Negative Beta coefficient

#1
Hello people , I'm studying the effect of 3 independent variables on one dependent variable, the correlation result indicated that all of the independent variables show a positive relationship with the dependent variable but after I ran the multiple regression analysis , the beta coefficient of one independent variable was negative, can anyone explain why ? can such negative value be justified ?
 

Karabiner

TS Contributor
#2
If it changed from beta=0.03 to beta = -0.02 or so, personally I would not mind.

Otherwise you could have a look for suppressor effect.

With kind regards

Karabiner
 
#3
Hello Karabiner, thank you for your answer.
I tried to square the semi part in SPSS and compare with the R square to see if it's greater, but the value was smaller therefore this is not a suppressor variable , not sure what's wrong though
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
Provide your output, so we can see if it a trivial variability issue or more substantial. The big question is likely, what is the relationship 'causally' between these variables? Issues are likely related to: confounding, mediations, or collider bias (an IV is a common effect of other IV and the DV). You should be able to hypothesize the data generating process of your data and realize what the problem is, right! So please describe how the variables are related.
 
#7
Greta, not necessarily if you check "A Primer for Soft Modeling" article by (Falk & Miller, 1992) you will see that there can be other reasons than multicollinearity!
 
#9
Greta, not necessarily if you check "A Primer for Soft Modeling" article by (Falk & Miller, 1992) you will see that there can be other reasons than multicollinearity!
OK, but then you know that multicollinearity can be one cause of the problem. And I thought that that was what you asked for.

the beta coefficient of one independent variable was negative, can anyone explain why ?