Hello Mohammad,

I'm not certain that I fully understand your query, but it appears that you are concerned that a given independent variable (IV) in a regression model has a negative regression coefficient estimate associated with it, but possibly a positive regression coefficient estimate with other, related models.

If you build a regression model with just one IV, then the sign of the regression coefficient (both standardized and unstandardized) will always be the same as that of the zero-order correlation (e.g., Pearson correlation) between scores on the IV and scores on the DV. In fact, the standardized regression coefficient will be identical in value to the Pearson r for that variable pair.

When you have more than one IV, however, things can change. The usual culprit for signs of regression coefficients no longer matching those of the bivariate Pearson correlations between a given IV and the DV is multicollinearity. When the explanatory power of IVs overlaps (e.g., there is redundancy), then it is quite possible that both the magnitude and sign for regression coefficients may seem not to correspond well to the Pearson correlations with the DV. As well, suppressor variables can sometimes result in regression coefficients that do not correspond well to the zero-order correlations.

Good luck with your work.

Mohammad -

If you have access to Google Translate, you might try that. Or those responding could try Google Translate. Either way, start by writing your question in simple words, with no idioms. That might help improve the translation, I think.

Best wishes - Jim