Hi Anis,

First of all I should know what do you mean with "I have done do a correlation and all my variable is significant". If you check the correlation between each IV and the DV, significance of this correlation is a good indication for the significance of the linear regression model. However, if you check all the correlation among the IV and find them significant, it is a serious problem called multicollinearity.

Before interpret these result, I think you must sure that all multiple-linear-regression assumption (i.e. no multicollinearity, homoscedasticity, etc) are met. I think if there are no problem with these assumption, the negative value should be interpret as an inverse-proportionally relation (that is, an increase in this IV will decrease the DV).

Hope this help.

When you say regression, do you mean doing separate regressions for each pair, so still looking at the relationship, unconditionally, between two variables, or do you mean some other kind of regression? If you mean looking at a regression where you use all IV to predict a DV, then the estimates associated with the individual IVs are the association after conditioning on the other variables, so will be different.

Are you interested in the relationship between each IV and the DV, or are you interested in something else?

His Anis,

In regression analysis, coefficients cannot be chaged because they indicate the relationship between IV and DV. Instead check if all assumptions of the regression are met (linearity, normal distribution of the residuala, multicolineality, homoscedacity,etc).