The unit STATIS offers some of the most fundamental procedures of mathematical statistics. Along with the Gamma and the Beta function, all major types of distributions are implemented (normal, t, F, and chisquare distribution). A major benefit of this package is that the distribution functions are calculated analytically and not by approximations, or by table lookup. Thus the precision is at least eight digits.
Further, the unit STATIS offers many often needed statistical tests, a cross validation engine, a random generator, as well as classes to create multiple linear regression and partial least squares models.
More details can be found in the online help pages.
The unit STATIS is part of the SDL Component
Suite.
Delphi Bits  Fast Calculation of Quantiles  An efficient way to speed up the calculation of quantiles. 
SDL TechNotes  Kurtosis and skewness do not match the values of Excel  The calculations of statistical parameters differs dedpending on whether you look at a sample or at a population. 
DataLab  DataLab uses many routines of the SDL Component Suite  If you want to test the statistical and mathematical routines available within the SDL Suite without writing a single line of code you could download the evaluation copy of DataLab and play with your data before you start programming your particular application. 
What's new:
 Release 12.0 [Dec12, 2023]
 the new class TMLRModel supports the creation of MLR (multiple linear regression), LDA (linear discriminant analyis) and RR (ridge regression) models
 implemented the function AkaikeInfCrit
 implemented the function GiniCoeff to calculate the Gini coefficient
 implemented the function TheilSenEstimator for robust linear regression
 implemented the properties TCrossValidator.FNCnt, TCrossValidator.FPCnt, TCrossValidator.TNCnt, and TCrossValidator.TPCnt
 implemented the two properties TCrossValidator.FNRate and TCrossValidator.TNRate
 implemented the BartlettTest
 implemented the binomial sign test BinomSignTest
 implemented a test for kurtosis: CalcAndTestKurtosis
 implemented a test for symmetry of distributions: CalcAndTestSkewness
 implemented PerformSiegelTukeyTest to perform a SiegelTukey test for equal variances
 implemented a second overloaded version of PerformKSNormalityTest to support dynamic arrays
 the function GrubbsTest has been changed and extended to support dynamic arrays as well
 the new function CorrTest tests whether the correlation of two datasets is significantly different from a reference value
 implemented the function PearsonCorrCoeff for calculating Pearson's correlation coeffcient from two arrays
 the class TPLSModel has been extended by the property CvdMCC
 the class TCrossValidator provides now the array property MatthewsCorr
 implemented a second overloaded version of PerformMannWhitneyUTest
 implemented a third overloaded version of Perform2SampleTTest
 implemented PerformFTest, Perform1SampleChi2Test, Perform1SampleTTest and WilcoxonSignedRankTest
 the function PerformLillieforsTest is now available in a second overload version supporting open arrays
 implemented an overloaded version of DeanDixonTest and of ShapiroWilkTest to support open arrays
 the function CFInformedness returns the informedness of a binary classifier
 the functions CFAccuracy and CFMatthewsCorr calculate the classifier accuracy and Matthew's correlation
 the function ExtractClassifPerformance extracts a classifiers performance from a multiclass confusion matrix
 the new members TPLSModel.SetObjTags and TPLSModel.ObjTag support object tags
 the new members TPLSModel.ObjAttrib, TPLSModel.NameObj and TPLSModel.SetObjAttribs support object attributes and object names in PLS models
 bug fix: BinomDistriQuantile did not return the ActualProb value
 bug fix: FisherExactTest changed the frequency table
 bug fix: the function UDistriIntegral returned a wrong probability (actually returned p[U1] instead of p[U])
 Release 10.7 [Aug24, 2020]
 the new function MedianTest implements the median test
 the new functions Chi2OfContTab and Chi2OfContTabYates calculate the chisquare value of a contingency table
 the new class TCFEvaluator provides an easy way to calculate binary classifier metrics
 the new function AdjustedRandIx calculates the adjusted Rand index
 the new function FriedmanTest provides Friedman's test for homogeneity of groups
 the new function LeveneTest provides Levene's test for equal variances
 the new function Lambda calculates the Lambda value of contingency tables
 the new functions DurbinWatson and DurbinWatsonCrit5pct support the application of a DurbinWatson test for serial correlation.
 the new function Prob2x2Contingency calculates the probability of a contingency table
 the new function FisherExactTest calculates Fisher's exact test
 the new function PerformChi2DistComp performs a chi^{2}test to compare empirical distributions
 bug fix: TCrossValidator did not create the matrices XMat and YMat
 bug fix: TCrossValidator.Execute did not correctly recognize the cases when calculating TPRate and FPRate
 Release 10.6 [Mar09, 2018]
 no changes
 Release 10.5 [Oct10, 2016]
