1. Highlights
● produced in 2003
● 4 new directives, 21 new procedures and 9 new functions
● binomial and Poisson tests (BNTEST, PNTEST)
● statistical process control (SPCAPABILITY, SPCCHART, SPCUSUM, SPEWMA, SPPCHART, SPSHEWHART)
● predictions in REML (VPREDICT)
● analysis of circular data (CDESCRIBE, DCIRCULAR, WINDROSE)
● response-surface designs (AFRESPONSESURFACE)
● survival-analysis enhancements (RLIFETABLE, RSTEST)
2. What’s new
2.1 Directives
AFRESPONSESURFACE uses the BLKL algorithm to construct designs for estimating response surfaces.
BASSESS assesses potential splits for regression and classification trees.
VCYCLE controls details of the REML algorithm.
VPREDICT forms predictions from a REML model.
2.2 Procedures
BNTEST calculates one- and two-sample binomial tests.
CDESCRIBE calculates summary statistics and tests of circular data.
DCIRCULAR plots circular data.
FACSORT sorts the levels of a factor according to an index vector.
FSTRING forms a single string from a list of strings in a text.
PNTEST calculates one- and two-sample Poisson tests.
PRWILCOXON calculates probabilities for the Wilcoxon signed-rank statistic.
RCOMPARISONS calculates comparison contrasts amongst regression means.
RLIFETABLE calculates the life-table estimate of the survivor function.
RSTEST compares groups of right-censored survival data by nonparametric tests.
SPCAPABILITY calculates capability statistics.
SPCCHART plots c or u charts representing numbers of defective items.
SPCUSUM prints CUSUM tables for controlling a process mean.
SPEWMA plots exponentially weighted moving-average control charts.
SPPCHART plots p or np charts for binomial testing for defective items.
SPSHEWHART plots control charts for mean and standard deviation or range.
SVSTRATIFIED analyses stratified random surveys by expansion or ratio raising.
TABSORT sorts tables so their margins are in ascending or descending order.
TALLY forms a simple tally table of the distinct values in a vector.
WINDROSE plots rose diagrams of circular data like wind speeds.
XOPOWER estimates the power of contrasts in cross-over designs.
2.2 Functions
MFRACTION(x;p;m) |
returns the period within a month that date-time value x belongs to; p is the length of the period (e.g. 5 for pentade, 10 for decade), and m is the starting month (default 1). |
|---|---|
NVRESTRICTED(x) |
synonym of NVALUES. |
NVUNRESTRICTED(x) |
number of values of x ignoring restrictions (i.e. gives the full length of x). |
RMEANS(x;p;q) |
running means of x using a window around each unit that includes p preceding and q succeeding observations; no default for p (must be set) default for q is 0. |
RNOBSERVATIONS(x;p;q) |
number of observations contributing to computation of running mean or total involving p preceding and q succeeding observations about each unit of x; no default for p (must be set) default for q is 0. |
RTOTALS(x;p;q) |
running totals of x using a window around each unit that includes p preceding and q succeeding observations; no default for p (must be set) default for q is 0. |
RUNS(x) |
length of run of values up to each unit in x. |
VEC(x) |
stacks columns of a matrix x into a single variate (VEC operator). |
VECH(x) |
stacks columns of the lower triangle of a matrix x (VECH operator). |
3. What’s changed
Most of the changes are completely compatible with Release 6.1, the previous release. There are a few commands, however, where new options or parameters have been inserted into the existing lists. These may cause problems in statements where option or parameter names have been omitted or abbreviated To avoid any difficulty, the name of the option/parameter after the new option/parameter should be given explicitly, and not abbreviated to fewer than four characters. There are also a few commands where options or parameter have been renamed, or their actions have been modified.
Any command, where changes in Release 7.1 may cause incompatibilities in existing programs, is marked below by the symbol †.
3.1 Directives
CVA can now save the adjustment terms (using new parameter ADJUSTMENTS, added at the end of the parameter list).
KRIGE can now save the Lagrange multipliers from the kriging solutions (using new parameter LAGRANGEMULTIPLIER, added at the end of the parameter list). If independent information is available about measurement error, you can now use this in punctual kriging (using new parameter MEASUREMENTERROR, added after LAGRANGEMULTIPLIER).
†RKEEP can now calculate exact confidence intervals for nonlinear models based on the likelihood surface rather than by assuming a quadratic surface around the optimum (new option CIMETHOD inserted before SAVE).
3.2 Procedures
The SEED option settings have been redefined in the design selection procedures to match usage e.g. in the RANDOMIZE directive. Standardized residuals now available instead of simple residuals diagnostic plots from analyses of variance. New selection methods are available for classification trees. Accelerated confidence limits are now available for bootstrap statistics. Exact probabilities are now available for Wilcoxon tests. The procedures concerned are listed below, with asterisks against those that may cause incompatibilities in programs from earlier releases.
