VECM lag length after VAR/ADF testing

[hlearner]

1) I have three I(1) variables--I tested and found that they were individually I(1) via ADFs on each series.
2) When I run a VAR on LEVELS, I find that the optimal lag length is lag = X.
2a) When I run a VAR on FIRST DIFFERENCES, I find that the optimal lag length is lag = Y.
Should I be running my VAR on levels or first differences knowing that the series are I(1)?
3) I want to determine how many cointegrating relationships I may have (0, 1, or 2) via the Johansen Integration Test.
4) I know I need to input the number of lag lengths of the DIFFERENCED endogenous variables.
Does this correspond to Y, i.e., the number of lag based on a VAR run on first differences?
5) More generally, is it correct to determine the number of cointegrating equations based on the optimal lag length from a VAR run on LEVELS or DIFFERENCES?

Thanks.
PS I'm using Eviews 7 and I've read Book II, Chapter 32 (VAR and ECM)

[BobJ]

The lag length of the VAR/VECM has nothing to do with the number of cointegrating relationships. If think you're confusing optimal lag length with the number of cointegrating relationships. Variables can't be differenced and have a cointegrating relatiohsip. You would wipe out the cointegrating relation by first differencing the data.