Latest / This Is Why with Dr. Busti / Power Analysis Explained: Type II Error, Beta, and Sample Size

Transcript

Power Analysis Explained: Type II Error, Beta, and Sample Size · This Is Why with Dr. Busti

• 0:00
  SPEAKER_00: All right, well, welcome.
• 0:01
  SPEAKER_00: We are in the biostatistics series on statistical analysis plans or SAPS SAP.
• 0:07
  SPEAKER_00: Okay, in this episode, we're going to discuss power analysis.
• 0:11
  SPEAKER_00: And this topic, guys, is relevant whether you're the participating in the research yourself or reading the clinical literature because it is really about not making a type 2 error.
• 0:23
  SPEAKER_00: And most people kind of glaze over this when they come to power analysis.
• 0:28
  SPEAKER_00: However, okay, especially when a study fails to find a difference, it matters.
• 0:34
  SPEAKER_00: So I'm gonna address not only what power is, what is it made up of, but also how to calculate it in a few minutes.
• 0:43
  SPEAKER_00: And I promise to make it easy for what.
• 1:35
  SPEAKER_00: However, there is a little bit of progression on some.
• 1:39
  SPEAKER_00: So if you need that, they're available out of the playlist on YouTube, our podcast channels, and at thisiswhy.health to make it easy if you need that guidance.
• 1:47
  SPEAKER_00: Now, we are down towards the bottom here of things that are done before the study's technically supposed to start by the researcher.
• 1:56
  SPEAKER_00: Hopefully the researcher tells you that.
• 1:58
  SPEAKER_00: If you're reading the paper as a clinician, right, or as a researcher reading other people's research, these are things that are done before the study starts.
• 2:07
  SPEAKER_00: Okay, you have to determine the type of data, how many groups, what type of groups, what type of analysis you're gonna do, and then you set the alpha, which becomes important also for power analysis.
• 2:17
  SPEAKER_00: So the two can't really be fully separated without some degree of context.
• 2:22
  SPEAKER_00: And I'll build some of that in here, but if you need a separate review on alpha, that is available for you on the channel.
• 2:28
  SPEAKER_00: Okay.
• 2:29
  SPEAKER_00: Part two is really about what happens when you finish summary statistics, those descriptive and inferential statistical analysis.
• 2:38
  SPEAKER_00: Okay.
• 2:39
  SPEAKER_00: And then the last one being about diagnostic tests, that has to do with you know the individual, how the test performs, but also you will use some of these principles if you're comparing two different tests.
• 2:50
  SPEAKER_00: Okay?
• 2:51
  SPEAKER_00: All right, now let's jump into power analysis and what is it?
• 2:56
  SPEAKER_00: By definition, it is one minus beta.
• 3:00
  SPEAKER_00: And I'm I'm gonna just pause here for a second for those of you that may be listening to this, not just for learning the topic and the concepts, whether you're learning to do research or active and involved in research or you're interpreting literature.
• 3:14
  SPEAKER_00: But if those of you studying for board exams, this is really important.
• 3:18
  SPEAKER_00: The boards love to throw these definitions, kind of curve balls all over the place.
• 3:23
  SPEAKER_00: Alpha does that uh interpretation as well.
• 3:26
  SPEAKER_00: It's the risk of making a type 1 error.
• 3:28
  SPEAKER_00: What is power?
• 3:29
  SPEAKER_00: 1 minus beta.
• 3:30
  SPEAKER_00: What is beta?
• 3:31
  SPEAKER_00: The risk of making a type 2 error.
• 3:34
  SPEAKER_00: So the smaller, if you just look at the math, right, just sit there and look at the equation and do the math, the smaller the beta is, the larger the power.
• 3:48
  SPEAKER_00: All right, well, that means it's also harder for you, okay, to do the study because you typically are going to need more patients.
• 3:56
  SPEAKER_00: We're gonna see that here in a minute when I show you a table, and I'm gonna walk you through that as you change the power, what it does to the sample size.
• 4:04
  SPEAKER_00: All right, again, I'm gonna show you how to do some basic calculations.
• 4:07
  SPEAKER_00: I'll show you the complicated way, not in depth, because who wants to know that unless this is what you do.
• 4:13
  SPEAKER_00: I'm not a biostatistician, but I trained and got a master's degree in evidence-based healthcare from Oxford, and I learned some of those principles, I did some of those calculations, and it's not necessary in all cases, especially if you're reading a lot of papers, okay?
• 4:28
  SPEAKER_00: All right, but uh power should be determined by the researcher a priori.
• 4:35
  SPEAKER_00: That means before the study starts.
• 4:37
  SPEAKER_00: That's why it's in part one, the statistical analysis plan, right?
• 4:41
  SPEAKER_00: Now the reader also should be able to determine that if it's not set.
• 4:46
  SPEAKER_00: Okay?
• 4:47
  SPEAKER_00: You might have, you might not agree that a power of 80% is sufficient.
• 4:52
  SPEAKER_00: You want to see a power of 90%.
• 4:54
  SPEAKER_00: Well, do they have the sample size to actually generate a power of 90?
• 4:59
  SPEAKER_00: Well, you could calculate the power analysis and determine that.
• 5:02
  SPEAKER_00: Right?
• 5:03
  SPEAKER_00: But what does it mean?
• 5:05
  SPEAKER_00: What is it?
• 5:05
  SPEAKER_00: Right?
• 5:06
  SPEAKER_00: What's it indicates the probability that a statistical test can detect a difference that's significant when it in fact truly exists?
• 5:18
  SPEAKER_00: Remember, we're trying to see if that difference is really there.
• 5:21
  SPEAKER_00: So said another way, kind of in plain English, is you know, what are the chances that my study, if I'm a researcher, will find a difference when in fact I uh when I'm going out to look for it, right?
• 5:35
  SPEAKER_00: So you're trying to set yourself up to be able to identify that so that, yes, when you say there's a difference, one, you don't want to make a type 1 error, but you also want to reduce the chance of making a type 2 error, right?
• 5:48
  SPEAKER_00: So you have to have enough patience to find that difference.
• 5:50
  SPEAKER_00: Okay, so another angle in this is since beta indicates the probability of making a type 2 error, then the power calculation tells you that the probability that you will not make a type 2 error.
• 6:07
  SPEAKER_00: One minus beta, one minus not making a type 2 error, because that's what beta is.
• 6:14
  SPEAKER_00: And remember, the smaller that number of beta, the larger your power is.
• 6:20
  SPEAKER_00: I know I'm saying some of these things a couple times, and I'm kind of throwing a little bit of what sounds like confusing language, but the reality is I'm saying the same things different ways.
• 6:32
  SPEAKER_00: And that's how you have to kind of get used to doing it, thinking about it, hearing it, especially if you're trying to prepare for one of those board exams.
• 6:40
  SPEAKER_00: But for those of you that have seen this in the hypothesis lecture, where I talk about null and alternative hypothesis, how to read them as it relates to your primary objective, because it helps you to orient around the type of data, the question you're asking, because that influences the inferential statistical analysis that you're going to do.
• 6:59
  SPEAKER_00: Right?
• 7:00
  SPEAKER_00: But in doing that, we don't want to make a type 2 error.
• 7:04
  SPEAKER_00: So this has to do with the you know, two by two with you know the decisions versus the reality, our decision as the researcher or someone reading the paper, but what is the truth, the reality?
• 7:17
  SPEAKER_00: Okay, and what we want to do is match those up.
• 7:20
  SPEAKER_00: But if you choose to accept the null, which means, remember, the null hypothesis, there's no difference in the exposure intervention that was done between the two groups at minimum.
• 7:31
  SPEAKER_00: There's no difference.
• 7:32
  SPEAKER_00: That's what you say, that's what the research says.
• 7:34
  SPEAKER_00: The paper published and says nothing different here.
• 7:37
  SPEAKER_00: But if the reality is that the null is actually false, then that is when you have made a type 2 error.
• 7:47
  SPEAKER_00: Okay.
• 7:48
  SPEAKER_00: Again, I'm just saying the same things.
