, counts, 0), 
      "full-range fit" = fitted(fit.logPrice.gr, full=TRUE))
sum(counts) - sum(fitted(fit.logPrice.gr))  
sum(counts) - sum(fitted(fit.logPrice.gr, full=TRUE))  # must be "nearly 0" 
#---
# Use first entry in Table 3 of Possolo et al. (2019) and do similar fitting
# to the *probability* values, not observation counts
afcrc59 <- 1.141
breaks <- c(-Inf, afcrc59 - 0.033, afcrc59, afcrc59 + 0.037, Inf)
prob <-  c(0.16, 0.50, 0.84) 
cum.percent <- c(0, prob, 1)*100 
fitSN <- fitdistr.grouped(breaks, counts=diff(cum.percent), family="SN") 
print(coef(fitSN))
print(rbind(target=c(prob, 1)*100, fitted=cumsum(fitted(fitSN))), digits=5)
# Note: given the nature of these data (i.e. probabilities, not counts), 
#       there is no point to use vcov, logLik and summary on the fitted object.