To: 人人
Your problem (sorry for this word) is that you do not fully understand the meaning of a test statistic. It is just a r.v. following some distribution (such that we may have its pdf or cdf). If it happens to be a distribtuion that is well-known, e.g. t, F, Chi-square, or Gaussian, then statistical inference can be done. either significance level ,p-value will be a simple by-product. This is partly why in some textbook, say for simple linear model, people will aruge that in order to do statistical inference we need a distributional assumption on the error terms of the model. or in an asypmtotical case, the modeler needs to impose some sort of moment conditions (iid, mixing) such that an appropriate CLT or functional CLT can be applied to derive the asymptotic distribution of the test statistic.
As long as the distribution of the test statistic is known, tail area of the distri can be obtained. If it is a t, fine, every textbook provides the table. If one has a distribution which is not well-known, the corresponding table can be generated by simulation (the empirical CDF). ADF test is a case of this.
I am not trying to earn the money from you. But from your comment in Floor3, I was stimulated to make some comments.