• 7:50
  SPEAKER_00: I'm just looking at it from a number of different angles.
• 7:53
  SPEAKER_00: Some people find information clicks better when they see things like this, but but that's ultimately what we're trying to avoid here.
• 8:02
  SPEAKER_00: Whereas type 1 error is you have chosen to reject the null.
• 8:07
  SPEAKER_00: That means you're saying, hey, wait a minute, there is a difference.
• 8:11
  SPEAKER_00: We are rejecting the null.
• 8:13
  SPEAKER_00: We found a difference.
• 8:14
  SPEAKER_00: But the reality is the null is true, right?
• 8:18
  SPEAKER_00: And that has to do with our alpha value that is also set a priority.
• 8:22
  SPEAKER_00: And we'll see that the alpha value plays directly into a component of the calculation for power analysis.
• 8:30
  SPEAKER_00: That's why you can't ignore the two variables in the part one of this statistical analysis plan.
• 8:39
  SPEAKER_00: Okay.
• 8:40
  SPEAKER_00: All right, so what are the components needed to make this calculation?
• 8:45
  SPEAKER_00: There's a there's several.
• 8:46
  SPEAKER_00: So I'm gonna walk you through them individually.
• 8:48
  SPEAKER_00: But these components also directly influence the power.
• 8:55
  SPEAKER_00: All right?
• 8:55
  SPEAKER_00: So the sample size.
• 8:57
  SPEAKER_00: You can calculate from the sample size that was done between the in the each group and go backwards and determine the power that you the study has.
• 9:09
  SPEAKER_00: That's why I was saying if you, let's say somebody you thought that the study needed to have m a greater power.
• 9:15
  SPEAKER_00: Well, obviously, and to do that, which we'll see here in a minute, you have to have a larger sample size.
• 9:20
  SPEAKER_00: Well, does that study have enough patience in each group to justify the power that you think it needs to have, versus what the researcher a priori defined and hopefully put in a well-written format in the methodology section?
• 9:35
  SPEAKER_00: But the the sample size is huge.
• 9:37
  SPEAKER_00: Okay?
• 9:38
  SPEAKER_00: The standard deviation, right?
• 9:41
  SPEAKER_00: That variance of the significance that you're looking for, right?
• 9:46
  SPEAKER_00: Because nothing is ever a pinpoint and only that point, right?
• 9:50
  SPEAKER_00: That's that standard deviation recognized as sigma, and you'll see this symbol show up in the literature, or if you and well, I'll show it to you on the more complicated uh formula, not to overwhelm you or scare you off, but it's important to recognize why do we gather this stuff?
• 10:09
  SPEAKER_00: They should tell you that in a good protocol.
• 10:12
  SPEAKER_00: And then there's the practical significance or the effect size, and that's known as delta.
• 10:18
  SPEAKER_00: Okay.
• 10:20
  SPEAKER_00: What is the expected mean difference that you're looking for that for you is clinically relevant?
• 10:27
  SPEAKER_00: If you make that number small, you make it harder on yourself to find a difference, right?
• 10:34
  SPEAKER_00: And if it's too big, bigger, you don't need as many people to find a difference.
• 10:39
  SPEAKER_00: That's what it does to the power, that's how it influences power analysis.
• 10:43
  SPEAKER_00: Okay?
• 10:44
  SPEAKER_00: But you want an appropriate, fair, clinically relevant effect size.
• 10:49
  SPEAKER_00: So if I'm talking about, which I'm going to use an example here in a minute, when you're doing an analysis of a new antihypertensive medication compared to a placebo, if I'm expecting at least a mean reduction or would prefer a mean reduction of five millimeters of mercury pressure, that's the difference that I feel like is clinically significant enough to use this drug over the placebo or do nothing.
• 11:15
  SPEAKER_00: Right?
• 11:15
  SPEAKER_00: But even within that five millimeter mercury pressure drop, there is a variance around that.
• 11:21
  SPEAKER_00: Some people will get more, some people will get less, and that's that standard deviation piece.
• 11:25
  SPEAKER_00: So you see all those two really become important.
• 11:28
  SPEAKER_00: Now, if you remember in the alpha lecture that I did, alpha and looking at it, you have to put it in the context of the hypothesis.
• 11:37
  SPEAKER_00: Is it a one-sided or one-tailed direction thought?
• 11:42
  SPEAKER_00: Or is it a two-tailed, which most studies, well-done studies, let me say, that are published in most good journals with high impact factors, are going to be two-tailed.
• 11:54
  SPEAKER_00: Remember that one-tailed said you know the direction that the result is, you just don't know the magnitude of it, maybe, right?
• 12:00
  SPEAKER_00: That there's a degree of bias, like you already know something about this, the direction.
• 12:05
  SPEAKER_00: You're not willing to accept that it could be on the other side, potentially worse.
• 12:10
  SPEAKER_00: So a good, transparent, objective researcher doing good studies will try to consider the two-tailed.
• 12:17
  SPEAKER_00: But remember, you make it harder on yourself in order to actually find that difference because the area under the curve where that difference lies is now smaller for two-tailed because you divide by two.
• 12:29
  SPEAKER_00: That 0.05, that alpha 0.05 is divided by 2.
• 12:33
  SPEAKER_00: And that changes our z-scores.
• 12:36
  SPEAKER_00: Okay, that's what these, these, that's what this is.
• 12:38
  SPEAKER_00: The z-scores that are used in the formula, okay?
• 12:44
  SPEAKER_00: And then there's also the beta, the z-score for our beta, which is based on the power calculation.
• 12:49
  SPEAKER_00: Is it 80%, which is the most minimum that most people will accept?
• 12:54
  SPEAKER_00: And then you know, some of you navigate, we should be moving towards 90%.
• 12:58
  SPEAKER_00: Okay?
• 12:59
  SPEAKER_00: Because again, we want to reduce the chance of making a type 2 error.
• 13:03
  SPEAKER_00: So what do these things look like as it relates to sample size?
• 13:08
  SPEAKER_00: Okay, so if your beta, remember, one power is 1 minus beta.
• 13:15
  SPEAKER_00: So remember, the smaller beta is, the more patient, the greater the power is.
• 13:22
  SPEAKER_00: Okay?
• 13:23
  SPEAKER_00: So look here, but you're going to need more patience to do that.
• 13:27
  SPEAKER_00: So you can see here, a a we would typically accept most people say a beta a power of 80%, which is beta of a 0.2 or 20%.
• 13:39
  SPEAKER_00: 1 minus 0.2 is 0.8 or 80%.
• 13:43
  SPEAKER_00: That's a power of 80, right?
• 13:45
  SPEAKER_00: That's where most people typically reside.
• 13:47
  SPEAKER_00: Well, you can see that the sample size is 16.
• 13:49
  SPEAKER_00: And of course, you have your Z beta score here, right?
• 13:53
  SPEAKER_00: But as you reduce the beta, it gets smaller, your power goes up.
• 14:01
  SPEAKER_00: Well, so does the number of patients that you need.
• 14:04
  SPEAKER_00: Well, the practical relevance of understanding why that matters is time to complete the study, cost, feasibility.
• 14:13
  SPEAKER_00: Do you have access to those patients?
• 14:15
  SPEAKER_00: I mean, I've been I've been involved in NIH funded and other studies, randomized controlled trials.
• 14:21
  SPEAKER_00: It's not always easy to get a patient to meet inclusion exclusion criteria and get him enrolled in a study at a location.
• 14:27
  SPEAKER_00: It's a lot of work.
• 14:28
  SPEAKER_00: And so when we start seeing thousands of patients in a study, which is what we all want, right, that's a lot of work or requires a lot of sites with a lot of people doing a lot of work.
• 14:39
  SPEAKER_00: Now, let's look at the formulas.
• 14:41
  SPEAKER_00: Again, warning.
• 14:43
  SPEAKER_00: I'm not trying to scare you, irritate you, frustrate you, but you need to know at some level what the original or standard calculation looks like.
• 14:54
  SPEAKER_00: And we would use this particular, I want to get context, the formulas that I'm going to present to you are the more common ones, okay, for power analysis, especially when you're comparing two different groups and the means between those groups.
