Thursday, April 7, 2011

Aanalysis or a limited set of tests statistical Packges1

PS -- a well-implemented Windows program for power and sample size calculations from Vanderbilt Univ Med Ctr. Handles dichotomous, continuous,or survival response measures, which are analyzed by chi-square or Fisher Exact tests, Student t tests, and log-ranks tests, respectively. The alternative hypothesis may be specified either in terms of differing response rates, means, or survival times, or in terms of relative risks or odds ratios. Studies with dichotomous or continuous outcomes may involve either a matched or independent study design. The latest version also handles Mantel-Haenszel tests. Can determine sample size for a specified power, power for a specified sample size, or the specific alternative hypotheses that can be detected with a given power and sample size. Produces graphs of relationships between power, sample size and detectable alternative hypotheses (with any two of these variables on x & y, and the third variable generating a family of curves on a single graph). Linear or logarithmic axes may be specified. Can print professional-quality power charts. The latest version also provides a concise and precise verbal description of each power analysis, which you can copy and paste into the Power/Sample-Size section of your proposal or protocol.

G*Power 3 -- a very general Power Analysis program for Windows and Macintosh. Performs exact analysis for 6 types of correlation tests, 3 types of bivariate regression tests, 1-group and 2-group comparison of means tests (parametric and non-parametric), 4 types of multiple regression tests, logistic regression, poisson regression, ordinary and repeated-measures ANOVAs, ANCOVAs, MANOVAs, multivariate T2 and MANOVAs, 8 types of tests of proportions (McNemar, Fisher, etc.), 1-group and 2-group variance tests, and completely generic tests involving the binomial, normal, t, chi-square, and F distributions. Computes power, sample sizes, alpha, beta, and alpha/beta ratios. Has a comprehensive web-based tutorial and reference manual.

Factor  -- a comprehensive factor analysis program. Provides univariate and multivariate descriptive statistics of input variables (mean, variance, skewness, kurtosis), Var charts for ordinal variables, dispersion matrices (user defined , covariance, pearson correlation, polychoric correlation matrix with optional Ridge estimates). Uses MAP, PA (Parallel Analysis), and PA - MBS (with marginally bootstrapped samples) to determine the number of factors/components to be retained. Performs the following factor and component analyses: PCA, ULS (with Heywood correction), EML, MRFA, Schmid-Leiman second-order solution, and Factor scores. Rotation methods: Quartimax, ,Varimax , Weighted Varimax, Orthomin , Direct Oblimin, Weighted Oblimin, Promax, Promaj , Promin, and Simplimax. Indices used in the analysis: dispersion matrix  tests (determinant, Bartlett's, Kaiser-Meyer-Olkin), goodness of fit: Chi-Square ,non-normed fit index, comparative fit index, goodness of fit index, adjusted GFI, RMS error of approx, and estimated non-centrality parameter (NCP), reliabilities of rotated components , simplicity indices: Bentler’s, and loading simplicity index. Provides mean, variance and histogram of fitted and standardized residuals, and automatic detection of large standardized residuals.

KEYFINDER -- a menu-driven interactive program for generating, randomizing and tabulating blocked and/or fractional-replicate factorial designs in completely general situations. It can generate blocked and/or fractional-replicate designs with userspecified confounding and aliasing properties. KEYFINDERruns on all versions of Windows. You can download the Version 3.3 Overview document, in PDF format, here. To obtain a free copy of the program and manual, send an e-mail to the author: Peter.Zemroch@shell.com

Weka -- a collection of machine learning algorithms for data mining tasks, implemented in Java. Can be executed from a command-line environment, or from a graphical interface, or can either be called from your own Java code. Weka contains tools for data pre-processing, classification, regression, clustering, association rules, and visualization, and is well-suited for developing new machine learning schemes.

StatCalc -- a PC calculator that computes table values and other statistics for 34 probability distributions. Also includes some nonparametric table values, tolerance factors, and bivariate normal distribution. A help file is provided for each distribution.

Scientific Calculator - ScienCalc program contains high-performance arithmetic, trigonometric, hyperbolic and transcendental calculation routines. All the function routines therein map directly to Intel 80387 FPU floating point machine instructions.

Distributions -- Windows program allows for the analysis of discrete single dimension distributions. The program is based on various manipulations of the poisson, binomial and hypergeometric distribution. Available are the probability of an observed number of cases given a certain null hypothesis, the calculation of exact poisson, binomial or hypergeometric confidence intervals, the exact and approximate size of a population using catch-recatch methodologies, the full analysis of a Poisson distributed rate ratio, Fieller analysis, and two versions of the negative binomial distribution can be used in various ways. Beside the exact procedures there are also various approximate procedures available. From the Downloads section of the QuantitativeSkills web site.

