The time-series variable should be the variable that marks the time period for each observation, which sounds like the year column to me. You probably want to include a spline of time with, say, 3-5 knots in order to impute from flexible trend lines within each site. You probably also want to include the cross-section variable (site code) to account for mean-shifts from site to site.

matt.

On Fri, Apr 29, 2011 at 5:56 PM, Tariq Abbas <tariq_uvas@hotmail.com> wrote:

Hi,

There are 535 sites in Pakistan where wild birds are counted by volunteers once per year usually in the mid of winter. I have data from 1987-2007. The data contain missing values that I want to impute through Amelia II.

I have restructured subset of data into three columns (site code, year, count). Would you please guide me what should be the model specifications (options) [knots, priors, and numeric options if required]? I get error message when I set "count" as time series variable and try to impute it.

I can provide you data if required.

Regards

Tariq Abbas

Free university of Berlin

Germany