NHST2

Summary

The video explores the process of hypothesis testing, focusing on the comparison between alternate hypotheses (H1) and null hypotheses (H0). It uses relatable examples, like the impact of red wine on health, to explain the concepts. The video emphasizes the importance of complementarity in hypothesis sets, discusses directional vs. non-directional tests, and highlights real-world examples like cereal consumption and body mass index. It also delves into potential pitfalls when reading statistical results, such as the difference between correlation and causation, the role of confounding variables, and the significance of experimental design and blinding in research.

Highlights

• Hypothesis testing helps determine the truth between competing claims. 📚
• Alternate hypotheses suggest something is happening, contrasting with null hypotheses. 🤔
• The red wine example highlights how hypotheses contrast, showcasing different outcomes. 🍇
• Understanding probability through marble selection aids in grasping hypothesis concepts. 🎲
• Directional tests offer predictions, whereas non-directional tests do not specify an effect's direction. ➡️
• Real-life examples like cereal and BMI help illustrate how hypotheses function in the real world. 📊
• The distinction between correlation and causation is critical; many fall into this trap. 🚫
• Explanatory versus response variables: knowing their roles in research is vital. 🔄
• Placebo and blinding in studies help safeguard against bias and misinterpretation. 👀
• Being mindful of complementary sets when testing hypotheses ensures comprehensive analysis. 🧠

Key Takeaways

• Understanding the difference between alternate (H1) and null hypotheses (H0) is crucial in hypothesis testing. 🔄
• Red wine hypothesis serves as a classic example to explore these concepts. 🍷
• Directional vs. non-directional tests offer different insights; directionality is key. 🚦
• Correlation isn’t causation! Beware of making causal claims from observational data. ⚠️
• Confounding variables can mislead; identifying them is essential for accurate interpretation. 🔍
• Experimental design, including blinding, ensures the validity and reliability of research findings. 🧪

Overview

The video delves into hypothesis testing, differentiating between alternate (H1) and null (H0) hypotheses. Through examples like red wine's effect on heart health, it illustrates the necessity of understanding these concepts, as well as the importance of exploring both directional and non-directional tests.

Erin Heerey discusses real-world instances such as cereal consumption's impact on body mass index, highlighting common statistical misinterpretations where correlation is mistaken for causation. The lecture accentuates the critical role of identifying confounding variables in hypothesis testing to reach accurate conclusions.

In conclusion, the video stresses the importance of robust experimental design and methodology, including blinding techniques in studies, to ensure findings' credibility. It serves as a comprehensive guide for students and researchers to navigate the complexities of statistical analysis and hypothesis formulation.

Chapters

• 00:00 - 00:30: Introduction to Hypothesis Testing In this chapter, the concept of hypothesis testing is introduced. It explains the process of testing between competing claims through hypotheses. The focus is on the 'hypothesis of interest', also known as the 'alternate hypothesis' or 'research hypothesis', typically abbreviated as H1, which suggests that there is some effect or phenomenon occurring in the study.
• 00:30 - 01:00: Hypothesis and Competing Claims The chapter explores the hypothesis that red wine consumption improves cardiovascular health. Researchers conclude that a glass of red wine daily benefits cardiovascular health. However, the chapter emphasizes the need to contrast this hypothesis with complementary predictions, indicating the importance of investigating competing claims and maintaining a balanced perspective in scientific research.
• 01:00 - 01:30: Null Hypothesis The chapter titled 'Null Hypothesis' discusses the concept of the null hypothesis, commonly denoted as H0. It is defined as a statement that there is no effect or no difference, essentially asserting that 'there's nothing going on.' An example is given in which the consuming of a glass of red wine every day is said to have either no effect or a negative impact on cardiovascular health. To evaluate the null hypothesis, data is collected and analyzed to decide whether to reject or retain it based on the evidence provided by the data.
• 01:30 - 02:00: Understanding the Probability Concept In this chapter, the concept of probability is introduced using the example of selecting marbles from a bag. The scenario presented involves a bag containing 14 marbles, with only one orange marble. The probability of picking the orange marble is calculated as 1/14. Additionally, the chapter covers the probability of selecting a marble of any color except orange, emphasizing the fundamental principles of probability in determining these outcomes.
• 02:00 - 02:30: Defining Hypotheses and Statistical Observations The chapter focuses on defining hypotheses within the context of statistical observations. It explains the concept by using the example of picking an orange among other objects. The probability of picking an orange is 1 in 14, while the probability of picking anything else is 13 out of 14. This analogy is used to explain the null hypothesis as being the 'not Orange' or the default assumption, in contrast to the alternative hypothesis (H1), which aligns with the research hypothesis.
• 02:30 - 03:00: Directional vs Non-directional Tests The chapter introduces the concept of hypotheses, particularly focusing on directional versus non-directional tests. It begins by explaining a basic probability scenario involving the selection of oranges. The null hypothesis is described as covering the sample space that excludes the research hypothesis. The importance of defining hypotheses in a way that encompasses the full sample space is emphasized. Additionally, it is noted that hypotheses are often derived from observations, such as those made by organizations like Statistics Canada.
• 03:00 - 03:30: Understanding Hypotheses in Real-world Context The chapter titled 'Understanding Hypotheses in Real-world Context' discusses the process of forming a hypothesis using real-world data. It begins with an observation from 2021, where female employees aged 25 to 54 earned 89 cents for every dollar earned by male employees. This observation is used to form a hypothesis that in Canada, being female causes an employee to earn less than being male, prompting a discussion on the components and implications of such a hypothesis.
• 03:30 - 04:00: Explanatory and Response Variables The chapter discusses the concept of explanatory and response variables using an example of gender wage disparity. It presents a hypothesis suggesting that female employees earn less than male employees, emphasizing the scope of this hypothesis as being specific to Canada, based on observations reported by Statistics Canada. The scope or generality of this hypothesis limits its applicability to the Canadian population.
• 04:00 - 04:30: Understanding Confounding Variables The chapter focuses on understanding confounding variables, using an example related to gender and salary discrepancies. The text suggests a causal relationship, stating that being female causes an employee to earn less than being male. This hypothesis is presented as a provocative starting point for discussion. Additionally, the notion of identifying a causal mechanism is introduced. However, it is noted that this is not a hypothesis that would necessarily be tested, highlighting the complexities and ethical considerations in social research.
• 04:30 - 05:00: Placebos and Blinding in Experiments The chapter discusses the concept of causal mechanisms in scientific hypothesis formulation, emphasizing the importance of identifying the chain of events that explains observations in experiments. It highlights the role of hypotheses in explaining the 'why' or 'how' behind phenomena, which is crucial for understanding the causal relationships and mechanisms underlying the observations.
• 05:00 - 05:30: Conclusion on the Importance of Double-Blind Designs This chapter underscores the critical role of double-blind designs in experiments. It highlights the need to differentiate between speculations and actual testable mechanisms within experimental contexts. The chapter emphasizes that readers should be cautious about interpretations of statistical results, ensuring that the methods applied are appropriate and allow for accurate causal inference.

NHST2 Transcription

• 00:00 - 00:30 so when we're testing hypotheses one of the things that we're doing is we're testing whether competing claims which one of those competing claims is true so we have a hypothesis of Interest which is also known as an alternate hypothesis a research hypothesis and often it's abbreviated as H with a subscript of 1 or H1 this is the idea that there's something going on out there so this is a paper
• 00:30 - 01:00 that suggests that red wine consumption improves cardiovascular health and what these the researchers in this paper conclude is that a glass of red wine every day is good for cardio cardiovascular health now I'm sure that some of the people listening to this lecture maybe myself included would like for that to be true but really so we have to contrast our hypothesis of Interest with a complementary prediction
• 01:00 - 01:30 and that complementary prediction is called the null hypothesis also known as h0 and it means there's nothing going on here so a glass of red wine every day has either no effect or a negative effect on cardiovascular health would be the competing claim here what we do then is we collect data and we make a decision that either allows us to reject or retain this null hypothesis based on the observed data
• 01:30 - 02:00 so in order to understand this process we need to revert back to our probability lecture so pretend that we have a bag of marbles that has these colored marbles in it so what's the probability of picking an orange marble there's only one in this bag out of a total of 14 marbles so the probability of picking an orange is 1 and 14. what's the probability of randomly picking a marble of any color except orange and the probability so that's the probability of not selecting an orange which would be
• 02:00 - 02:30 13 out of 14. so that should be pretty clear and easy it makes total sense here that if the probability of picking an orange is 1 in 14 the probability of picking not orange or any one of these marbles over here is 13 and 14. so the null hypothesis is a bit like the not Orange so our what's one way to think about this is our research hypothesis or our alternate hypothesis or H1
• 02:30 - 03:00 is like what's the probability of picking orange and our null hypothesis is what's the probability of picking not work so the null hypothesis is the sample space that does not include the research hypothesis and it's really important that we cover the full sample space when we're thinking about defining hypotheses so how do we Define hypothesis what's a hypothesis well oftentimes hypotheses are derived from observations so observations that we might make about the world for example statistics Canada
• 03:00 - 03:30 reported that in 2021 female employees age 25 to 54 earned 89 cents for every dollar earned by men okay well there's an observation now we can use that observation to make a hypothesis so our hypothesis might be in Canada being female causes an employee to earn less than being male and there are a couple of things I want to impact unpack here about this hypothesis
• 03:30 - 04:00 one of them the first one is the direction of this hypothesis goes in we've suggested that female employees earn less than male employees we also have a scope so we talk about in Canada because this is an observation made specifically by stat can um and Stat can really reports Canadian statistics so the generality or scope here this is the population to which the hypothesis applies this is the population of people who live in Canada
• 04:00 - 04:30 or employees who live in Canada I guess it's probably a better way to state that and we also have a mechanism associated with this hypothesis here I'm identifying a causal mechanism we're going to talk about that in a minute so being female causes an employee to earn less than being male and what I'm suggesting here is that this is causal I'm being a little provocative with this hypothesis by the way this isn't a hypothesis I would want to test and in fact it's not a hypothesis that one would be able to
• 04:30 - 05:00 test based on this observation but a causal mechanism and sometimes they're there sometimes they're not it's something to think about when you're when you're formulating hypotheses what is the chain of events that explains the observations we often use our hypotheses to try to explain why or how something happens and that is our mechanism or our causal mechanism of the of the observation now interestingly
• 05:00 - 05:30 we spend a lot of time talking about causal mechanisms that make sense and that don't make sense and those causal mechanisms may or may not be accurate so these are speculations about why something happens or how something happens that may or may not be testable in the context of the experiment this is something you need to be really careful about when you're reading statistical when you're reading statistical results you need to make sure that the method allows this the
• 05:30 - 06:00 researcher to talk about these specific cause otherwise what they are doing is something we talked about last lecture which is they're extending Beyond they're making a a non-methodological error which is they're extending beyond the data that they can that they collected they're extending their conclusions beyond the data so we need to be really careful here with different kinds of hypotheses this specific one has a causal mechanism in it and I will tell you right now it's a causal mechanism that we cannot test
• 06:00 - 06:30 but let's move on to another element of hypotheses directional versus non-directional tests so it's important to know what is being specified by a particular hypothesis because remember we need to complete the full sample space when we're talking about hypothesis testing non-directional tests do not specify the direction of an effect so we could say the treatment group
• 06:30 - 07:00 differs from the control group or the treatment group will differ from the control group on whatever our measure is and that is a non-directional test it does not specify a direction in general when we write these things you'll see them sometimes written like using the parameter notation so the MU or mean of the treatment group does not equal the MU or mean of the control group and the null hypothesis is the complement to this the mean of the
• 07:00 - 07:30 treatment group equals the mean of the control group or is not different from is another way to read that the null hypothesis or the the treatment group will be the same as the control group and that's our null hypothesis directional tests do specify a prediction a prediction direction for the effect so on our previous slide when we talked about that hypothesis about being a female employee in Canada causes you to earn less that causes you to earn
• 07:30 - 08:00 less or the earn less that specifies a direction of the effect so directional tests do specify a prediction Direction so we might say the treatment group will perform better than the control group if we have a medicine that's designed a new medicine that's designed to improve treatment for a particular illness or improve outcomes for a particular illness what we should see is that people who get that treatment actually are doing clinically better than people who are getting a control treatment
• 08:00 - 08:30 so this is in in terms of sort of these little this hypothesis notation we can talk about the MU or mean of the treatment group being better or performing better than the control group this is greater than we can also talk about we can we can specify the complement here which is that the MU or of for the null hypothesis which is the MU of the treatment group is either less than or equal to the mean of the control group we can also specify the opposite direction the treatment group will form
• 08:30 - 09:00 worse than the control group so we have treatment group lower than the control null hypothesis is that the treatment is greater than or equal to the control so you should be familiar with reading hypotheses and figuring out which direction they go in because I often put test questions about those things I will also remind you that sets of predicted events must be complementary hypothesis testing only works
• 09:00 - 09:30 if these sets of events are truly complementary to one another let's take an example here's an example in which there's a headline here actually this is another example that might be related to the statistics in the news assignment as well so here's a study being reported in fact it's being reported by CBS News and the study title or this the article title is study colon cereal keeps girls
• 09:30 - 10:00 Slim so this was published in September of 20 of 2005. so it's pretty old now and I'll just read you the first bit of this of this article girls who regularly ate breakfast particularly one that includes cereal were Slimmer than those who skipped the morning meal according to a study that tracked nearly 4 200 girls for 10 years girls who ate breakfast of any type had a lower average body mass index which is
• 10:00 - 10:30 a common obesity gauge than those who said they didn't the index was even lower for girls who said they ate cereal for breakfast according to the findings of the study conducted by the Maryland Medical Research Institute the study received funding from the National Institutes of Health and cereal maker General Mills hmm this is a hypothesis suggesting causation cereal keeps girls Slim
• 10:30 - 11:00 this is a study that tracked nearly 2 400 girls for 10 years and presumably they tracked body mass index and a bunch of other metrics they also asked these girls what they ate for breakfast or what probably whether they ate breakfast and if so what now this is a study in which the method there's no manipulation here so the method here is observational
• 11:00 - 11:30 claim that CBS News is making is that eating cereal keeps girls slim it's promoting a causational [Music] method rather than a correlational method this headline of course is much more eye-catching than serial is associated with lower body mass index might be a better way of stating this article the article title
• 11:30 - 12:00 um but in the grand scheme of things we need to think about how hypotheses work in the real world and we need to be really careful about what is being hypothesized relative to the methods so this is just an observational study this is a study that tracked and asked girls to self-report what they ate for 10 years nobody in this study was told you eat breakfast you don't eat breakfast you skip breakfast every day and because nobody was told we cannot make a
• 12:00 - 12:30 causal claim you cannot say cereal keeps girls slim there's no causal Association now of course this was also funded by the serial maker General Mills who might have a conflict of interest in this article title it might wish to sell you more cereal there are a number of possible explanations we could identify one is that eating breakfast causes girls to be thinner that's one possible explanation another is that being thin causes girls to eat breakfast maybe if you're thin
• 12:30 - 13:00 you get hungrier over the night and so when you wake up in the morning you're really hungry and you eat breakfast there's also the possibility that there's another variable that is connected to both of those first two that might be known as a confounding variable or an extraneous variable and that confounding our extraneous variable might be the explanation for both of these effects there might be something else going on that both causes girls to be thinner and that causes them
• 13:00 - 13:30 to eat breakfast so an extraneous variable that affects both the explanatory variable and the response variable might make it look like there's a direct relationship when in fact there's none for example perhaps girls who play sports where they have still a lot of running or where they're really athletically engaged maybe those girls are thinner because they expend a lot of calories they build a lot of muscle they work really hard and maybe that causes them to both be thinner and to desire to eat breakfast
• 13:30 - 14:00 every day they need to fuel up for their morning workout so that might be an extraneous variable that would explain these this hypothesis so when we are talking about hypothesis testing in addition to these elements of actual hypothesis testing which we will get into later in this lecture we also need to think about the methods because you cannot make a causal claim based on a correlational design
• 14:00 - 14:30 so let's talk about some vocabulary here the first piece of vocabulary I want to tell you about are explanatory and response variables so these are sort of how we think about how we explain how studies work an explanatory variable is the variable that you suspect is the cause of an effect we sometimes call that the independent variable or IV we sometimes call it the predictor variable
• 14:30 - 15:00 we also have response variables this is the variable you think is being influenced by the explanatory variable it's the variable that you are measuring so this is your dependent variable sometimes how we call that or Criterion variable now I want to make really really clear that these labels do not guarantee any kind of causal Association even if there's a relationship identified between the explanatory and response variable does not mean that the
• 15:00 - 15:30 explanatory variable is causal and that is extremely important order to show a causal relationship between an explanatory and a response variable you must have an experimental design in which you manipulate the explanatory variable so if we did a study in which we manipulated which girls ate breakfast and which girls did not randomly determined right we took our sample of people we randomly assigned them to the
• 15:30 - 16:00 breakfast group and to the not breakfast group then we could actually ask the question does eating breakfast cause girls to be thinner but here we cannot ask this question the hypothesis I proposed to you earlier that being female causes you to earn less this is another one of these that we cannot claim a causal relationship about there are lots of reasons why women might earn less than men these include the type of jobs women
• 16:00 - 16:30 tend to get or tend more often to get these include sort of all kinds of elements of working and work relationships claim causation there so we need to be really careful if a if an explanatory or independent variable was not manipulated you will have a very hard time claiming that a cause that something causes or that it causes the response variable so you need to be really
• 16:30 - 17:00 careful there because a lot of studies will try to do that they will try to trick you in that way let's talk about a few other definitions a confounding variable as we said is a very that co-varies meaning it is related to both the independent variable and that might and the dependent variable and that might provide an alternate explanation for what you are seeing as the effect that your that
• 17:00 - 17:30 you're measuring so variable a might relate to variable B that's the effect that you're measuring a copy causal it could be just observational but there might be another confounding variable that that is actually causing or related to um to both of those effects we also want to talk about placebos so when we are doing hypothesis testing it's very common to include a placebo or
• 17:30 - 18:00 control group so this is a treatment that is missing the active ingredient if we think about this in the context of uh trial that studies in medicine we could think about a sugar pill that looks just like the active medicine but it doesn't contain any of the active medicine so it contains all the sort of usual filler ingredients that that come up in various in various preparations of a particular medicine so usually there's a bit of chalk in there and a bit of this and that it's inside your medicines
• 18:00 - 18:30 turns it into a little pill that you can swallow um but is not active and so instead of having a pill with medicine in it it's a pill that looks just like the medicine group or that looks just like the medicine um that participants take if they and and they'd maybe perhaps even believe it to be active um these are often used in control groups because what they can then do is they can look at the effects of believing that you're taking a medicine
• 18:30 - 19:00 and actually taking a medicine versus the effects of believing you're taking a medicine and not actually taking a medicine and so the placebo effect you've probably heard of that in the placebo effect participants who are receiving Placebo show Improvement because they think they are getting an active treatment when in reality they're not so imagine that I think I'm getting a medicine and I therefore feel better sometimes Placebo effects also include you can see in fact we saw this a lot
• 19:00 - 19:30 with the coveted vaccines um people in the placebo condition reported lots of side effects of these vaccines even though all they were getting injected with was a little bit of a saline solution so they were not actually getting any dose of medicine and they were still reporting side effects because they thought they were getting the vaccine so that's the placebo effect people showing a change or showing an effect of the treatment even though in reality they're getting no treatment
• 19:30 - 20:00 blinding a study that is single blinded is when participants do not know if they are in the controller treatment groups so if I'm getting a trial covert vaccine am I actually getting vaccine or am I just getting a saline solution injected and when I do not know the answer to that I don't know whether I'm in the control of the treatment groups then that's a study that is called single
• 20:00 - 20:30 blinded a double-blind design is where neither the researchers who interact with participants nor the participants know who is in the control versus the treatment groups this is a very powerful technique double blinding and it's really important to look for in methods where experimental where experimental manipulations are administered because if the researcher knows who's in what group they can inadvertently bias the data and
• 20:30 - 21:00 we've seen that in our own lab where researchers who thought they were administering the good condition to people their participants were rated them as being more trustworthy and more friendly so they were actually changing their behavior without meaning to none of our researchers believed they were doing anything they thought they were administering administering the study truthfully and properly and in fact they were not it's really important in experimental designs to look for how
• 21:00 - 21:30 researchers describe what they did to double blind the design and that's going to be that's really important when you're evaluating research if there's an experimental design and you don't know how the double blinding was done that can be problematic and I'm going to start there for now and we'll pick it up in the next section