Hi All, I am doing a linear regression analysis using stata software, which is new for me. In the interpretation of the regression output, are we using coefficient or Adjusted R squared to predict the contribution of independent variable to the dependent one? Usually we used to give the comment on coefficient but what I found in the provided link that they used R squared. https://statistics.laerd.com/stata-tutorials/linear-regression-using-stata.php.
Coefficients and R-squared show different characteristics of the model.
R-squared shows the variance in the dependent explained by the independent variables. It is a good practice to report R-squared.
Interpretation of coefficients is generally required.
Contribution of a variable is a complicated topic.
Thanks Mehmet for your kind replay. I have adjusted R -squared = 0.200, while the coefficient is very low (.0007) with significant p value (0.004). According to my understanding that the independent variable can explain the variability of the dependent variable by 20%, Is that right? Is this have practical value even with very low coefficient?. This is the confusion point.
Do you have just one predictor?
Do not forget that magnitude of coefficient is affected by the scale of independent variable. Namely, if you convert a variable measured in meters to a variable in centimeters, the coefficient is changed.
This model is for one predictor and it is adjusted for the other variables. The unit used for the independent variable (physical activity measurement) is minute/day (which is standard). I can convert it to min/week, even it will not change a lot, in the clinical application in the discussion, it will have the same impact as without conversion.
Yes, it seems it has a very low (or even no) impact.
Does your model have other variables (because you say adjusted) ? Although the answer to this question is not so important regarding the low coefficient of your variable, it might hint the interpretation of R-squared.
Thanks a lot for your time and your effort to help.
Yes, the model has multiple variables, as it was the full adjusted model.
yet the effect is found statistically different from zero with very high confidence. can it be due to the measurement scale of your dependent variable? that also influences the magnitude of the coefficient.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers