MyRegression

A simple function on LS linear regression with many informative outputs

Description

After the Dimitri's comment I deeply reviewed this file. Now this file use the Regress function (but the results with Polyfit were the same). Anyway I wrote this function exclusively for the common use of a one estimator linear regression; a common laboratory use where a simple regression is required.
Now this function checks the presence of outliers (as the Robustfit).
This function computes a least-squares linear regression suppling several output informations:
     - Presence of outliers
     - Slope with standard error an 95% C.I.
     - Intercept with standard error an 95% C.I.
     - Pearson's Correlation coefficient with 95% C.I. and its adjusted form (depending on the elements of X and Y arrays)
     - Spearman's Correlation coefficient
     - Regression Standard Error
     - Total Variability
     - Variability due to regression
     - Residual Variability
     - Student's t-Test on Slope (to check if slope=0), with power
     - Student's t-Test on Intercept (to check if intercept=0) with power
     - Power of the regression
     - Modified Levene's test for homoschedasticity of residuals
     - a plot with:
         o Data points
         o Least square line
         o Red dotted lines: 95% Confidence interval of regression
         o Green dotted lines: 95% Confidence interval of new y evaluation using this regression.
     - the residuals plot

The function requires the Statistics Toolbox because calls TINV and TCDF to perform the Student's t-test.
To compute power, Myregr requires powerStudent by Trujillo-Ortiz, A. and R. Hernandez-Walls. If this function is not present on the computer, it try to download it from FEX

My profile on XING http://www.xing.com/go/invita/13675097
My profile on LinkedIN http://it.linkedin.com/in/giuseppecardillo

• Required Products: Statistics Toolbox
• MATLAB release: MATLAB 7.3 (R2006b)

[Dimitri Shvorob]

REGRESS of Statistics Toolbox already does most of the calculations, and is not restricted to the single-regressor case. The author is apparently unaware of the 'b = inv(x'*x)*(x'*y)' formula, and computes OLS coefficients with POLYFIT.(?!). Redundant and badly written.

[Kai M]

This is a great file. It's way simpler than the theoretically overloaded MatLab code. Very nice gimmicks include outlier exclusion, plotted confidence intervals etc. This really saves time, because all you need is plugging in the two vectors you want to correlate. Thanks a lot to the author!!!

[C C]

It might already be in stat toolbox, but for those of use that can't afford the toolbox, this is great.

[Ben]

 

[Oleg Komarov]

In spite of the many useful fcns written by giuseppe, I don't see much use for this one, especially if we consider that the Statistics TB is mandatory (not to talk about fcns by other authors...). Regress, regstats and my personal enhanced variation do already the most...

[Kushsairy]

useful for me

[Andre Guy Tranquille]

 

[hamed nozari]

very good. thanks

[Suryo Kusumo]

 

[Chris Choi]

Having a few statistics that MATLAB default functions do not provide is awesome especially when it's too cumbersome to write a new one. Good job!