Public Member Functions

JackKnifeList Class Reference

list of JackKnife data More...

#include <mclist.h>

Inheritance diagram for JackKnifeList:
PsiMClist

List of all members.

Public Member Functions

 JackKnifeList (unsigned int nblocks, unsigned int nprm, double maxldev, std::vector< double > maxlest)
 constructor
unsigned int getNblocks (void) const
 get the number of blocks in the current experiment
double influential (unsigned int block, const std::vector< double > &ci_lower, const std::vector< double > &ci_upper) const
bool outlier (unsigned int block) const
 is block an outlier?

Detailed Description

list of JackKnife data

JackKnifeing is not suggested for the assessment of confidence intervals or variability. Instead the close link between jackknife samples and individual data points is useful to determine influential data points and outliers.

Constructor & Destructor Documentation

JackKnifeList::JackKnifeList ( unsigned int  nblocks,
unsigned int  nprm,
double  maxldev,
std::vector< double >  maxlest 
) [inline]

constructor

Parameters:
nblocks number of blocks in the experiment
nprm number of parameters in the model
maxldev deviance of the maximum likelihood estimate on the full dataset
maxlest maximum likelihood estimate of the full dataset

Member Function Documentation

double JackKnifeList::influential ( unsigned int  block,
const std::vector< double > &  ci_lower,
const std::vector< double > &  ci_upper 
) const

determination of influential observations is performed by checking whether a parameter changes significantly (as defined by the confidence intervals) if one observation is omitted. Thus, if leaving out one observation results in significant changes in the estimated parameters, this observation is considered "influential".

Parameters:
block index of the block to be checked
estimate point estimate of the parameters in the model
ci_lower lower confidence limits for each parameter in the model
ci_upper upper confidence limits for each parameter in the model
Returns:
a number indicating the influence of the block. Values > 1 correspond to point estimates for that block that are precisely on the CI limits
bool JackKnifeList::outlier ( unsigned int  block  )  const

is block an outlier?

determination of outliers is based on the following idea: We add a new parameter that fits the data in block perfectly. If this "modified" model is significantly better than the original model, then this block is considered an outlier.

Parameters:
block index of the block to be checked
Returns:
true if block presents an outlier
The documentation for this class was generated from the following files:
Generated on Mon Jul 4 2011 14:52:04 for Psi++ by  doxygen 1.7.1