When working with healthcare cost data, it’s common to encounter skewed distributions with outliers. In such cases, estimating the average treatment effect might not be enough. Instead, we should consider estimating median treatment effects (MTE) or median treatment effects on the treated (MTT) to get a more comprehensive understanding of the data.
One simple approach to estimate MTE or MTT is to use quantile regression instead of ordinary least squares (OLS) regression. This is easily achievable using R packages like MatchIt and quantreg.
However, when using propensity score matching followed by regression on the matched data, things get a bit more complicated. Calculating valid confidence intervals for MTE or MTT requires a more nuanced approach. Bootstrapping seems like a promising method, especially when combined with propensity score matching (PSM) or other methods like g-computation.
Why Median Treatment Effects Matter
Estimating median treatment effects can provide valuable insights into how a treatment affects the median individual, rather than just the average. This is particularly important in healthcare, where individual responses to treatments can vary greatly.
The Challenge of Skewed Data
Skewed data with outliers can lead to biased estimates of treatment effects. By estimating median treatment effects, we can mitigate this issue and gain a more accurate understanding of the data.
The Role of Quantile Regression
Quantile regression is a useful tool for estimating median treatment effects. It’s a simple and fast method, especially when using R packages like MatchIt and quantreg.
Confidence Intervals for MTE or MTT
When using propensity score matching followed by regression on the matched data, calculating valid confidence intervals for MTE or MTT is crucial. Bootstrapping is a promising approach, but it’s essential to consider other methods like g-computation to ensure robust results.
Conclusion
Estimating median treatment effects can provide a more comprehensive understanding of healthcare data. By using quantile regression and robust methods for calculating confidence intervals, we can gain valuable insights into how treatments affect individuals. Don’t settle for just estimating average treatment effects – go beyond average and explore the median treatment effects in your healthcare data.
Further reading: Quantile Regression