"output_years" argument in multi-state function for non-single year time intervals.

[shajiang93]

When my input data is structured in 5-year intervals, such as 2020, 2025, 2030, and 2035, specifying output_years = c(2020, 2025, 2030, 2035) causes an out-of-bound error. This happens because the multi-state function attempts to select input matrix lists at positions c(1, 5, 10, 15) instead of the intended c(1, 2, 3, 4). What I am doing currently is to give output_years as c(1,2,3,4) to make the function work and then recode the year- and age-related variables in the output later. But this seems not to be convenient. A potential solution to make this argument more flexible is to rename the list of input matrices based on the corresponding years, or any better solutions?

[IvanWilli]

Hi @shajiang93 , this is on the kin_multi_stage_time_variant_2sex ?
I think indexing is the right way. No matter the time and age jump (that must be equal), should work
What do yout hink @ButterickJoe ?

[ButterickJoe]

Hi both, I don't know if you recall Ivan, but the original code had a variable "n_inc" which effectively captured the age/time-increments. The result was to go through a list such as 2020, 2025, 2030 and return c(1,2,3); year=2020,2025,2030. Yourself and Diego didn't think it was necessary so I removed it.

If you want going forward, I could re-incorporate into the function and re-fork the package, or simply send you the raw files. The code as it stands will obviously work, but will require re-naming of years/ages as Sha noted.

[IvanWilli]

Hi Joe, thanks
Yes, you are right. That was suggested because the rest of the time-variable functions works indexing. I did not tested 5-years jump so did not see the issue with selecting.
I think we can keep it like that and rename. If this is used more frequently we can do the change. Is ok @shajiang93 ?
Best

[ButterickJoe]

Hi both (again). I have found the issue. On line 711 of the main function "kin_multi_stage_time_variant_2sex" I loop through the output years, and convert entries in this list of output years to the corresponding output data:

df_year_list <- list()
for(j in years){
ii <- as.numeric(j) - start_year + 1 ... [rest of code] .....

Assume, like Sha, we have times 2020, 2025, 2030. Then we get ii = 1, 6, 11 instead of the required ii = 1, 2, 3.

A a simple fix to ensure that when using 5/10-year time-series, we index the output data by:
ii <- 0
for(j in years){
ii <- ii + 1... [rest of code]...

Sorry about this. As stated, I wrote only for single-year models.

[IvanWilli]

Yes, that could be a solution. Aplied to both functions create_cumsum_df and create_full_dists_df . I can do the implementation in next days and come back to you

[ButterickJoe]

Whichever you prefer. I don't mind editing the code -- although not this week. I have looked at the file and there are 3 lines in each of the functions create_cumsum_df and create_full_df.

As mentioned, we loop through years. Also, a few lines down (e.g., 793), we assign the age of kin:

female_kin$age <- rep(seq(0, (na-1)), each = ns) [fine in 1-year age intervals].

This needs to be extended to include the age increment; e.g.,

female_kin$age <- rep( seq( 0 , n_inc*(na-1) , by = n_inc ) , each = ns) [n_inc-year age intervals]

Best

`

[IvanWilli]

Ok. Go for it when you have some time, no hurry for this week.
Thanks!

[IvanWilli]

Hi @shajiang93
Joe made some changes in #43 covering the issue, already merged.

[ButterickJoe]

Hi both,

I just wanted to check that having the function inputs output_years and model_years is what we want? Essentially, restricting the model to output_years introduces issues when plotting kin as a function of Focal's cohort of birth, e.g.,

if cohort = 1910 and output_years = c(1965,1975,1985,1995,2005) as in the Vignette, then Focal can only be one of 5 ages: 55,65,75,85, or 95. Is this discretisation what we require?

Best wishes all
Joe