In our research finding using AMOS, the regression coefficient between the two constructs is negative (-0.238). However, the correlation coefficient between these two constructs is positive (0.017). What would be the proper explanation to justify these results?
A positively-correlated regressor should have a positive coefficient if it is the only regressor in a linear model. If not, that is a sign you did not account for confounding variable(s). The following could render relevant insights.
Statistics How To. (2020, July 7). Confounding variable: Simple definition and example. https://www.statisticshowto.com/experimental-design/confounding-variable/
Good luck,
@Deepak Same query has been discussed on RG in the thread below and might help your cause.
https://www.researchgate.net/post/Is-it-possible-to-have-a-negative-correlation-but-positive-regression-in-SPSS
Hello all,
the magical word is path tracing. The correlation between two variables is a result of all indirect and direct open paths connecting the two variables.
If your X variable has a negative effect but either a) correlates positively with one or more other variables who also have negative effects or b) correlates negatively with one or more other variables who have positive effect then, depending on the size of one or more of these components, the correlation will have the opposite direction.
This may be the result of *successful* control for confounding or other relationships, e.g., mediation.
For instance, if you have a partial mediation model with X affecting Y via M, and the indirect effect is positive and the direct effect is negative, the correlation between X and Y may be negative or zero or whatever. Correlations are the results of pieces of effects and relationships.
As said, google for path tracing. This is a fundamental principle underlying all causal structures and observed associations resulting from these.
I attached slides from a lecture on SEM. You will find path tracing on the first 10 slides (the rest is essential too).
HTH
Holger
When correlation among exogenous variables is high (greater than 0.6) it can lead to collinearity. Collinearity is an issue because it can increase standard errors and biases indicator weights (even sign changes)... Check correlation among exigence constructs.....
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS