Okay, In this video, we're going to talk about how we can increase the statistical power of an experiment. This might seem like a really silly question to ask because you might immediately say, well, shouldn't we just increase our sample size? The answer is no. That's not the first thing that we should do. And that's be for a number of reasons. It could be because there might be ethical considerations. That's probably most important. There'll be financial considerations. And also, the larger an experiment becomes, so the more subjects that we have, the more difficult it becomes to run. And if an experiment is more difficult to run, we're more likely to make mistakes and jeopardize all of our hard work. So what else can we do to increase our sample size? Sorry, What else can we do to increase our statistical power other than increase sample size? For answer that question, I want to point you to this really nice book by Nicole graven, grand Ruxton, power analysis, an introduction for the life sciences. I've based this video off of Chapter 3 in this book. Overall, the book is really nice. It provides a great overview of power analysis. It talks about how to perform power analysis using simulations. And it's very accessible. It's written for undergraduates, but it will help anyone. So I want you to be aware. Now, how can we increase the statistical power of an experiment? Answer that question. We need to first consider what statistical power depends upon. Statistical power depends on the size of an effect, the amount of variation that's inherent to our data, and on the sample size. So our approaches to increase statistical power will depend on these variables. Okay? So how can we increase statistical power? The first thing I might consider is whether or not we want to choose one particular variable over another as a metric that we're using to represent some biological phenomenon we're interested in. I know that sounds very cryptic. Basically what I'm trying to say is that when we're trying to study some phenomenon, there might be a number of different things that you could measure that you think would be indicative of the phenomenon you're trying to measure or the phenomenon that you're trying to understand. And some of those things you might measure might be more prone to bias or they might be more than, might be measured less precisely than other variables. So the first decision you can make when thinking about how to increase your statistical power to choose the variable that you're measuring carefully so that you can avoid bias as much as possible. And so that you can be measuring something that says biologically relevant as possible, as well as being able to measure consistently and well, so precisely and accurately. The second thing you might do is try to increase the precision with which we can measure the variable that we're interested in. So for example, we might use better instruments in order to collect our data. Can often be very costly. And so another thing we might consider just improving the protocols that we use in our experiment. We can also try to decrease the inherent variation in our data using two very related techniques. I've well, CO given Ruxton refer to them as subsampling and repeated measurement. I'm going to illustrate these ideas just with a simple example. Let's imagine you wanted to study the qualities of brain cells. So you're a neuroscientist and you're interested in the qualities of brain cells. Brain cells might be difficult entities to measure, to measure consistently, okay, there might be a lot of inherent variation in our measurements of brain cells. So there's, we can help decrease that inherent variation. First of all, by choosing, say, one cell and measuring that one cell multiple times. And then using the average value of that measurement or using some other technique that would account for pseudo replication that could arise from measuring the same thing more than once. That point is if we measure that exact same. More than once. And that can help us to achieve more, a more reliable over all estimate for the qualities of that focal cell. Subsampling is very related. Where in this case, what we might do is if we're trying to understand the qualities of say, the average cell in a particular brain. Than what we might do is sample a variety of cells within that brain, measure each one, and then use the information from all of those cells together represent the typical cell for that brain. So we might use an average or we might analyze our data in a way that accounts for studio replication. Okay, so what else can we do? Well, we can be very selective about what experiments and material we choose to use. Specifically might use experiments and material that is likely to reduce inherent variation. So we might work with inbred lines or clones. If we have the option of working in a lab versus working out in a forest, a lab might provide a more consistent environment. And as a result, we could decrease inherent variation in that environment. If we're studying a population, we might choose to focus on a particular demographic, okay, as opposed to studying the entire population. All of these approaches will help to reduce in here and variation in our data. There's a trade off, however, that we've talked about in other videos, which is that the more restrictive that we make, our type of sample that we're using or the location of our experiment. The less we're able to generalize about the results from our experiment. But that trade off is one that we should think about carefully. And it may very well end up causing us to choose things like inbred lines are clones as a means to decrease variation and increase our power. The next thing we can do, it shouldn't be surprising is we can change the design of our experiment or use a different type of statistical analysis. So what do I mean by changing the design of the experiment? Well, consider these two experiments, okay? In both cases, we have a treatment group and a control treatment. Okay, So here's treatments and here's control. And in both these cases we have the, we actually have the same number of mice being used in our experiment. But if we were to do a power analysis to compare the power of these two experiments, we might find that one of these experiments is more powerful than the other. Okay? So when we perform a power analysis, we rarely want to consider only one experimental design. We probably want to consider a variety of experimental designs and choose the most powerful among those possibilities. We can also incorporate other features in our experimental design that would help us to control inherent variation. So we can incorporate blocking or we can include covariates in our analyses. These are topics that we talk about in our video series on experimental design. If you want to learn more about those options. And now finally, we end with our option of increasing sample size. So those are a variety of options. And you can see that certainly increase in sample size is not our only option. We should probably consider all of these other questions first before choosing to increase our sample size as a means of increase in statistical power. I'm going to end the video there and say, I hope it's been helpful and thank you very much.