63785_en-US

liliconcom

Crossing and missing data

Hello !

I need your help!
I performed repeated mass measurements on the same individuals for 5 years.
I tried to repeat this data every month but some months I have no data. So I have missing data.

However, I want to see how the mass evolves in terms of months, year and month * year.
So I have dependent variable "mass", then independent variables: "month", "year" and "month * year"; "individual" is in random variable.

However, since I have missing values for some months, when I put the "month * year" cross in my model, the model does not work (there are no values).
 
I would like to know if you have a solution to this problem.

Thanks in advance !!!

Re: Crossing and missing data

With only 5 years of data, I'd suggest starting with a graphical view of the data - one time series for all 60 months.  If there seems to be a seasonal pattern, it will be important to include month in your model (or perhaps some combination of months).  Trend should be easy to spot if it is present.  More complicated is a seasonal pattern that changes over time - meaning that the seasonal variation is different in each year.  That sounds like what you are attempting to model - but with 5 years of monthly data, I think it will be difficult for any statistical test to find a changing seasonal pattern - although graphically you should be able to describe it.  Finally, whatever model you end up running, I'd suggest using the Fit Model platform so that you can select the "Informative Missing" option to deal with the missing data.  Alternatively, you might want to impute the missing data.

Re: Crossing and missing data

How do you code the Month values? 1-12? If so, then nest Month within Year. I thought that Month ran from 1 to 60.

Re: Crossing and missing data

Thank you for your reply.

Only, if I put only the "month" in my model, I will get a mass average for each month, all years combined. Which is good for having a general pattern of mass variation over time.

If I put only the year, I will get a mass average for each year but in my case I want to know if there is a difference in the annual mass pattern so I think I need to include the interaction ' month * year '.

Re: Crossing and missing data

I think that you do not have independent predictors. Month and Year are just different intervals of the same predictor, time. That is like entering a model C + T + F(T) + T*f(T).

 

You should use Month or Year but not both at the same time in the linear predictor.