My IV is self-reported waiting time for an event occurrence (say waiting to enter a restaurant), rounded to the nearest minute. The range is 0 to 15 mins with a mean of 5 mins - positively skewed distribution.
Should I apply any transformation to the time variable? I'm wondering how the skewed distribution will affect my results.
The issue is usually is the response distribution ( conditioning on the IVs) not the IVs per se.
Kelvin's point is correct - you need to look at the residuals of the model with your predictors within it. That said it can be useful to transform time variables if you do have positively skewed residuals. The log transformation is common as it isn't uncommon for the effect of time to be multiplicative rather than additive. As you have zeroes you'd need to add a constant before transforming. If the zeroes are from scores rounded to the nearest minute (and you don't have the time in seconds) hen adding 0.5 might be reasonable (otherwise a smaller constant such as 1/60 is sensible). However, check the residuals of the model first.
data transformation performed with both IV and DV so normally distributed data
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics