Close Home Forum Sign up / Log in

regression question on R-square

S

help Sneaks, somebody, anybody,
the R-squared value of my model is not significant.
But the F-value (of the ANOVA) is significant.
what does this mean

thanks very much
love satchi

The R Square just refers to the amount of variance your model accounts for - they only report significance for when you have a hierarchical regression - so you can say there was a 'significant change when xyz variables were added to the model' so it doesn't matter.

F-value means that your model is significantly better at predicting the outcome than the 'best guess' (the mean) so that's good!

So just report R2 and adjusted R2, don't report significance for them - only report the significance if you have 2 or more blocks and then report the 'change statistics' i.e. whether adding things in block 2 etc significantly changed the model.

Then report the significance for the ANOVA.

And then go on to do the beta values later (up)

S

thanks Sneaks
but I havent reported beta values before. From the SPSS output, its the B values reported, right? the coefficients.

HOw do I report beta values, is it after reported B values. If I remember correctly the beta values are for standardised values, something like that.

thanks a lot
love satchi

S

I have also a B value for a predictor which is 0.00, so this means the predictor does not account for any variance at all in the dependent variable. This also means it's a lousy predictor?

thanks again
love satchi

Have a read through this...

http://www.statisticshell.com/multireg.pdf

For reporting regression you need to firstly report the model statistics (R2, adjusted R2 and ANOVA) then you need to report how each predictor contributed for the model. Here you can report B and standardised beta (see the pdf - there is a table in it that shows how to report in a journal/thesis). You also need to report the T values for every one.

A B value of 0 is bad - it means that every time you change your predictor by 1 unit, your outcome changes by...erm 0. So your predictor has no effect - but do check the whole model and t-statistics.

If you were expecting this variable to have an effect, then maybe check multicollinearity - so that you know another variable isn't stealing all the effect away from it.

S

thanks so much Sneaks
will have a look at it now
:-) :-) :-)
love satchi

B

sneaks perfect reply........you are correct......sneeks have tried Structural Equation Modelling Using AMOS? i intend to use it for my analysis

No I haven't, although I was going to use it for my PhD. My sup has paid for me to go on a course later in the year, but its not to do with my PhD.

15762