Summarizing time series data

kmj

New Member
#1
Hello,

Thank you very much for taking the time to read this!

I have many (~70) time series showing the average body size (kg) for the ~50 years. Each time series is for the same species but from a different region/area.

I would like to look at the change in body size over time and how this correlates with different regions and variables (e.g., temperature).

I am in over my head! I have scoured through Google and forums but have not found my answer! I would greatly appreciate any help or direction to other resources that you can provide.

Thank you!
 

noetsi

Fortran must die
#2
I don't know any simple approaches (time series is never simple). One possibliity, probably the easiest for what you want to do is regression with autoregressive error. If you just wanted to do a univariate analysis, predict future values, with no driving variable than exponential smoothing (the time series method not smoothing) is the fastest and some research suggests the most accurate. But this does not allow you to have a predictor. It just predicts the future of a series.

I am not sure, I have never seen, how you combine different series this way.
 

kmj

New Member
#3
Thank you very much for your response!

Have you heard of a dynamic factor analysis? Do you think this would be an appropriate way to answer my question?

Thank you!
 

noetsi

Fortran must die
#4
I have not heard that term although structural equation models can be used for some types of time series and that is a form of factor analysis (or more accurately factor analysis is a form of SEM). I am not enough of an expert in SEM to advice you if you can use that for this.

You might try spunky or lazar and see what they say. They know SEM very well.

I suspect repeated measure anova or regression with autoregressive error is where you will end up.
 
#5
Dynamic factor models seems to be interesting models. Stock and Watson are two persons related to it. There are R programs - if you are familiar with that.

However I did not understand the description of the data by kmj, so I don't know if it is relevant here.
 

noetsi

Fortran must die
#6
If you are asking about structural equation models then this is not simple for someone who has not worked with that before in my very humble opinion anway. It is not just the stats that are complex, you have to learn the software as well of which mplus is probably the best although there are many options. But none of the possibilities are simple.