[Rsiena-help] clarification on varying covariates waves / fixed parameter for cycle4 effect

[Ruth M. Ripley]

Dear Marco,

Varying covariates use M-1 values, as each is used for the whole of the
interval from time M to M+1. The help page is misleading. I will fix it!

The 4-cycles effect is either square rooted or not, but the statistic does
not appear to be so. This seems to be the reverse of what the
documentation suggests. I do not know which is correct: the names or the
code!

Regards,

Ruth


On Wed, 19 May 2010, marco tonellato wrote:

> Dear all,
>
> while importing attribute data into my model I've noticed a difference
> between the RSiena manual  and the help of the function "varCovar" regarding
> the number of covariate waves that must be imported:
>
> In the RSiena manual
> - paragraph 4.3 ("individual covariates"): "If observation moments for the
> network are t1, t2, ..., tM, then the changing covariates should refer to
> the M-1 moments t1 through tM -1, and the m-th value of the changing
> covariates is assumed to be valid for the period from moment tm to moment
> tm+1. The value at tM, the last moment, does not play a role".
>
> Whereas in the varCovar help
> - Arguments: val: Matrix of covariate values, one row for each actor, one
> column for each period.
>
> I noticed that RSiena takes M-1 columns, leaving out the last one.
>
> What it is not clear to me is whether individual changing covariates should
> be measured at each time period or, ideally, between each of the time
> periods (consequently  assuming one value for each time period minus 1).
> Could you please clarify on this? How should the varCovar columns be coded?
>
>
> Coming to a different topic, I noticed that the 4-cycles effect requires a
> parameter to be set, but in the 4cycle statistic there seems to be no such
> fixed parameter.
> Could you please clarify on this as well? What role does this fixed
> parameter play in the estimation?
>
> Thanks a lot,
>
> Marco
>