Multinomial -- This Windows program is the exact solution to the Chi-square Goodness of fit test of testing for a difference between an observed and an expected distribution in a one-dimensional array. For example, the test can be used to compare the distribution of diseases in a certain locality with an expected distribution on the basis of national or international experiences using an ICD classification. In a two-category array the multinomial test provides a two-sided solution for the Binomial test. For example, Multinomial {10 20 0.20 0.80} gives the two-sided probability (0.105) for the single sided Binomial {0.20 10 30} probability (0.061). The multinomial allows you to work with empty '0' observation cells although you must have an expectation about a cell. From the Downloads section of the QuantitativeSkills web site.

Tables -- a Windows program for the analysis of tables with up to 2*7 and 3*3 cells. The program allows for exact and approximate statistics to be calculated for traditional, ordinal and agreement tables. Fisher exact, Number Needed to Treat, Proportional Reduction in Error Statistics, Normal Approximations, Four different Chi-squares, Gamma, Odds-ratio, t-tests and Kappa are among the many statistical procedures available. From the Downloads section of the QuantitativeSkills web site.

MorePower -- another well-implemented power/sample-size calculator for any ANOVA design, for 1- and 2-sample t-tests, and for 1- and 2-sample binomial testing (sign test, chi-square test).

EqPlot -- Equation graph plotter program plots 2D graphs from equations. The application comprises algebraic, trigonometric, hyperbolic and transcendental functions.

BlockTreat -- a Java program that implements a very general Monte Carlo procedure that performs non-parametric tests (based on random permutations, not ranks) for block and treatment tests, tests with matching, k-sample tests, and tests for independence between any two random variables. Designs may be incomplete and unbalanced, or even have supernumerary entries. The tests are "exact", in the Monte-Carlo sense -- they can be made as accurate as desired by specifying enough random shuffles.

PCP (Pattern Classification Program) -- a machine-learning program for supervised classification of patterns (vectors of measurements). PCP implements: Fisher's linear discriminant, dimensionality reduction using SVD, PCA, feature subset selection, Bayes error estimation, parametric classifiers (linear and quadratic), L-S (pseudo-inverse) linear discriminant, k-Nearest Neighbor, neural networks (Multi-Layer Perceptron), SVM, model selection for SVM, cross-validation, and bagging (committee) classification. Supports interactive (keyboard-driven menus) and batch processing.

PEPI -- a collection of 43 small DOS / Windows programs that perform a large assortment of statistical tests. They can be downloaded individually, or as a single ZIP file. (A new Windows version is being developed; the test version can be downloaded here.) They were written to accompany the book Computer Programs for Epidemiologic Analyses: PEPI v. 4.0, by Abramson and Gahlinger, which is available for purchase. A freely-accessible article describing the new features of WinPEPI can be accessed here. The programs include: p-value adjustments for multiple significance tests; Attributable and Prevented Fractions: Case-Control Studies; Analysis of 2 x 2 Tables; Chi-square Tests of Association; Combining Measures of Association or Probabilities; Confidence Intervals; Aids to Use of Pearson's Correlation Coefficients; ifference Between Rates, Proportions or Means; Direct Standardization; Exact Test for a 2 x K Table; Tests for Goodness of Fit ; Fitting of Poisson and Binomial Distributions; Appraisal of Frequency Distribution ; Indirect Standardization; Agreement Between Categorical Ratings; Life Table Analysis;  Logistic Regression Analysis (Unconditional and Conditional); Wilcoxon-Mann-Whitney Test and Related Procedures ; Extended Mantel-Haenszel Procedure: Trend Analysis; Multiple Matched Controls; Correcting for Misclassification in 2 x 2 Tables; Analysis of Paired Samples ; Poisson Probability: Observed vs Expected Events; Poisson Regression Analysis; Power of a Test Comparing Two Proportions or Means; Probability and Inverse Probability Values: Z, t, Chi Square, F; Procedures using Random Numbers; Association Between Ordinal-Scale Variables; Comparison of Two Rates or Proportions; Comparison of Person-Time Incidence Rates; ower and Sample Size for Regression and Correlation Analyses; Comparison of Several Related Samples; Sample Size for Estimation of Proportion, Rate, or Mean; Sample Sizes for Comparison of Two Samples ; Internal Consistency of a Scale; Screening and Diagnostic Tests ; Seasonal Variation ; Smoothing of Curves and Median Polish Procedure; Kaplan-Meier Life Table Analysis, Log-rank and Logit-rank Tests; Calculation of Elapsed Time; Trend Analysis and Multiple Comparisons, and two special calculators: WHATIS and WHATS.

TETRAD (from the TETRAD Project at CMU) -- a free program for creating, simulating data from, estimating, testing, predicting with, and searching for causal/statistical models of categorical (or ordinal) data and to linear models ("structural equation models') with a Normal probability distribution, and to a very limited class of time series models. Provides sophisticated methods in a friendly interface. It performs many of the functions in commercial programs such as Netica, Hugin, LISREL, EQS and other programs, and many discovery functions these commercial programs do not perform. TETRAD is limited to models The TETRAD programs describe causal models in three distinct parts or stages: a picture, representing a directed graph specifying hypothetical causal relations among the variables; a specification of the family of probability distributions and kinds of parameters associated with the graphical model; and a specification of the numerical values of those parameters.

No comments:

Post a Comment