Conclusions
- Lessons learned:
- rounding has an effect, especially for large data
sets
- if you are interested in the "real" numbers
- use small data sets
- use qualitative tools (e.g. Q-Q plot)
- use adapted methods
- read "Interval Estimation of a Normal Process Mean
from Rounded Data" by Lee and Vardeman
- General lessons:
- numerical experiments help to
- find facts
- test ideas
- identify relevant parameters
- "much data is better" is not always true
- "reasonable" assumptions or approximations might be
wrong
- always question everything - even if it has been used
for decades
- Is body height really normally distributed?
- of course not, it is always positive
- but: this is irrelevant in our example
- its has to be measured → values are always rounded
- but: you can always increase the precision
- it varies in time - you are larger in the morning
- but: you could define h(t)
- "body height" doesn't exist
- where is the upper limit of your body?
- ask the highest electron
- don't forget Heisenberg!