I don't know if I understand well. Do you have independent variables data,which don't predict dependent variable?

I would check if every data individually has an impact on dependent variable, because sometimes independent varaibles can spoil their effect on dependent variable when they are all in the biggest group.

Yeah i have independent variables data. According to spss output my independent datas do not have significant effect on the dependent variable; but seperately some of them have. Should i specify this result also ?

My data has 2517 students.

Dear Tugba, model signification in multiple regression aims to lower the number of significant IVs by chance. For example, if you have 7 IVs vs one DV and want to test the significance of each at 0.05 alpha level, it is probable that you reach a significant relationship at that significance level by chance.

F statistics perform a familywise test on all coefficient to keep alpha level constant. So, if you want to test each coefficient, you can do so. But keep two things in mind:

1- there should exist a logical causal relation between each analysed IVs and your DV.

2- Use a familiwise correction test on your coefficient if it showed significant just below 0.05 level.

It is calculated as: 1-(1-alpha)^k

which k is the number of considered coefficients.

At last you can check the correlation between your IVs. if there exist any significant correlation, then it can be the cause of such results.

Cheers

EK

Dear Tugba

Check also the co-linearity between independent variables. Co-linearity may causes your logical independent variables insignificant..

regards

D.

If you suspect that some of the independent variables should influence your dependent variable I would check for simple correlation first.

Tugba, I may tend to disagree here. If you conduct a joint test of significance, such as the F test, for the overall effect of the model, then you may not want to test the predictors independently. Multi-collinearity will not cause the F-test to be nonsignificant. While you might have two (for example) correlated predictors knock each other out in terms of significance tests that won't affect the F-test. So if the model that is theoretically important to you contains the multiple predictors, there is no reason to break it down if you are looking at the joint test. If you have other theoretical models of importance, then sure, test the variable independently. Most output these days will yield tolerance or VIF statistics that will give you an idea if collinearity is playing a role or not. The reason I tend to disagree with any kind of single testing, one variable regressions, or correlation coefficient is to avoid testing numerous hypotheses that might not be of theoretical interest to you. I don't suggest doing all those tests just to find out. Bob

thank you Robert. i do not go into details since i have found that the model is not significant.

Tugba, I forgot to say that multicollinearity gives an adverse result in model significance. Which means, If it exist, total mode is significant, but your coefficients with high collinearity have not significant t test and their standard errors are high.

thank you for your consideration Fausto Galetto, but unfortunately, i gave my analysis to my instructor, i wish you could see my question 2 days before..

Even if they are not significant you have to report the magnitude of their low effect high effect or no effect otherwise they should not be placed in the study if they have no role

Don' t give up whatever!!

Using correlation, find out the individual strenghts of your sub-variables and try to do the regression without the weaker variables. I experienced a similar challenge and, this way, the variable in question became significant after I droped the weak sub-variable.