## Core Features

- Exceptionally
**easy**to use **Statistics guide**built into the software- Includes
**all common statistical methods** - State-of-the-art computational methods for
**fast and reliable**results (e.g. bootstrap around 100 times faster than R) - Can link easily with Microsoft
**Excel** - Delivered and supported directly via the
**web** - Designed for both front line
**research**and**education** - Built upon twenty years of research and development

## Data Import and Export

- Workbooks (similar to Excel spreadsheet)
- Maximum worksheet capacity is 64K columns by 1M rows
- Read and write Microsoft Excel files (2013/2010/2007 or 97-2003)
- Import any common spreadsheet data
- Import text based data (formatted or plain)
- Read/write reports in portable rich text format (RTF) or HTML
- Windows metafile scaleable graphics (can edit components in Word)

## Help

- Statistics guide/text on-line
- Comprehensive scientific references for methods
- Context-sensitive help with results/reports
- Powerful educational tool
- Global academic user base

## Data Preparation

- On-line algebraic calculator
- Apply user-defined functions/formulae to data in worksheet
- Many arithmetic, trigonometric, algebraic and logical functions
- Sort in worksheet (like Excel) or as separate function
- Rotate/transpose blocks of data in worksheet
- Ranks and normal scores (van der Waerden, Blom and expected normal order)
- Combine or split data by group identifier or separate columns
- Transformations (many, including ladder of powers)
- Standardization
- Anthropometric standardization (child growth charts)
- All pairwise differences, means and slopes
- Dummy/design/indicator variable generation
- Convert text to numbers
- Generate random numbers (uniform, normal, chi-square, F, t, binomial, Poisson, gamma, exponential)
- Categorise a continuous variable
- Extract a subset of data by search rules
- Tabulate and detabulate row/column classified data

## Descriptive Statistics

- Univariate descriptive statistics (count, mean, standard deviation, standard error, confidence interval, skewness, kurtosis, median, quartiles, range and a user defined quantile)
- Frequencies
- Tabulations and crosstabs
- Time series summary with area under time-concetration curve for pharmacokinetics

## Confidence Intervals

- Strong emphasis on confidence interval inference
- Exact (profile likelihood) methods used wherever practical
- Additional mid-P coverage given with many intervals
- Help on interpretation

## Graphics/Charts/Plots

- Bar/column charts
- Frequency distribution histograms
- Spread plots, dotplots
- Box & whisker plots
- Normal plots, nomal scores
- Forest (Cochrane, 'blobogram') plots
- Scatter plots, line charts and error bar plots
- Ladder, pairwise change plots
- Agreement plots
- Survival curves
- Residual and diagnostic plots with regressions
- Area under ROC curves
- Population pyramids
- Control (statistical process) charts

## Parametric Methods

- Student's t tests for single, paired (including agreement stats) and unpaired samples
- Normal distribution (Z) tests
- Reference range (normal, log-normal and percentile-based)
- F test, variance ratio test
- Shapiro-Wilk, Shapiro-Francia and Royston tests for (non) normality

## Non-parametric Methods

- Exact P values and confidence intervals
- Mann-Whitney (Wilcoxon rank sum)
- Wilcoxon signed ranks
- Spearman and Kendall rank correlations
- Cuzick's test for trend
- Two sample Smirnov
- Exact confidence intervals for median and other centiles
- Homogeneity of variance, including Breslow-Day test
- Friedman, Cochrane Q, Kruskal-Wallis with multiple contrasts
- Chi-square goodness of fit
- Gini coefficient of inequality with bootstrap confidence intervals
- Simpson and Shannon diversity indices with bootstrap confidence intervals
- LOESS curve fitting
- Nonparametric linear regression

## Regression and Correlation

- Simple linear regression and Pearson's correlation
- Multiple/general linear with best subset selection
- Principal components analysis (PCA) with Cronbach's alpha
- Influential data identification, residual diagnostics and plots
- Grouped linear regression with analysis of covariance
- Various linearized estimates
- Probit analysis (probit or logit), dose response, ED50, LD50
- Polynomial regression with area under curve and back interpolation
- Logistic regression with confidence intervals for cross classification and odds ratios, and bootstrap
- Conditional logistic regression for matched case-control studies
- Poisson regression, relative risk, incidence rate ratio
- Curve fitting
- Cox regression, proportional hazards, hazard ratio
- Kendall and Spearman rank correlations with confidence intervals

