Functions in Streams

Functions in STREAMS

Below is a quick overview of the major functions in STREAMS.

Data in STREAMS are stored in projects, with information about variables and datasets stored in a project dictionary. The project dictionary is stored as an Access database. The database also stores the computed matrices. Imported raw data files are stored in SPSS format.

Data (i. e., raw data files and computed matrices are stored in a hierarchical system with Folder and Datasets. Several functions are available for importing data, inspecting datasets, and deleting datasets. These functions are available on the Data tab of the so called Project Window:

Importing Data and Computing Matrices

The STREAMS Data Wizard guides the user through the process of importing data.

The following main functions are available for getting data into STREAMS:

	Importing of raw data and computation of covariance matrices with data input in SPSS-, PRELIS- and Asci-format.
	Computation of a single covariance matrix, with facilities for selecting subsets of cases.
	Computation of separate covariance matrices for each code value of a classification variable.
	Computation of one covariance matrix for each combination of missing values.
	Computation of matrices for two-level analysis, including calculation of intraclass correlations
	Importing of raw data for a subset of cases.
	Importing of separate raw data files for each code value of a classification variable.
	Importing of externally computed covariance, correlation, and polychoric correlation matrices. For the latter type of matrices weight matrices, if any, are imported as well.
	Imputation of missing values through substitution with means, computed either for the total group of for subsets.

Other Data Functions

	View computed matrices or imported raw data.
	View listings with descriptive results from computation of matrices.
	Remove datasets and folders.

Top

Tools for Model Specification

STREAMS offers a large set of tools for specification of models.

The Model Building Language

The MB language is a meta-language for SEM, which has been designed to be as simple as possible, and yet allow specification of a wide range of models.

The following are some characteristics of MB:

	Four types of variables may be used as independent and dependent variables: manifest and latent variables, and residuals in manifest and latent variables.
	Four types of statements about relations, variances, covariances and means of the variables.
	In multiple group modeling, groups are refered to with labels in the MB statements.
	Constraints of equality over groups and variables are specified through enclosing lists of variable and group names in parentheses.
	A general equality constraining statement may be used to impose constraints of equality on any two free parameters.
	Fixed parameter values are assigned through supplying the parameter value in the statement.
	Special extensions to allow specification of two-level latent variable models.
	The MB instructions are translated into statements in the Amos 4, EQS, LISREL, or Mplus languages, and the model is estimated with one of these programs.
	If one or more previously estimated models are available which involve the same relations, start values may be copied from these.

Types of Models Supported

With STREAMS all the usual types of SEM models are easily specified and estimated with all SEM programs, e. g.:

	Regression and path models with observed variables.
	Confirmatory factor analysis.
	Structural equation models
	Growth curve models.
	Multiple group models.
	Models with or without means.

STREAMS also supports some special types of complex models for all or most SEM programs:

	Two-level latent variable models, in which SEM models are specified for group and individual level data simultaneously. Two-level SEM models are available with Amos 4, LISREL and Mplus, but not with EQS.
	Missing data models. In STREAMS a procedure is available to compute covariance matrices for all patterns of missing data. The program also constructs the setups which allow these matrices to be put together into one complete model.

STREAMS also supports some special types of models which only are available for one SEM program, e. g.:

	Latent variable mixture modeling (Mplus).
	Observed categorical variables, and mixtures of categorical and continous variables (Mplus and EQS).

An Intelligent Editor

Model specification is done with an intelligent editor, offering push buttons for choice of commands, list boxes for choice of variables, and forms for choice of options. The editor also keeps track of the logic of the model, and supports the editing process in several ways.

The MB statements are created in the so called Model Building Window, which is an ordinary text editor. Instructions may be entered and edited either through the keyboard, or through filling out forms, such as the Relations form.

The following functions, among others, are available to support model specification:

	A Model form to define basic characteristics of the model
	An Options form to specify options for the selected SEM program
	A Datasets form to specify one or of the datasets in the project to be included in the model.
	Forms for specifying manifest and latent variables
	Forms for creating REL, COV, VAR, MEA and EQ statements
	Automatic editing of instructions when manifest and latent variables are removed.
	A form to impose equality constraints over a range of statements
	Joining of two or more models into one.

Path Diagrams with Amos 4

There is also an experimental interface between STREAMS and Amos 4 which transforms an MB model to a path diagram, and vice versa: STREAMS constructs Amos path diagrams from MB models, which may be edited with the Amos diagramming tools. The model may then be estimated with Amos.

An example:

REL Verb -> WORD READ GI ERC
REL Reas -> DS DTM
COV Reas Verb

It should be observed that the path diagram interface is not yet fully developed.

Top

Estimation of Models

When the model is to be estimated the MB instructions are in a first step translated into instructions in the language of the selected model fitting program. The following programs are supported:

	Amos (3.6 and 4.0)
	EQS (5.7-6.1).
	LISREL (8.50 - 8.70)
	Mplus (Version 1-3).

STREAMS constructs a complete setup, with data specifications and model instructions. Normally the model is run within STREAMS, but the instructions may also be brought into the SEM program, for further work in that environment.

Start Values

Start values may be copied from one or more previously estimated models. This offers several advantages:

	Start values are easily supplied when they are needed..
	A program with less well developed algorithms for computing start values may take start values from a model estimated with another program.
	When successive changes are made to a large model estimation is much quicker if start values are taken from the model estimated in the previous step.
	Models which are difficult to estimate are typically easily dealt with if first a constrained version of the model is fitted and the constraints are then successively relaxed. In each estimation start values are taken from the previously estimated model.

Top

Presentation of Output

After the model has been estimated the results are brought back into STREAMS and translated into the MB representation. For many types of models additional calculations are made (e. g., computation of reliability coefficients in CFA models, and decomposition of variance in two-level models).

STREAMS 3 also features a Model Viewer whuch has been designed to be a flexible tool for organizing and inspecting modelling output. With the Model Viewer information may simultaneously be dealt with from:

One or more groups of cases.

One or more estimation methods (EQS only)

One or more models.

Different estimated statistics.

The information is presented in a spreadsheet table, with a tree view to control the organization of the presented information, e. g.:

The rows and/or columns of the spreadsheet may be split so that information from different groups, estimation methods, models or types of statistics is presented adjacently, e. g.:

The Model Viewer thus allows the mode of presentation to be tailored to suit the needs of the actual model. There also are functions for exporting the information to, e. g. EXCEL format, and for printing it.

Top