com.itemanalysis.psychometrics.irt.estimation

Class MarginalMaximumLikelihoodEstimation

java.lang.Object

com.itemanalysis.psychometrics.irt.estimation.MarginalMaximumLikelihoodEstimation

public class MarginalMaximumLikelihoodEstimation
extends java.lang.Object

Marginal maximum likelihood estimation (MMLE) for item parameters in Item Response Theory (IRT). Estimation is accomplished using teh EM algorithm. Computation is done in parallel to substantially reduce the amount of time needed for the EM algorithm. Bayes modal estimates of item parameters are computed when a prior distribution has been included in an item response model. The priors maye be specified for any or all item parameters. Any item can use priors. As such, you can have some items with one or more priors and other items without any priors.

Constructor Summary

Constructors

Constructor and Description
MarginalMaximumLikelihoodEstimation(ItemResponseVector[] responseVector, ItemResponseModel[] irm, DistributionApproximation latentDistribution)

Method Summary

Modifier and Type	Method and Description
void	addEMStatusListener(EMStatusListener listener)
double	completeDataLogLikelihood() Computes the complete data log-likelihood.
void	computeG2ItemFit(int nbins, int minExpectedCount)
void	computeItemStandardErrors()
void	computeRaschItemFit()
void	computeSX2ItemFit(int minExpectedCount)
void	estimateParameters(double converge, int maxIter) Estimate parameters using the specified convergence criterion and maximum number of iterations.
void	estimateParameters(double converge, int maxIter, DensityEstimationType densityEstimationType) The EM algorithm for estimating item parameters is conducted with this method.
void	fireEMStatusEvent(int iteration, double delta, double loglikelihood, java.lang.String termCode)
void	fireEMStatusEvent(java.lang.String message)
ItemFitStatistic[]	getFitStatistics()
IrtObservedScoreDistribution	getIrtObservedScoreDistribution()
ItemResponseModel	getItemResponseModelAt(int j)
DistributionApproximation	getLatentDistribution()
int	getNumberOfItems()
int	getNumberOfPeople()
java.lang.String	printItemFitStatistics()
java.lang.String	printItemParameters()
java.lang.String	printLatentDistribution()
void	removeEMStatusListener(EMStatusListener listener)
void	setVerbose(boolean verbose)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MarginalMaximumLikelihoodEstimation

public MarginalMaximumLikelihoodEstimation(ItemResponseVector[] responseVector,
                                           ItemResponseModel[] irm,
                                           DistributionApproximation latentDistribution)

Method Detail

estimateParameters

public void estimateParameters(double converge,
                               int maxIter)

Estimate parameters using the specified convergence criterion and maximum number of iterations. If using this call the latent distribution is not estimated.

Parameters:

converge - convergence criterion should be a positivie number close to zero such as 1e-3. If it is a negative number algorithm will run to maximum number of iterations.

maxIter - maximum number of iterations. Algorithm will stop if the maximum is reached evne if convergence criterion is not satistfied.

estimateParameters

public void estimateParameters(double converge,
                               int maxIter,
                               DensityEstimationType densityEstimationType)

The EM algorithm for estimating item parameters is conducted with this method. You can display the iteration number (iter), current maximum change in item parameter estimates (delta), and the value of the loglikelihood from each cycle by using DefaultEMStatusListener. You can also write a custom DefaultEMStatusListener to write this information to a log file.

Parameters:

converge - maximum change in parameter estimate convergence criterion.

maxIter - maximum number of EM cycles.

densityEstimationType - method to estimate (or not estimate) latent distribution

getNumberOfItems

public int getNumberOfItems()

getNumberOfPeople

public int getNumberOfPeople()

getItemResponseModelAt

public ItemResponseModel getItemResponseModelAt(int j)

getLatentDistribution

public DistributionApproximation getLatentDistribution()

completeDataLogLikelihood

public double completeDataLogLikelihood()

Computes the complete data log-likelihood. The kernel of the log-likelihood is computed incrementally in the Estep. This method incorporates the prior probabilities of item parameters into the log- likelihood function.

Returns:

log-likelihood value

computeItemStandardErrors

public void computeItemStandardErrors()

getIrtObservedScoreDistribution

public IrtObservedScoreDistribution getIrtObservedScoreDistribution()

computeSX2ItemFit

public void computeSX2ItemFit(int minExpectedCount)

getFitStatistics

public ItemFitStatistic[] getFitStatistics()

computeG2ItemFit

public void computeG2ItemFit(int nbins,
                             int minExpectedCount)

computeRaschItemFit

public void computeRaschItemFit()

printItemParameters

public java.lang.String printItemParameters()

printLatentDistribution

public java.lang.String printLatentDistribution()

printItemFitStatistics

public java.lang.String printItemFitStatistics()

setVerbose

public void setVerbose(boolean verbose)

addEMStatusListener

public void addEMStatusListener(EMStatusListener listener)

removeEMStatusListener

public void removeEMStatusListener(EMStatusListener listener)

fireEMStatusEvent

public void fireEMStatusEvent(java.lang.String message)

fireEMStatusEvent

public void fireEMStatusEvent(int iteration,
                              double delta,
                              double loglikelihood,
                              java.lang.String termCode)