## Analysis of Variance

- Randomized Block: one way, two way, two way with repeats
- Multiple comparisons: Tukey(-Kramer), Dunnett, Neuman-Keuls, ScheffĂ© and Bonferroni
- Nonparametric: Kruskal-Wallis and Friedman with multiple contrasts
- Crossover
- Latin squares
- Nested/Hierarchical two way
- Homogeneity of variance
- Analysis of agreement

## Agreement Analysis

- Agreement of continuous data (intra-class correlation etc.)
- Agreement of categorical data (kappa or two or more raters and two or more categories etc.)
- Reliability and reducibility (Cronbach's alpha for scale reliability etc.)
- Berry-Mielke Universal R

## Meta-analysis

- Exact pooled estimates given wherever practical
- Odds ratio
- Peto odds ratios
- Relative risk
- Risk difference
- Effect size (d, g)
- Incidence rate difference and ratio
- Proportion
- Correlation
- Summary data meta-analysis
- Forest plots and exact confidence intervals
- L'AbbĂ© plots
- Funnel plots
- Egger et al. and Begg & Mazumdar bias tests
- I-square confidence intervals and heterogeneity summary

## Survival Analysis

- Kaplan-Meier estimates with survival plots
- Confidence intervals for mean and median survival times
- Follow-up (Berkson-Gage) life tables
- Abridged current life tables
- Logrank test with trend, exact hazard ratios, stratified
- Generalized Wilcoxon tests (Peto-Prentice, Gehan-Breslow or Tarone-Ware weights)
- Wei-Lachin multivariate comparison of two groups
- Cox regression, proportional hazards, multivariate hazard ratio

## Distribution Functions

- Normal, Gaussian
- Student t
- F
- Chi-square
- Studentized range Q
- Binomial
- Poisson
- Spearman rho
- Kendall tau
- Non-central t
- Gamma
- Many other distribution functions behind other methods: hypergeometric, Kolmogorov-Smirnov, Mann-Whitney U etc.

## Chi-square Contingency Tables

- 2 by 2 with confidence interval for odds ratio or relative risk
- 2 by K with trend
- R by C with trend and G square
- McNemar matched pairs (and k by k extension)
- Maxwell (agreement, equivalence)
- Mantel-Haenszel and Woolf with plots
- Generalised Cohrane-Mantel-Haenszel for R by C by K tables
- Goodness of fit

## Exact Tests on Counts

- Fisher exact (2 by 2 tables)
- Fisher-Irwin (R by C tables)
- Gart confidence intervals for odds ratios
- Liddell exact alternative to McNemar for matched pairs
- Sign test, binomial test
- Poisson rate or count confidence interval

## Proportions

- Single
- Paired
- Two independent
- Exact confidence intervals

## Rates

- Exact Poisson confidence intervals for rates or counts
- Exact analysis of rate ratios
- Direct standardization
- Indirect standardization and standardized mortality ratio (SMR)
- Standardize and compare two rates
- Incidence rate meta-analysis

## Sample Size Estimation

- For comparison of means (paired, unpaired)
- For comparison of proportions
- For population surveys
- For survival analysis
- For correlation

## Randomization

- Uses robust random number generator (Mersenne-Twister)
- Intervention-control pairs (with balanced allocation)
- Allocation to independent intervention and control groups
- Block randomization
- Preference group allocation
- Random series of integers from X to Y
- Generate random numbers from distributions (uniform, normal, binomial, Poisson, beta, F, t, chi-square, gamma, lognormal, Cauchy, Weibull, exponential, logistic, negative binomial) for simulation

## Clinical Epidemiology

- Prospective risk: relative risk, risk reduction and population attributable risk
- Retrospective risk: odds ratio, population attributable risk
- Diagnostic test analysis
- Likelihood ratios in diagnostic test series
- Number needed to treat (NNT) with exact confidence intervals
- Screening test error probabilities (Bayes)
- Kappa (optional user defined weights) and Scott's pi agreement
- Incidence rate analysis
- Growth charts (anthropometric z-scores)

## System Requirements

- Any version of Microsoft Windows that is supported by Microsoft running .Net Framework version 6.1 or later
- At least 1GB memory free
- Some users run StatsDirect on a Mac with Microsoft Windows installed using virtualisation (e.g. via Vmware Fusion or Parallels)
- Previous StatsDirect version 2 is available for old Windows NT, 2000, XP, Vista, or 7 without Service Pack 1, including 32-bit versions