To all users of Amelia,

Using basis functions in case of time series cross section data seems to improve amelia imputations substancially (at least in my datacase).

**In Amelia manual on define polynomials used as b0 + b1* t + b1* t^2 + b1*t^3. I understand it is a typographic error and coefficients are different (b0, b1, b2, b3..)

Spline function looks like improves slightly imputations (so far splinetime=4) and it´s generally preferred for interpolation (wikipedia). I`m just testing.

But LOESS or local regression, where on make a different regression for each element based on the k-neighbors, using least squares regression over a polynomial is really appealing. The tecnique is not widely used (I guess) because until a few years ago computers didn´t allow for it.

James Honaker/Gary King achieve better imputations in "What to Do About Missing Values in TSCS Data" with LOESS.

Meanwhile, and as far as I keep testing I have also changed from limit the draws (, emburn=c(20,100)) to increase tolerance (,tolerance=0,001) and results should now be more solids.

** Also trying to subset overimpute  (a.out, subset=country=="france", "Vtas") gives me this error:

Error in `[<-`(`*tmp*`, , j - 1, value = c(NA, NA, NA, NA, NA, NA, NA,  : 
  subscript out of bounds

I think it could be for having variables (defined as "idvars" at imputation but not at overimpute instruction) with "NA". 

Someone has experience in changing basis functions or could give some help?

Jesus Borruel

PHD student

San Pablo-CEU University

Spain(EU)

--
Amelia mailing list served by HUIT
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
Amelia mailing list
Amelia@lists.gking.harvard.edu

To unsubscribe from this list or get other information:

https://lists.gking.harvard.edu/mailman/listinfo/amelia