• 15:09
  SPEAKER_00: Okay?
• 15:10
  SPEAKER_00: And so you see that the first one, which is causes a seizure, lowers the seizure threshold by just looking at it.
• 15:17
  SPEAKER_00: Okay.
• 15:18
  SPEAKER_00: There's a lot of math here, there's a lot of Z's, alphas, betas, sigmas, deltas.
• 15:23
  SPEAKER_00: But remember, I gave you those where they most good studies will tell you that.
• 15:28
  SPEAKER_00: Now you won't say the sigma is, and the delta is, they're using the meaning of those things.
• 15:34
  SPEAKER_00: The variability of the result is the desired mean reduction is the effect size.
• 15:42
  SPEAKER_00: They're telling that to you in the methods.
• 15:44
  SPEAKER_00: You just have to be able to read it and interpret the information appropriately.
• 15:48
  SPEAKER_00: But the one I love and that works just as well and that is used a lot of times in protocol development is this one.
• 15:58
  SPEAKER_00: They're both ends, sample size.
• 16:00
  SPEAKER_00: Which one do you like?
• 16:02
  SPEAKER_00: Right?
• 16:02
  SPEAKER_00: So when here's that 16, where'd that number come from?
• 16:06
  SPEAKER_00: Well, that's just the standard for this assessment, which we already know what if you set it up things correctly, you have an alpha 0.05, it's two-tailed, which again is the most common.
• 16:18
  SPEAKER_00: You set a power of 80, then that multiplier that you don't have to do all this math all the time, it's not as complicated now anymore, is it?
• 16:27
  SPEAKER_00: It's just it's roughly 16.
• 16:29
  SPEAKER_00: It's actually 15.6 or something, something like that.
• 16:33
  SPEAKER_00: And you we just round up to make it easier, right?
• 16:37
  SPEAKER_00: So it's not an exact number, but that is our multiplier.
• 16:42
  SPEAKER_00: And you get that by just simply doing the math and you round up just a tad bit.
• 16:47
  SPEAKER_00: Again, we like to work with whole numbers, it's just easier than decimal points and all that.
• 16:52
  SPEAKER_00: All right, and it doesn't change it significantly.
• 16:55
  SPEAKER_00: But what happens if you have a nine 90% power or you do a one-telled analysis?
• 17:01
  SPEAKER_00: Well, that 16 multiplier now becomes 21.
• 17:04
  SPEAKER_00: So you just replace the the 16 with 21.
• 17:06
  SPEAKER_00: That's it.
• 17:07
  SPEAKER_00: But you already know the variance, the delta, the effect size, and the variance that they should tell you.
• 17:14
  SPEAKER_00: Okay, so let's let's put this in just basic application.
• 17:19
  SPEAKER_00: Okay, so we have that new anti-hypertensive medication, we're gonna compare it to placebo, like many many studies will do.
• 17:26
  SPEAKER_00: We're expecting that five millimeter reduction in the systolic blood pressure with a variability of 12.
• 17:32
  SPEAKER_00: That's what they're gonna say.
• 17:34
  SPEAKER_00: They didn't define the delta as that way, they didn't say it was sigma was this.
• 17:38
  SPEAKER_00: They use these kind of words.
• 17:41
  SPEAKER_00: But they're telling you the components of the power analysis without saying it.
• 17:48
  SPEAKER_00: The really, really good ones will give you those variables and even maybe tell you the power analysis on some level.
• 17:54
  SPEAKER_00: But why don't they do that?
• 17:56
  SPEAKER_00: Well, one, there's assumptions that people understand this stuff and can know how to interpret information provided.
• 18:02
  SPEAKER_00: And then, second is the practical limitation of space in these papers and these journals.
• 18:08
  SPEAKER_00: Okay?
• 18:09
  SPEAKER_00: But again, when you apply this and you set the alpha a priori two-sided or two-tailed, that means you're willing to accept that the new anti-hypertensive medication could be achieve what you're wanting, desired, or it could go the other direction.
• 18:23
  SPEAKER_00: It could raise blood pressure.
• 18:24
  SPEAKER_00: I mean, hopefully that doesn't happen.
• 18:26
  SPEAKER_00: Right?
• 18:27
  SPEAKER_00: That's bad.
• 18:28
  SPEAKER_00: But you're open to it.
• 18:29
  SPEAKER_00: That's being more uh transparent, less bias, risk, and a power of 80% at minimum.
• 18:37
  SPEAKER_00: Okay?
• 18:38
  SPEAKER_00: So if you map that out going back to that very complicated seizure-provoking formula, that is not really as scary as it is when you go back to these things and you just know what the values are.
• 18:51
  SPEAKER_00: What is alpha?
• 18:52
  SPEAKER_00: Well, we told it was 0.05.
• 18:53
  SPEAKER_00: Well, you divide by two because why?
• 18:55
  SPEAKER_00: Because it's a two-tailed.
• 18:58
  SPEAKER_00: Your beta.
• 18:59
  SPEAKER_00: You got one minus beta, that's your power, and you get your z-score.
• 19:02
  SPEAKER_00: We already know what that is.
• 19:04
  SPEAKER_00: We've already defined those.
• 19:06
  SPEAKER_00: Those come from tables.
• 19:08
  SPEAKER_00: Okay.
• 19:10
  SPEAKER_00: And so they're already predefined.
• 19:13
  SPEAKER_00: So we have our as standard deviation, that was the 12, that was the effect size that we were looking for was five.
• 19:19
  SPEAKER_00: So that's our delta.
• 19:21
  SPEAKER_00: So you just plug in the numbers.
• 19:23
  SPEAKER_00: It's not, you know, it's not that big of a deal.
• 19:27
  SPEAKER_00: It just looks complicated, it looks scary.
• 19:31
  SPEAKER_00: But again, if you're on the research development side, protocol development, you may have to do this and actually provide it because many times the IRB protocols are more detailed or in depth than what you finally publish.
• 19:45
  SPEAKER_00: You just tell people in a summary format what you published, what you did, and if they really want to get down to it and press it, they can, you know, obviously try to look into it.
• 19:53
  SPEAKER_00: But you calculate it, you get 91 patients roughly per group, right?
• 19:59
  SPEAKER_00: It's not perfect, it's not exact science, but that's roughly what it is.
• 20:02
  SPEAKER_00: So that means for each group, if you're going to compare two groups, that means a total of 180 patients.
• 20:07
  SPEAKER_00: But 90 in each group.
• 20:09
  SPEAKER_00: So you have to enroll that many people in order to achieve a power of 80%.
• 20:15
  SPEAKER_00: And if you're going to find a difference, hopefully that's where it's going to occur after that.
• 20:20
  SPEAKER_00: So some people will over-enrol, which obviously improves the power because some people could drop out, and you don't want people to drop out for some weird reason, and then you fall below a power, and then you don't find a difference at the end, now you've got a huge risk of a type 2 error.
• 20:37
  SPEAKER_00: So if you compare that over here, that this is the math on that the simplified equation.
• 20:43
  SPEAKER_00: Again, assuming alpha 0.05, two-tailed, your multiplier power of 80, you're it's 16, so all you do is just plug in these two values, and you do the math, right?
• 20:54
  SPEAKER_00: So 16 times 12 squared, then you you take that and then you divide it by 5, which is your effect size, right?
• 21:05
  SPEAKER_00: Then you get 92.
• 21:06
  SPEAKER_00: So there's there's a small discrepancy, okay, that's one to two patients.
• 21:10
  SPEAKER_00: So you error and say, I'm gonna work, I'm gonna do 95.
• 21:14
  SPEAKER_00: All right, but I at minimum need 91 to 92 to do that.
• 21:18
  SPEAKER_00: Now, what if you want to go the other direction?
• 21:21
  SPEAKER_00: Maybe they didn't tell you the power.
• 21:24
  SPEAKER_00: It wasn't a good paper, it wasn't a good research.
• 21:26
  SPEAKER_00: And you want to know what is the power of the study?
• 21:29
  SPEAKER_00: Well, you can you can work the equation backwards, and you can calculate the power, right?
• 21:39
  SPEAKER_00: So if the let's say it was 40, there was only 40 people instead of the 91 minimum.
• 21:46
  SPEAKER_00: Well, you you can convert the power by using a standard table or calculator, right?
• 21:51
  SPEAKER_00: That's what kind of gives this value here.
• 21:55
  SPEAKER_00: But you can go backwards, and again, you're just filling in the sample size that you have.
• 22:00
  SPEAKER_00: Delta and the and the sigma that you've been given, and you know your alpha.
• 22:04
  SPEAKER_00: So you can plug it in and go backwards.
• 22:07
  SPEAKER_00: Okay.
• 22:08
  SPEAKER_00: And so I run through the synopsis right here for you if you want to dive into it.
• 22:14
  SPEAKER_00: Obviously, some of these have to do with some of the, if you use a Z approximation or if you use other tools to get you get roughly around this percent, right?
• 22:24
  SPEAKER_00: There's there are variations to the, there's usually a range, right?
• 22:28
  SPEAKER_00: But let's say it's 46%.
• 22:31
  SPEAKER_00: This is your power.
• 22:33
  unknown: Right?
• 22:33
  SPEAKER_00: So if you take that and subtract it from 100%, then your beta is obviously 54.
• 22:42
  SPEAKER_00: So your risk of making a type 2 error is now almost 60%.
• 22:48
  SPEAKER_00: So if you finish the study and there was no difference, and you failed to reject the null hypothesis in this situation, that would suggest that this is really not a useful study.
• 23:02
  SPEAKER_00: I should stop reading it right away.
• 23:04
  SPEAKER_00: I should never, it should never have gotten published.
• 23:06
  SPEAKER_00: The peer review process in an appropriate journal would never have allowed that to get published.
• 23:12
  SPEAKER_00: Right?
• 23:13
  SPEAKER_00: Because it's junk in, junk out.
• 23:15
  SPEAKER_00: I don't know what to do with that.
• 23:16
  SPEAKER_00: There's too high risk for a type 2 error here.
• 23:19
  SPEAKER_00: And so remember, power is that probability of not making a type 2 error.
• 23:25
  SPEAKER_00: And the smaller our beta, the larger our power, the larger sample size we need.
• 23:31
  SPEAKER_00: But insufficient power fails to allow you to rule that out.
• 23:35
  SPEAKER_00: Okay?
• 23:36
  SPEAKER_00: Alright, so you can access this topic along with the other ones that I kind of mentioned on hypothesis testing and getting oriented on that.
• 23:44
  SPEAKER_00: Alpha, okay, power, all these are put in a playlist for you and organize on YouTube and specifically at thisiswhy.health.
• 23:54
  SPEAKER_00: You can get other collections there, evidence-based medicine, foundations, and topics for those of you learning the process, you're learning how to do research, or maybe you've already seasoned at some level, but you need to sharpen some perspective, or you want to review, or you're preparing for a board exam.
• 24:12
  SPEAKER_00: Or you're the clinician reading the paper.
• 24:14
  SPEAKER_00: Either way, these tools are designed to help you get through that process.
• 24:18
  SPEAKER_00: If you're a faculty or residency program director, we do provide curriculum support for evidence-based medicine, biostatistics, clinical research that most programs require for your milestones and competency standards.
• 24:33
  SPEAKER_00: We do that for medicine and pharmacy and obviously nursing as well, does it at some level.
• 24:38
  SPEAKER_00: We provide that service at high yield med reviews.
• 24:40
  SPEAKER_00: We also have cue banks for you to practice.
• 24:42
  SPEAKER_00: And that's an investment to equip the next generation.
• 24:46
  SPEAKER_00: If you're interested in that, please check us out at highyieldmedreviews.com.
• 24:50
  SPEAKER_00: You can go to curriculum support, and we can help you out or reach out via social media or even post a comment.
• 24:59
  SPEAKER_00: So I hope you found it helpful.
• 25:01
  SPEAKER_00: We'll see you next time on the next topic within the biostatistical series.
• 25:05
  SPEAKER_00: Take care, everybody.