Thread Subject: Interpreting OLS Matlab results

Hi,
I'm new to Matlab and looking to interpret the below results. I'm not sure what the signiificance of things like the R squared, the coefficients, t stats are etc. I've used one dependant variable and six independent variables. Can anyone please help. Anything appreciated. thanks. Apologies if this is a double post.


Ordinary Least-squares Estimates
R-squared = 0.0644
Rbar-squared = 0.0457
sigma^2 = 0.0000
Durbin-Watson = 0.4464
Nobs, Nvars = 256, 6
***************************************************************
Variable Coefficient t-statistic t-probability
variable 1 0.146728 162.088033 0.000000
variable 2 0.000095 0.017054 0.986407
variable 3 -0.014683 -2.153490 0.032237
variable 4 0.011606 2.488331 0.013486
variable 5 0.000000 1.598341 0.111230
variable 6 -0.000000 -1.059647 0.290328

Wald F-test results

ans =

    1.0475 0.3832

kostas Lynch wrote:
> Hi,
> I'm new to Matlab and looking to interpret the below results. I'm not
> sure what the signiificance of things like the R squared, ...
> Ordinary Least-squares Estimates
> R-squared = 0.0644
...

This really isn't a ML question at all.

That value of R^2 implies the model explains only 6% of the total
variance of the data (which by inference leaves 94% unexplained).

Have you plotted these data and the model fit to look for
reasonableness?????

--

kostas Lynch wrote:
> Hi,
> I'm new to Matlab and looking to interpret the below results. I'm not sure what the signiificance of things like the R squared, the coefficients, t stats are etc. I've used one dependant variable and six independent variables. Can anyone please help. Anything appreciated. thanks. Apologies if this is a double post.

Kostas, this isn't really a MATLAB question; you really ought to pick up an introductory book on linear regression. But I couldn't help but notice this:

> Ordinary Least-squares Estimates
> R-squared = 0.0644

> sigma^2 = 0.0000

which seems rather odd, unless perhaps your data are pretty badly scaled. The first says you can't hardly predict y at all, the seond says the error in predicting y is very small (in absolute terms, anyway).