I am trying to run a multiple linear regression with two independent factors. These are biological markers that negatively influence the level dependent factor so when running the model, I expect to get negative coefficients. This is the case in three of my four data sets. For the fourth one I am getting one positive coefficient, which is in turn throwing off the whole calculation and positively affecting the level of the dependent variable.
I tried running the independent variables separately and they each give negative results, but when done together, one returns positive.
My argument is from a biological standpoint that theoretically this does not happen. I am however open to suggestions but suspect a statistical error. I checked for normality and everything seemed okay
Check the individual data points for any outliers that might be throwing off the correlation.
I was looking in the wrong variable but checking the dependent variable, there does seem to be some outliers indeed. I will try to sort these out
This could be due to multi collinearity in the affected data set.ie the two independent variables are highly correlated .This can sometimes result in a change of sign of regression coefficient.
It makes sense that these two would be highly correlated because the both measure the same thing (inflammation status). To use these in an equation though would it be acceptable to change the sign, or just run a simple regression (with the independent variables separately) and use those coefficients?
Update: I found few outliers int he dependent variable and removed them but am still getting a positive coefficient. I also tried log transforming the values in all the variables but still get a positive result. Does anyone else have another suggestion?
since the independent variables are highly correlated , i suggest you choose the one which is easiest to measure/ or is most reliablyand accurately measured, and omit the other from the model..
Unfortunately I need coefficients from both. One measures early stage inflammation and the other late stage so both coefficients are part of the equation.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running