Hi Shagun,
When estimating 95% confidence intervals (CI) for non-normally distributed data in MATLAB, bootstrapping combined with the 'prctile' function is a robust method that doesn't assume a specific data distribution. This approach involves resampling your data with replacement, calculating the statistic of interest (e.g., mean) for each sample, and then using 'prctile' to find the 2.5th and 97.5th percentiles as the CI bounds.
Comparatively, transforming data (e.g., log transformation) to approximate a normal distribution and then calculating CI can yield different results. This method requires the data transformation to be effective in normalizing the distribution and may affect interpretability and accuracy if the transformation introduces bias.
In essence, bootstrapping with 'prctile' is generally preferred for its versatility and reliability with non-normally distributed data, while data transformation followed by parametric CI estimation could be considered when the transformation is appropriate and well-understood. The choice between these methods depends on the dataset and analysis requirements.
