I have conducted a population based survey and have weighted the data based on gender, age and location. I have conducted logistic regression analysis on the unweighted data, should I also have conducted the regression analysis on the unweighted data also?
If gender, age, and location are co-variables, you should just include them in your regression model. I don't understand why you want to "weight" your data based on these variables (and how you do this). Can you explain what you did?
Perhaps similar current discussion: https://www.researchgate.net/post/Weighted_vs_Unweighted_Regression_Results
Data can be weighted because of the probabilities of selection (survey weights) and/or because of modeling heteroscedasticity.
See https://www.researchgate.net/publication/319434665_Comparison_of_Model-Based_to_Design-Based_Ratio_Estimators for examples.
Calibration weights can combine those two weights.
Are you accounting for survey weights; is that what you are doing? I'm not familiar with this with logistic regression.
You might also model as Jochen suggested.
Weight in statistics is controversial and it depends upon the objective of the analysis. Logistic or linear regression generates coefficients inducting the importance of your variables. Could you please clarify the application of weight you are referring to what is the aim of applying weight?
- Badges
- Science topic