AFCARRYOVER and AGCROSSOVERLATIN: CARRYOVER parameter renamed CARRYOVERFACTOR.
†AGALPHA, †AGBIB, †AGBOXBEHNKEN, †AGCENTRALCOMPOSITE, †AGCROSSOVERLATIN, †AGCYCLIC, †AGDESIGN, †AGFRACTION, †AGHIERARCHICAL, †AGLATIN, †AGMAINEFFECT, †AGNEIGHBOUR, †AGSEMILATIN, †AGSQLATTICE: change to the SEED option. A zero value now continues any existing sequence of random numbers or, if none, selects a seed automatically using the system clock. Negative values suppress randomization. Default when running the procedures in batch is now -1 instead of 0.
†APLOT: new RMETHOD option (inserted before INDEX) allows you to choose standardized residuals instead of simple ones.
†AUNBALANCED and AUDISPLAY: new RMETHOD option (as final option) to allow you to choose standardized residuals instead of simple ones, and TPROBABILITY (inserted between TPROBABILITY and PLOT) to print probabilities for t statistics.
†BCLASSIFICATION: new options METHOD (between PRINT and GROUPS) to request use of the mean posterior improvement criterion instead of Gini information to select characters, and ANTIENDCUTFACTOR (at end of option list) to request adaptive anti-end-cut factors; see Taylor & Silverman (1993). New parameter ORDERED (at end of parameter list) to indicate whether factors for use in the tree have ordered or unordered levels.
†BOOTSTRAP: new option CIMETHOD (between METHOD and VCOVARIANCE) to select bias-corrected and accelerated comfidence intervals.
BREGRESSION: new parameter ORDERED (at end of parameter list) to indicate whether factors for use in the tree have ordered or unordered levels.
†CUMDISTRIBUTION has been rewritten to allow for groups. The new version will fit all the models provided by the previous version, except for those using the calculation setting of TRANSFORMATION (which is now no longer available). Several parameters and options have been added to make the procedure more flexible and to extend the models that can be fitted. Option DENSITY has been renamed DISTRIBUTION, and option PARAMETER has been deleted as it is no longer applicable. The shift transformation of the previous version is now achieved by setting the new LAG option. Graphs are now in high resolution, and controlled by the PLOT option. Parameter NUNITS has been replaced by the more comprehensive SAVE parameter. NPOPULATIONS now has an upper limit of 3. The form of the DATA input is more flexible: counts need not be accumulated, and the first count need not be 0. Initial values are no longer required. Results may change slightly as the new procedure removes some constraints due to limits in the mathematical functions available in previous releases of GenStat. In addition, degrees of freedom may vary, as the new procedure now explicitly includes a count of 0 at TIME=0 if no DATA are supplied for TIME=0.
JACKKNIFE: new parameter ACCELERATION (at end of parameter list) to save the acceleration parameter requires to calculate accelerated confidence limits for bootstrap statistics.
LRVSCREE: can now generate high-resolution plots of the latent roots (using new options PLOT, TITLE and WINDOW, added at end of option list).
†MVARIOGRAM: can now fit cubic and stable (i.e. powered exponential) models (new settings cubic and stable for the MODEL option, and new option SMOOTHMESS added between MODEL and WEIGHTING).
PROBITANALYSIS: PRINT option has a new setting effectivedoses to control printing of effective (or lethal) doses.
†RPAIR: a new setting estimable has been added to the COMBINATIONS option and is now the default (as in the PREDICT directive).
†TTEST: the CIPROB option has been renamed CIPROBABILITY. The INTERVAL parameter has been replaced by new parameters LOWER and UPPER.
†WILCOXON: NORMAL parameter replaced by PROBABILITY parameter (to save probabilities, calculated using the new PRWILCOXON procedure, rather than Normal approximations).
†XOEFFICIENCY: NPERIOD and CONTRASTS options renamed NPERIODS and CONTRASTTYPE.
3.3 Functions
INVERSE now works on diagonal matrices (as well as square and symmetric matrices).
GINVERSE now works on diagonal matrices (as well as square and symmetric matrices).
NDAYINYEAR(x;m) provides the number of the day in year corresponding to date-time value x, and starting the year at the beginning of month m (default 1).
NWEEKINYEAR(x;s) is the number of the week through the year for date-time value x. The default setting for s is 'iso'; this uses the definition of ISO Standard IS-8601 (1988) in which any week (starting on Monday) that lies in more than one year is assigned a week number for the year in which most of its days occur. The alternative setting, 'simple', takes the first week of the year as the one containing 1st January.