I have some data of the diet of an omnivorous fish diet collected over 20 years. This data is divided in two seasons, the dry period (drought) and the wet period (flood), and I want to analyze the differences between these periods over the years. Unfortunately, some of these years have an unbalanced sampling design between the drought and flood years, and some periods within the years that we have no samples. If I use a random variable (like the sampling month or the year) in a generalized linear mixed model to control this sampling problem, can I compare the two periods?
Dear Ana Lucia,
if I understand you correctly, you're aiming to compare fish diet over 20 years between two seasons (wet and dry period, respectively). Generally speaking, I don't see any problems in comparing these two seasons/periods, if there are only minor data gaps (like, perhaps, just two or three seasons missing or single data points within the seasons). However, could you be more precise on what exactly you mean by "unbalanced sampling design between the drought and flood years"? If sampling techniques and/or data collection/recording were fundamentally different between different years or seasons, then this could potentially pose issues of comparability. But otherwise I don't see problems.
Hope this helps a bit.
Best,
Julius
Hi, As per the data type you mentioned, I believe it should be possible.
Thank you everyone, it was very helpful! Regarding the "unbalanced sampling design", I was referring to different number of stomach content analized per year, i.e., more stomachs analized in one period than in other, and, of course, one or two years with stomachs only in one period (drought or flood). For this reason, I believe that a random variable could help to control this missing data, since the data base is large.
Hi Ana Lúcia,
I don´t see any problem in comparing your data between dry and wet periods with only minor data gaps. If you have some year without any data on one of the periods you can take out that year of the analysis or add some random variable to take that into account.
Many thanks for the clarification, Ana Lucia!
Yes, I think you can go ahead with the analysis that way. I don't see any obstacles in there being different numbers of statistical subjects (in your case: stomach content) between different years and/or seasons. And if, as you say, there are only one or two years, for which entire seasonal data points are missing, this should be possible to "bridge" or extrapolate by statistical means.
By the way: Where in Brazil is your field site located?
Good luck and keep us posted about the progress of your research :-)
Julius
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Sampling Design