Statistical Commands in Blossom

Blossom currently has six statistical commands, MRPP, SP, MEDQ, LAD, OLS, and COV. The MRPP command can specify one of three multiresponse permutation procedures.

1) Multiresponse permutation procedures (MRPP)

2) Multiresponse randomized block permutation procedures (MRBP)

3) Permutation tests for matched pairs (PTMP)

These procedures (MRPP, MRBP, and PTMP) are distribution-free techniques for making inferences about grouped data. Their advantages over many classical techniques include the ability to select an analysis space commensurate with the geometry of the data as perceived by the investigator. Several classical univariate and multivariate parametric and rank tests can be emulated with these procedures as well. The simplest MRPP analysis is for data consisting of two or more observations on objects in two or more groups. The MRBP and PTMP variants are for similar data that are blocked or paired.

Since the MRPP command can emulate so many different statistical tests, the specification of the command line can be quite complex. However, Blossom uses default values, which for routine analysis makes the command easy to use.

The MEDQ command calculates univariate or multivariate medians and distance quantiles either by groups specified by a grouping variable or for the entire data file being used. Options allow you to specify quantiles to report that differ from the default quantiles.

The SP command calculates the multiresponse sequence procedure to test for first-order autoregressive patterns (serial dependency). The default value produces an analysis in Euclidean space. A sequencing variable that determines the order of the data can be selected or Blossom assumes by default that the order in the file is the sequential order of interest.

The LAD command estimates a least absolute deviation regression or an optional quantile regression. The model specified in the LAD command line is considered the full parameter alternative model for hypothesis tests. The associated command, HYPOTHESIS, can be used to specify a reduced parameter null model that is tested against the model specified by the LAD command.

The OLS command estimates an ordinary least squares regression. It has an associated HYPOTHESIS command that performs a similar function in testing hypotheses as the associated HYPOTHESIS command does with the LAD command.

The COV command provides for tests of g-sample empirical coverage tests if used with a grouping variable and related goodness-of-fit tests if specified without a grouping variable.

The MRPP variants, MRSP, LAD, OLS, and COV are discussed in turn. MEDQ is discussed with MRPP as it provides descriptive estimates that are useful for interpreting results of hypothesis tests with MRPP.

Statistical Commands