I have annual maximum water levels of a river station for 60 years from the water year 1959/60 to 2019/20, but there is a period of missing data for five years from 1965/66. Can I take remaining 55 years, as a continuous period, in my frequency analysis since the data from each year is considered independent?
I want have expertise view on the influence of missing data on the accuracy of results.
For me, as a person directly related to statistical analyses, I will say yes! you can do that without any significant problem in your result.
I too think you can do this providing you are willing to make the Missing at Random assumption, in effect the section that is missing is unrelated to the response being modelled ; eg it is not the wettest year that caused the destruction of the measuring instrument and therefore the lack of data. Article What is the difference between missing completely at random ...
Hi Prema Hettiarachchi !
Some questions:
>The "annual maximum water levels of a river station" is a single value per year, right?
>Are you 100% that the annual data are independent?
There are some things you could do:
-Fill the missing data with something representative (e.g. mean value of the other years, or the mean value of the two 5-year periods before and after them)
-Linear Interpolate the data, in an increasing/decreasing type: If year 1965 has a value of 10 (a random example) and year 1971 has a value of 20, the remaining steps could increase by (20-10)/5 = 2 units per year.
-You could also ignore the years before 1970, and have a continuous time series of water level values
Thank you sir. For your two questions, my answer is yes. However, since annual maximum water levels are generally independent, I have doubt about the validity of using linear interpolation for filling gaps as you suggested. Further, if I skip all the data before 1970, I will loose another 6 years of my data unnecessarily. I would like to have the views of other experts, including Mr. Medhat and Mr. Kelvin who answered my question before. Thanks again.
Dear Prema
From my limited statistical knowledge, I would agree with Medhat and Kelvyn and rather than trying to infill the missing years, it would be better to treat the 55 years of annual maximum water levels you have as a continuous period in your frequency analysis.
I hope this is of help.
Darren
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs