If you're seeing this message, it means we're having trouble loading external resources on our website.

If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.

Main content

Examples of null and alternative hypotheses

AP.STATS:
VAR‑6 (EU)
,
VAR‑6.D (LO)
,
VAR‑6.D.1 (EK)
,
VAR‑6.D.2 (EK)
,
VAR‑6.D.3 (EK)
,
VAR‑6.D.4 (EK)
,
VAR‑6.D.5 (EK)
,
VAR‑7 (EU)
,
VAR‑7.C (LO)
,
VAR‑7.C.1 (EK)
The null and alternative hypotheses are both statements about the population that you are studying. The null hypothesis is often stated as the assumption that there is no change, no difference between two groups, or no relationship between two variables. The alternative hypothesis, on the other hand, is the statement that there is a change, difference, or relationship.

Want to join the conversation?

  • blobby green style avatar for user maroom253
    Sal said 'lets remind ourselves what null hypothesis is' did he cover this topic somewhere else?
    (20 votes)
    Default Khan Academy avatar avatar for user
  • blobby green style avatar for user Warren Sunada-Wong
    Does the null hypothesis need to have the equality symbol only?

    In the second question about average hours of sleep, Sal writes the null hypothesis H0: u >= 8 hours. However, in every other null hypothesis I've seen on Khan Academy, the null hypothesis has a "=" not a ">=" or "<=". Is it required to write only "=" or are the other two equality symbols okay as well?
    (7 votes)
    Default Khan Academy avatar avatar for user
    • leafers ultimate style avatar for user Cory Cascalheira
      The short answer is yes: the other symbols are fine. It really depends on your research question or the test you conduct.

      Your null hypothesis is simply what you assume to happen at baseline when everything is going as it should be. Often, this means no difference, which is the same as equality, =.

      But you could be comparing reviews of popular American movies in, say, Russia. Consider Avatar (2009), the highest grossing American movie of all time. It is reasonable to assume, at baseline, that most Americans would rate Avatar favorably. But do Russian sensibilities differ significantly?

      How we approach this question (one-sided or two-sided) depends on how we articulate our hypotheses. If we have evidence that Russian ratings of American movies tend to be lower than American ratings of American movies--maybe we read studies about this trend, or observed it while surfing the Internet--then our hypotheses would be:

      Null: Russian ratings < American ratings
      Alternative: Russian ratings >= American ratings

      But maybe we have no clue how Russians rate American movies. If the ratings were the same, then that would not be a very interesting finding, right? Equal ratings mean that the film was good. But if we found a difference in ratings, that could be interesting because now we have a new question: why are the ratings different? But before we get ahead of ourselves, it is possible to frame the hypotheses as statements of equality:

      Null: Russian ratings = American ratings
      Alternative: Russian ratings =/= American ratings

      So, yes: you can use logical comparisons beyond simple equality. It depends how you frame your research question.
      (17 votes)
  • blobby green style avatar for user word3381
    Helloo.
    Can someone help me to explain how null hypotheses works. I don't understand? I need help urgent.
    (0 votes)
    Default Khan Academy avatar avatar for user
    • leafers ultimate style avatar for user Cory Cascalheira
      The null hypothesis is what happens at baseline. It is the uninteresting hypothesis--the boring hypothesis. Usually, it is the hypothesis that assumes no difference. It is the opposite of your research hypothesis.

      The alternative hypothesis--that is, the research hypothesis--is the idea, phenomenon, observation that you want to prove.

      If you suspect that girls take longer to get ready for school than boys, then:

      Alternative: girls time > boys time
      Null: girls time <= boys time

      If you think that your sibling gets more expensive presents than you on the holidays, then:

      Alternative: sibling presents cost more than my presents
      Null: the cost is not different

      I think of null as not, nothing, void, absent, uninteresting, normal, not worth reporting. If the null hypothesis were true, and you explained the null hypothesis to a friend, they might say, "Well, duh! Who cares?"

      Imagine going up to your friend and saying, "I had to reject the claim that my sister gets more expensive presents than me around the holidays! It turns out that my parents treat us the same!"

      Your friend: "Well, duh! Who cares! Parents are supposed to love their kids equally."

      The null hypothesis is the who-cares hypothesis.
      (22 votes)
  • winston baby style avatar for user Yash Singh
    At the second question didn't the statistics class assume that the average sleep time at their high school is less than 8 hours?
    (4 votes)
    Default Khan Academy avatar avatar for user
  • blobby green style avatar for user max liu
    shouldn't the null and alternative be mutually exclusive and collectively exhaustive? If H0 is =, then H1 can only be ≠. To test if H1 is "<" then H0 has to be ≥.
    (3 votes)
    Default Khan Academy avatar avatar for user
  • starky ultimate style avatar for user horus.scope
    It's such an abuse of language to suggest that your so-called hypothesis is the exact thing you think is false. If you suspect they're sleeping less than 8 then obviously your hypothesis is not that they are sleeping 8 or more. Don't worry, I understand just fine; I just don't respect it and won't be able to respect anyone who uses this statistical language for the rest of my life.
    (2 votes)
    Default Khan Academy avatar avatar for user
  • blobby green style avatar for user ankurkabra17
    What happens if we interchange null and alternate hypothesis? How does the test change
    (2 votes)
    Default Khan Academy avatar avatar for user
  • leafers ultimate style avatar for user charles john
    What is the meaning of simulation given in the previous exercise of simple hypothesis testing. I am not able to capture the idea of frequency given in the simulation table.
    (1 vote)
    Default Khan Academy avatar avatar for user
  • blobby green style avatar for user Lena Turner
    How would I find the level of significance for weight loss using 12 individuals 6 loss more weight
    (1 vote)
    Default Khan Academy avatar avatar for user
  • old spice man blue style avatar for user michael.szabo
    Anybody else think it's weird our null hypothesis for the second example is based on a recommendation stating teenagers should sleep >=8hrs. It doesn't say they do.

    I thought the point of null hypothesis and alternative hypothesis was to challenge established ideas. So if we had an established population average of teenagers sleeping >=8hrs of sleep, that'd be the null hypothesis that someone could go and challenge with their alternative hypothesis. The idea of setting up a null hypothesis from a recommendation and not an actual statistic throws me off.

    Any thoughts?
    (1 vote)
    Default Khan Academy avatar avatar for user

Video transcript

- [Instructor] We are told a restaurant owner installed a new automated drink machine. The machine is designed to dispense 530 milliliters of liquid on the medium size setting. The owner suspects that the machine may be dispensing too much in medium drinks. They decide to take a sample of 30 medium drinks to see if the average amount is significantly greater than 500 milliliters. What are appropriate hypotheses for their significance test? And they actually give us four choices here. I'll scroll down a little bit so that you can see all of the choices. So like always, pause this video and see if you can have a go at it. Okay now let's do this together. So let's just remind ourselves what a null hypothesis is and what an alternative hypothesis is. One way to view a null hypothesis, this is the hypothesis where things are happening as expected. Sometimes people will describe this as the no difference hypothesis. It'll often have a statement of equality where the population parameter is equal to a value where the value is what people were kind of assuming all along. The alternative hypothesis, this is a claim where if you have evidence to back up that claim, that would be new news. You are saying hey there's something interesting going on here. There is a difference. And so in this context, the no difference, we would say the null hypothesis would be, we would care about the population parameter, and here we care about the average amount of drink dispensed in the medium setting. So the population parameter there would be the mean, and that the mean would be equal to 530 milliliters. Because that's what the drink machine is supposed to do. And then the alternative hypothesis, this is what the owner fears, is that the mean actually might be larger than that, larger than 530 milliliters. And so let's see which of these choices is this? Well these first two choices are talking about proportion, but it's really the average amount that we're talking about. We see it up here. They decide to take a sample of 30 medium drinks to see if the average amount, they're not talking about proportions here, they're talking about averages, and in this case we're talking about estimating the population parameter, the population mean, for how much drink is dispensed on that setting. And so this one is looking like this right over here. Only these two are even dealing with the mean. And the difference between this one and this one is this says the mean is greater than 530 milliliters, and that indeed is the owner's fear. And this over here, this alternative hypothesis, is that the, that it's dispensing on average less than 530 milliliters, but that's not what the owner is afraid of. And so that's not the kind of the news that we're trying to find some evidence for. So I would definitely pick choice C. Let's do another example. The National Sleep Foundation recommends that teenagers aged 14 to 17 years old get at least eight hours of sleep per night for proper health and wellness. A statistics class at a large high school suspects that students at their school are getting less than eight hours of sleep on average. To test their theory, they randomly sample 42 of these students and ask them how many hours of sleep they get per night. The mean from this sample, the mean from the sample, is 7.5 hours. Here's their alternative hypothesis. The average amount of sleep students at their school get per night is... What is an appropriate ending to their alternative hypothesis? So pause this video and see if you can think about that. So let's just first think about a good null hypothesis. So the null hypothesis is, hey there's actually no news here, that everything is what people were always assuming. And so the null hypothesis here is that no, the students are getting at least eight hours of sleep per night. And so that would be, that remember we care about the population of students. And so and we care about the population of students at the school. And so we would say well the null hypothesis is that the parameter for the students at that school, the mean amount of sleep that they're getting, is indeed greater than or equal to eight hours. And a good clue for the alternative hypothesis is when you see something like this where they say, a statistics class at a large high school suspects, so they suspect that things might be different than what people have always been assuming or actually what's good for students. And so they suspect that students at their school are getting less than eight hours of sleep on average. And so they suspect that the population parameter, the population mean, for their school is actually less than eight hours. And so if you wanted to write this out in words, the average amount of sleep students at their school get per night is less than eight hours. Now one thing to watch out for is one, you wanna make sure you're getting the right parameter. Sometimes it's often a population mean. Sometimes it's a population proportion. But the other thing that sometimes folks get stuck up on, but the other thing that sometimes confuses folks is, well we are measuring, is that we are calculating a statistic from a sample. Here we're calculating the sample mean, but that, the sample statistics are not what should be involved in your hypotheses. Your hypotheses are claims about your population that you care about, here the population is the students at the high school.