SPSS, Statistica - это все понятно, но существуют и другие, которые по своим возможностям (читай +) не уступают этим мастадонтам от математической статистики
Есть различного рода надстройки к MS Excel (мне вот понравились):
DeltaX for Excel 2000 (выявление достоверности различий выборочных средних)
StepRegression для Excel 2000 (линейный регрессионный анализ)
взять можно тут [
www.excelstat.newmail.ru]
Есть еще отличная программа KyPlot ver. 2 - бесплатна, а вот уже ver. 4. - стоит 400$.
там, где была, уже взять бесплатную нельзя, но кому надо, есть у меня на сайте
Даже вторая версия может:
Basic Statistics:
Calculates basic statistics parameters for data in the input range.
t-Test:
Performs a student's t-test for unpaired comparison, paired comparison or the mean of a group.
ANOVAs:
Performs one-way ANOVA or two-way ANOVA with or without replication.
Simple Regression Analysis:
Performs linear regression analysis for X, Y pairs based on the least squares method.
Model: Y = b0 + b1 * X + error
Standard Line Assay
Inverse estimation of X from Y in simple regression analysis.
Model: Y = b0 + b1 * X + error.
Given standard measurements (with replication), estimate X values for Y sample measurements.
Polynomial Regression
Fits the data to the polynomial equation by the least squares method.
Model: Y = b0 + b1 * X + b2 * X^2 + b3 * X^3... + error.
Multiple Regression Analysis
Fits the data to the multiple regression model by the least squares method.
Model: Y = b0 + b1 * X1 + b2 * X2 + b3 * X3... + error.
Multiple Comparisons
Parametric multiple comparisons:
1) Tukey test (refs.1,2)
Performs pairwise comparisons for one-way layout design.
The method for unbalanced cases is often called Tukey-Kramer test.
2) Dunnett test (ref.3)
Performs comparisons with a control.
Nonparametric multiple comparisons:
1) Steel-Dwass test (refs.4,5)
Performs pairwise comparisons for one-way layout design.
(Tukey-equivalent nonparametric test)
2) Steel test (ref.6)
Performs comparisons with a control.
(Dunnett-equivalent nonparametric test)
The calculation procedures for multiple comparisons are according to ref. 7.
The programs for calculating the studentized range distribution and multivariate t-distribution are based on ref. 8.
Factor Analysis
Options for Factor Analysis
Data Type
When "Correlation Matrix" or "Covariance Matrix" is selected, the lower triangular part is read and used as input.
Method
1) Non-iterated Principal Factor
When "Unity" is selected for "Initial Communalities", this gives the "Principal Component" solution (ref. 9).
With this setting, you can obtain a rotated solution of the principal components.
2) Iterated Principal Factor
When converged, this method gives the same solution as the unweighted least squares method.
3) Unweighted Least Squares (ULS)
The algorithm by Joereskog (ref. 10) is used.
4) Maximum Likelihood
The algorithm by Jennrich-Robinson (ref. 11) is used.
Optimization for both the "Unweighted Least Squares" and "Maximum Likelihood" methods is done using a sequential quadratic programming method with constraints 0<= uniquenesses <=1.
Initial Communalities
selects initial values of communality.
1) Unity: all initial communalities are set to one (uniquenesses=zero);
to be used with "Non-iterated Principal Factor" to obtain principal component solutions.
2) Squared Multiple Correlation (SMC)
MC(j)=1 - Inv(R)(j, j)
(Inv(R): inverse of the correlation matrix)
3) Joereskog's formula: =1 - {1 - m / (2p)} / Inv(R)(j, j)
(m: number of factors; p: number of variables)
Rotation
1) Orthomax rotations with a weight (W): Varimax (W=1); Quartimax (W=0); Biquartimax (W=0.5);
Equamax (W=M/2); and General Orthomax with arbitrary W.
2) Oblimin rotations for Simple Structure:
Covarimin (W=1); Quartimin (W=0); Biquartimin (W=0.5); and General Oblimin with arbitrary W.
3) Oblimin rotations for Simple Pattern: gives solutions corresponding to "direct" oblimin method.
4) Procrustean rotations: Orthogonal Procrustes; Oblique Procrustes with Target Pattern; and Oblique Procrustes with Target Structure:
performs Procrustean rotations with a user-given matrix as the target pattern or structure matrix;
you have to specify the location of the first element of the target matrix in the "Target Matrix for Procrustes" frame.
Normalize Loadings
applies to orthomax and oblimin rotations.
When checked, "normal" varimax, etc is performed.
When unchecked, "raw" varimax, etc is performed.
Normalize Target
applies to Procrustean rotations.
When checked, the target matrix is normalized.
When unchecked, the raw target matrix is used.
Promax Rotation
1) No Rotation
2) For Simple Pattern
3) For Simple Structure
first performs one of the orthomax rotations in the "Rotation" menu (usually varimax), raises the elements of the rotated loadings to some power, specified by "Power", and then performs an oblique Procrustean rotation with the modified loadings as the target pattern matrix (2) or structure matrix (3).
Factor Scores
specifies the method to estimate factor scores.
1) No Score Output
2) Regression Method
3) Bartlett's Method
Cluster Analysis
Options for Cluster Analysis
Clustering Method
"Nearest Neighbor", "Furthest Neighbor" and "Group Average" methods:
can be used for all data types and all measures.
"Centroid", "Median", "Ward" and "Flexible" methods:
can be used only for "Raw Data" or "Euclidean Dissimilarity Matrix" data type with "Non-standardized Euclidean" or "Standardized Euclidean" measure.
Output in Squared Distances
When checked, elements in dissimilarity matrix and the value axis of the dendrogram are given in squared distances;
available only for "Raw Data" or "Euclidean Dissimilarity Matrix" data type with "Group Average", "Centroid" or "Ward" method.
