Hi Regression Gurus
Its satchi again, if Rsquare is significant but adjusted Rsquare is not significant.
What does this mean.
When I report, what do I say.
If I have R2 0.80, p<.001 but adjusted R2 is 0.76, NS. Does this mean my model is not good.
please help, badly need advice.
thanks

I tend to go by p<.05 (5%) as the threshold significance value. If the adjusted R^2 doesn't fit at this level I would consider refining the model using the coefficient scores.

Rj24 please can you tell me how to adjust it then?
where can I read about it, adjusting by the coefficient
thanks
satchi

Good grief, your question has given me a headache Satchi! No help, sorry, just fascinated! I have to do some Stats for my thesis and did a course on stats and understood- ermmm - none of it - so hoping I can learn through your questions!

I didn't think R2 and adjusted R2 had significance values? - or am I going crazy? are you talking about the change statistics in a hierarchical regression? So confused!

hi stressed and sneaks
yes there is p value for both; but if its the same model then it should be on the same line as R2 and adjusted R2
oh no what do I do now. Its been weeks since I looked at my results and now I look at them and I've gone blank
satchi

sneaks, where is the change stats? are you taking about R2 change in hierarchical regression, if so where is it.
I'm quite lost. thanks satchi

if you have done a hierarchical regression (i.e. used block 1, block 2 etc) then you can request 'change statistics' when you run it. So when you get the table with R2 and adjusted R2, to the right in the same table, there are 'r2 changed' and significance values - to denote whether the change was significant.

Are you using SPSS?

R2 gives the amount of variance accounted for by the model. Adjusted R2 gives the same statistics, but estimated for the POPULATION rather than the sample (which is what R2 gives). If your R2 is close to Adjusted R2 then it means that your model is pretty good at estimating what is going on in the population, not just the sample. I've never seen significant values for these (just the change stats, when necessary).

When looking at the significance for the overall model, its the F value that has significance.

Although I could be wrong - anyone please feel free to contradict!

hi Sneaks
thanks very much; yes I'm using SPSS; but haven't got much experience on it.
Sometimes I even forget what I have done!
I really dont know where to select for change statistics; i will have a look now.
thanks a lot
satchi

Sneaks
thanks I found it :-)
satchi

sneaks help......are you there.. i have in my 2nd block, r2 change of 0.04 but not significant.
But F is significant for this model but has a lower value compared to block 1.

Does this mean that when I have the added predictors in the 2nd block, they are not useful
thanks very much
satchi

======= Date Modified 16 Mar 2010 20:50:54 =======
Yep it means that although the overall model is still significant i.e. it predicts the outcome better than using the best guess (the mean), but the variables you added in the 2nd? block don't really add much. i.e. they do not add to the variance in a significant way.

An increase in variance of .04 is pretty small, and if the F value is smaller in the second model, it suggests the first model may be better.

I help MSc students with regression - are you making sure that you have at least 10-15 cases per predictor? - its the most common mistake they tend to make.

hi Sneaks; thanks again.
wow your statistics must be really solid. My sample size used is 32. I had 5 predictors; when you say 10-15 cases per sample what does it really mean. If I had 32, is it 32x5 then for the predictors? I also had some negative B values for some of the predictors.

how can I improve the B values, is there any way, besides adding the number of cases.
thanks a lot
satchi

for every predictor you add into the model, you need 10-15 participants/cases so for 5 predictors you need 50-75 cases, otherwise you will be losing power.

the negative b values just show you the relationship of the predictors.

The b value means for every unit change in the outcome, then the beta value changes by X units - so you won't change a negative to a positive, its like correlation, it signifies the direction of the relationship. Standardised beta values show the same thing but in standard deviations.