I am new to statistics and I am trying to understand conceptually (and hopefully visually!) what is the difference when someone states "under the null" or "under the alternative". I guess this may stem from my lack of understanding of whether or not the null and alternative hypothesis can be based on two different likelihoods.

  • $\begingroup$ It would help if you could provide some context. What is it that someone is claiming, eg, "under the null". The meaning could differ depending on the larger claim. More narrowly, by "under the ____", people typically mean under the assumption that the ______ is true. $\endgroup$ – gung Dec 8 '16 at 21:44
  • 1
    $\begingroup$ I just want to add something that hasn't been mentioned yet. Often the null hypothesis is the uninteresting hypothesis to the researcher. Consequently he or she wants to reject it. We deal with an observed statistic. It is common to say that as when discussing a p-value that under the null hypothesis the probability of observing a value as extreme or more extreme than what was observed (the definition of p-value). We have two types of errors, Type I (falsely rejecting the null hypothesis) and type II not rejecting the null hypothesis when the alternative is true. $\endgroup$ – Michael Chernick Dec 8 '16 at 22:33
  • $\begingroup$ Continuation: In this set up since we control the Type I error (significance level) when we cannot reject the null hypotheses the common statement is "cannot reject" rather that "accept" the null hypothesis. As discussed in other recent posts hypothesis tests can be one-sided or two-sided and so too for the p-value. $\endgroup$ – Michael Chernick Dec 8 '16 at 22:40
  • $\begingroup$ Another point I want to raise is that in medical research the null and alternative hypotheses are reversed because the uninteresting hypothesis becomes the important one. This occurs when testing that a new treatment is either "non-inferior" or "equivalent " to a standard treatment. In clinical trials this standard in some circumstances is sufficient to register a new drug or a generic drug to be approved as effective by the US Food and Drug Adminstration. See for example the work of William Blackwelder. $\endgroup$ – Michael Chernick Dec 8 '16 at 22:53

The exact specification of the null and alternative hypotheses depend on the test, but they all share general features.

Each hypothesis is associated with a particular probability distribution. For example, suppose you have a sample population with two variables: a type that can take values A or B, and a continuous variable. In this example, you might want to ask whether two groups of people (men and women) have different mean heights.

In this example, lets assume that height always follows a normal distribution. Our null hypothesis is that the heights for both men and women follow the same distribution. This is specified as

       meanwomen = meanmen

Whereas your alternative hypothesis specifies

       meanwomen $\ne$ meanmen

So you can ask for the odds "under the null" -- as in, if the null hypothesis were true, what would be the odds that you obtain your observed data? You can also ask for the odds "under the alternative" -- as in, if the alternative hypothesis were true, what would be the odds that you obtain your observed data? These would be your two likelihoods.

So the null and alternative hypotheses are not based on different likelihoods, but they each imply different likelihoods because they assume different distributions.

Since we here assumed the data followed a normal distribution, these distributions are normal. In the null hypothesis, our data all follows one normal distribution with mean equal to the observed total sample mean. In the alternative hypothesis, each group follows a normal distribution with mean equal to their respective observed group means.

As an added point, it's worth noting there are two distributions at play here: the distribution of the data, and the distribution of the sample mean. I was describing the distribution of the data above. The test above is ultimately based only on the null and alternative distributions of the sample means, which are of course functions of the distributions of the data and the sample size. In this case the hypothesis actually only specifies the mean. Frequently, such as with Student's T test, the hypothesis specifies that both the mean and standard deviation are the same.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.