Dear All Professionals,
I am so wondering about how to approach to weight in data well? Thank you for your concern and lessons.
.........................
ID and Weight Adjustment for Flu Patients Using Korean Health Panel Data
Q1) How should the weight be applied when analyzing the factors influencing the choice of medical institutions among flu patients using the Korean Health Panel data?
Q2) Should weights be considered when analyzing causal relationships between variables, given that a national-level analysis typically requires weighting?
Q3) Depending on the weighting method chosen in Q2, how should weights be assigned to patients who have never used medical institutions or have never visited for flu?
I am so wondering about how to approach to weight in data well? Thank you for your concern and lessons.
.........................
ID and Weight Adjustment for Flu Patients Using Korean Health Panel Data
| A | 1.5 | 12 | 2 | 1.5 / 12 = 0.125 | 1.5 / 2 = 0.75 |
| B | 2.0 | 5 | 3 | 2.0 / 5 = 0.4 | 2.0 / 3 = 0.67 |
| C | 1.2 | 8 | 4 | 1.2 / 8 = 0.15 | 1.2 / 4 = 0.3 |
| D | 0.3 | 3 | 0 | 0.3 / 3 = 0.1 | NA |
| E | 0.7 | 0 | 0 | NA | NA |
Q1) How should the weight be applied when analyzing the factors influencing the choice of medical institutions among flu patients using the Korean Health Panel data?
- Maintain the original weight at the patient (ID) level.
- Divide the original weight by the number of total visits to medical institutions (hospitals and clinics).
- Divide the original weight by the number of visits to medical institutions specifically for flu.
- Keep only the flu patient sample and normalize the original weight.
- Post-stratification: This requires population data (age, gender, income, etc.), which may not be feasible.
Q2) Should weights be considered when analyzing causal relationships between variables, given that a national-level analysis typically requires weighting?
- When conducting a nationwide analysis, weighting is generally necessary.
- However, if the sample size is sufficiently large and the goal is to examine causal relationships between variables, is it still necessary to apply weights?
- If the analysis only focuses on causality between variables, can we assume that the observed sample data represents unbiased sampling?
Q3) Depending on the weighting method chosen in Q2, how should weights be assigned to patients who have never used medical institutions or have never visited for flu?
- If some patients have never used medical institutions at all or never visited for flu, how should weights be handled for such cases?
- Should they be assigned zero weight or be excluded from the analysis?
Comment