I am examining the relationship between a training dosage and subsequent efficacy, and comparing the relationship between a treatment (coded 1) and a control group (coded 0).
In a related earlier post -gung provided a nicely detailed explanation of how OLS uses a loss function to yield a line of best fit. Similarly, -Marco helped to distinguish how coefficients indicate change in the dep variable for a unit change in the ind variable.
I am looking for clarity on why my scatterplot results vary from my regression results. My scatterplot of training dosage and efficacy is shown below.
Specifically, OLS shows that the difference between treatment and control is -0.366, indicating the control group has higher efficacy than the treatment, controlling for training dosage. Yet the scatterplot, which compares Treatment and Control groups in the panels, shows that the treatment group has a higher positive correlation than the control.
In using a paneled scatterplot, I include the same three variables as used in the OLS. So, why are the correlations shown on the paneled scatterplot not (apparently) consistent with the results of OLS?