Muliple Variable Regression
Components for Delphi Programming

Advanced Regression Modeling & Analysis

The MLR components support transformation of the data contained in database table fields. For any variable x in the regression model:

y = b₀ + b₁x₁ + ... + b_nx_n + e ,

possible transformations of x_j include

• field1*field2 (product transformation)
• ln(field1) (log transformation)
• 1/field1 (multiplicative inverse transformation)
• field1^k (kth power transformation)

where field1 and field2 are any two fields within a database result set.

Response variables can also be transformed to achieve a curvi-linear regression model. Modeling the data transformations is explained in the MLR Help file.

References for Regression Analysis

Before programming a regression application, review of regression theory would be very helpful.

Both of the following texts are highly recommended as references for multiple variable regression techniques. The text by J. Wesley Barnes is more introductory in nature.

• Statistical Analysis for Engineers and Scientists by J. Wesley Barnes, 1994
  
• Applied Regression Analysis by Norman R. Draper, Harry Smith, 1998

How the Components Work

The initial version included a TTable descendent component (TMLRTable) class and a TQuery descendent (TMLRQuery) component class. Each of these required the presence of the Borland Database Engine on the target system.

Version 2.0 replaces the TMLRTable and TMLRQuery with a single TDatasource descendent, TMLRDatasource, making the component set independent of the database engine. This new component can be used with any TDataset descendent which implements the typical dataset navigation methods, e.g.

• TADOTable
• TClientDataset
• TDBISAMTable/Query
• TIBTable
• TTable/Query
• TYourOwnTDatasetDescendent

The TMLRDatasource component loads the data contained in its assigned Dataset property, and performs the developer-defined modeling and the associated required mathematical calculations.

TMLRegress Features

	Instantaneously performs regression on any numeric data accessible through any TDataset compatible database engine.
	No need for your customers to to export data from your application's Paradox, dBase, or InterBase file formats to SPSS, SAS, Excel, Quattro Pro, or 1-2-3 to analyze.
	Scatter plots quickly show whether a linear data trend appears to be present among any of the control variables and the response variable.
	A variety of residual plots are available to access whether the general assumptions of a linear regression model are satisfied.
	Plots of residuals, y vs xi, for each i, and the cumulative distribution function can be viewed using the component plotting methods, or via another third-party graphing utility using the component properties.
	Can identify outlying data and predict the value of the dependent variable for different inputs of independent variable values.
	Delphi VCL Components allow developers to set all the parameters visually at design time if appropriate, or to develop complex models at run time.
	Sample project with multiple data tables and demo Object Pascal source code.
	TMLRegress Online-Tutorial demonstrates the building of a regression model with a TMLRTable component and five (5) lines of event-handler code. Please click here to take the regression tutorial.

Consulting Assistance

We recognize that, under any circumstance, programming multivariate regression applications is not the easiest development activity. Applied Analytic Systems is available to assist developers with integrating MLR components into their applications, for statistical modeling assistance, for custom statistical programming, and and for code review.

In most cases, we can provide substantial assistance with three or four billable hours. For help, contact our Statistical Consulting Group by email at statistics@aasdt.com or by phone at 412.253.4160.

TMLRegress Component Set

Applied Analytic Systems, Inc. Carnegie Office Park 600 North Bell Ave Bldg. 2, Suite 2700 Carnegie, PA 15106
Phone: 412.253.4160 FAX: 412.253.4161

All prices in US dollars. TMLRegress Copyright © 1997-2003 Applied Analytic Systems Inc.  Borland, Delphi, and C++ Builder are trademarks of Borland/Inprise Corporation.