In case of Heteroskedasticity, and Dickey Fuller (Unit Root test), when do we reject a null hypothesis. ie. If Stat value > Critical value, or the opposite of it. Thanks for help
Are you sure in case of Unit Root ? Because, I just solved a topic test, and it claimed the following statement
"Because the unit root test statistic (–18.7402) is smaller than the critical value (–2.89), the AR(1) model does not exhibit a unit root" In unit root hypothesis testing, our null hypothesis is that it unit root exists. So this means that if Test stat > Critical value, then technically Ho should have been rejected, meaning unit root does not exist. However in case of above, opposite is seen.
Hi. Thanks a lot for replying. Could you answer my query in the 4 cases as below Suppose, Case 1: tstat=-3, Critical value = -2
Case 2: tstat=-1 Critical Value = -2
Case 3: tstat= -3 Critical Value = 2
Case 4: tstat= 2 Critical Value = -2 In the above cases, would we reject or accept the null hypothesis. Also would Case 3 and Case 4 exist, I doubt they would not. Your answer really means a lot to me.Thanks
#3 Fail to reject if one tail test (upper); if two tail, then -2,+2 are the critical values and reject Ho because -3 < -2
#4 Fail to reject if a one tailed test (lower) but reject if two tailed test since test stat =critical value (implies p-value = alpha)
Overall, your decision to reject the null is based on the same procedure (test statistics in the rejection region, p-value no larger than alpha). However, you need to be keenly aware of the null and alternative hypothesis in each test to know what the conclusion you make.
In Dickey-Fuller testing, the null is that beta = 0 = (1-rho) so rejecting Ho suggests rho is different from 1 which means there is not a unit root. Failing to reject Ho means the data are not strongly disagreeing with rho being 1 which means the data are somewhat compatible with unit root (hence fail to reject Ho means you can’t disagree with a unit root based on the data and particular alpha level).
Technically, yes if alpha and p-value are equal you would reject H0 because alpha is the threshold where you say “not more than x% of true null hypotheses will be rejected.” The issue is somewhat trivial though as most cases alpha and p value or test statistics and critical value are not actually equal. In another light, the size of the p-value conveys information regarding how much our sample contradicts what we expect under the null hypothesis. When the p-value gets smaller, there is more evidence contradicting the null. Even if you technically reject Ho with p-value .05000000… and alpha .05, do you really think this tells you something different than a p-value of .053? Hopefully, you say no because “significant” and “nonsignificant” are just ways of dichotomizing a continuous summary of evidence from the sample (the p-value). There is nothing magical about a test being significant versus not significant, it’s just a way for you to say “at this threshold, I’m willing to conclude the alternative.”