 the function PercentileOfArray calculates the percentile of an array of data values
 Release 10.4 [Jun02, 2015]
 the new method TPLSModel.AppendModelParamsToFile stores the model parameters in readable format in an open text file
 the new functions ExactRunsTest and RunsTest implement the WaldWolfowitz runs test
 the new function RunsTestSerial applies the WaldWolfowitz test to a vector of numbers
 Release 10.3 [Oct06, 2014]
 the readonly property TCrossValidator.IsExecuting returns TRUE while the crossvalidation is executing
 the new function Anova1F performs a onefactorial ANOVA (analysis of variances)
 Release 10.2 [May30, 2013]
 the new array properties TCrossValidator.FPRate and TCrossValidator.TPRate return the false positive and true positive rates of binary classifiers
 the new array properties TPLSModel.CvdFPRate and TPLSModel.CvdTPRate return the crossvalidated false positive and true positive rates of binary PLSDA classifiers
 Release 10.1 [Oct29, 2012]
 the new function ShapiroWilkTest performs the ShapiroWilk test for normality
 the function ShapiroWilkIntegral calculates the integral of the W distribution
 the function LoadPLSModelComment allows to browse PLS models on disk
 the property TCrossValidator.RMSEP is now an array property returning der rms errors of prediction of the individual response variables.
 the method TPLSModel.SaveModelCoefficients now stores the number of used factors, as well
 The array property TPLSModel.CvdRmsEP returns the individual root mean square errors of prediction of all response variables.
 The vector property ClassifThreshold returns the optimum classifier thresholds for PLS discriminant analysis
 The property IsDiscriminantModel allows to switch between PLS regression and PLS discriminant analysis
 the function WilcoxonSRQuantile calculates the critical threshold of the Wilcoxon signed rank test
 the new functions KSPValue and KSQuantile calculate the pvalues and the quantiles of the KolmogorovSmirnov test for normality
 the function LillieforsQuantile returns the critical value for the Lilliefors test
 the function LillieforsPValue estimates the p value for the Lilliefors test
 the functions PerformKSNormalityTest and PerformLillieforsTest calculate the KolmogorovSmirnov statistic and the Lilliefors statistic of a data vector
 the function LnBinomCoeff returns now a value of 1 for invalid parameters
 bug fix: TCrossValidator did not select a random sample but a consecutive block
 Release 10.0 [Oct4, 2011]
 new class TPLSModel provides the computational engine for partial least squares (PLS) analysis
 the new class TCrossValidator allows to cross validate arbitrary statistical models
 Release 9.7 [May31, 2010]
 two functions for appyling the Grubbs outlier test implemented: GrubbsCriticalValues and GrubbsTest
 bug fix: FDistriQuantile calculated incorrect quantiles for p values less than 0.5 and unequal degrees of freedom
 Release 9.51 [Dec01, 2008]
 no changes
 Release 9.5 [Oct31, 2008]
 new method EstimateProbDensity implements a variable kernel method to estimate densities of univariate distributions
 new method TRandGen.Standardize implemented
 TRandGen.tDistri, TRandGen.FDistri, and TRandGen.Chi2Distri implemented
 DeanDixonTest has been modified to test for the element which lies most outside of the bulk.
 bug fix: TRandGen.Random does not hang if all probabilities are set to zero values
 Release 9.0 [May28, 2007]
 no changes
 Release 8.5 [Feb18, 2006]
 function DeanDixonCriticalValues provides critical values of the DeanDixon outlier test
 new function DeanDixonTest performs the DeanDixon outlier test
 UDistriQuantile, UDistriIntegral, and UDistriDensity implemented
 MannWhitney UTest implemented (PerformMannWhitneyUTest)
 TSignifLevelEx and NumericSigLevel extend definition of levels of significance
 hypergeometric distribution implemented: HyperGeoDistriDensity
 KendallsTau implemented (calculates Kendall's taub rank correlation coefficient)
 KruskalGamma implemented (calculates Kruskal's gamma value)
 SpearmanRankCorr implemented (calculates Spearman's rank correlation
 Release 8.2 [Nov03, 2004]
 no changes
 Release 8.1 [Aug29, 2004]
 available for Delphi™ 8 for .NET
 class TRandGen moved from unit MATH1 to unit STATIS
 Release 8.0 [Apr15, 2004]
 new function Perform2SampleTTest for performing a twosample ttest implemented
 error function Erf and ErfApprox implemented
 bug fix: function IncompleteGamma is now declared in the interface section
 Release 7.2 [Mar23, 2003]
 STATIS is now part of MathPack
 improvement: Chi2DistriDensity does not crash anymore if called with invalid parameters
 Release 7.0 [Sep15, 2002]
 available for Delphi™ 7.0
 the unit is now CLX compatible
 Release 6.5 [May28, 2002]
 available for C++Builder™ 6.0

Release 6.0 [Aug06, 2001]
 available for Delphi™ 6.0

new functions implemented: FisherTransform, FisherTransformInv,
ConfidenceInterval

Release 5.5 [May01, 2000]

Release 5.0 [Oct07, 1999]

STATIS is now available for Delphi™ 5.0

Release 1.1 [Mar27, 1999]
 STATIS is now available for C++Builder™ 4.0
available for C++Builder™ 5.0