AES Consulting meeting on 7 Sep 2016
The data set discussed today concerned time series under 3 different treatments with 3 replicates under each treatment under an apparent completely randomized design. The time series has ~1000 observations with, apparently, many missing values and, in particular, many missing values for 3 of the time series. We had many questions about the design of this experiment including
- the actual design and
- what is going on with the apparent missing data,
but the main question to be addressed here is what to do with a time series when the client has little statistical background.
Simplify the times series
The main suggested approach is to create statistics from each experimental unit that are scientifically meaningful and perform some relatively simple analyses on these statistics. For this data set some possible statistics are
- average across the time series,
- mean absolute error relative to averaged out measurements, and
- average daily range.
The client should be consulted about other relevant statistics that are scientifically meaningful.