So in this section, I want to say a few words about sampling, taking a sample from a population. Now obviously investigating a whole population would be best. But even if we lived in a moderately sized city of, say two million people. If, I wanted to get data on all two million people, it could be very time consuming. Time consuming to send investigators out to get data on those two million people. It would be very expensive as well. You can imagine, and if we upscaled it. It gets even worse. Now fortunately that's not always required. We do have what we call inferential statistics and that's what we're going to learn about. We take a sample of from the population and we investigate that sample only as a representation of the larger population. Imagine we knew what a patient parameter was, by some divine intervention we knew what that was and we took a sample and we calculated sample statistics and those two differed from each other completely. Then we know something went wrong. Some error happened. There was some, a bias was introduced. Now, the number one solution for this is to make the sample as big as possible. Large sample size has solved many problems when it comes to random sampling, now think about these things, if you read a journal article, again as to say an infer it to the population that you are seeing. In order to do that think about certain things is that population from which the sample was taken even the same as the population that I'm dealing with. Let's take an example of gastric carcinoma cellular phase in Japan. In your beautiful study there, absolutely correctly randomly sampled, good statistics can infer to the population, but is the population of patient with gastric carcinoma in Japan the same as the population with gastric Carcinoma in Europe or Northern America. You've got to ask yourself that question. Remember, not all populations and not all diseases really are the same throughout the world. And then always look for introduction of bias in the study, even if the populations are comparable. Was there any introduction, Was there any bias in selecting those patients? Very important to look out for. Now when we talk about a sampling error that is actually what happens when everything goes right. The random sampling, selection, it was really done properly. But because it is just a sample, there is a chance that it slightly differs from the population from which it was taken. And that can happen under the best of circumstances. That's when we refer to just sampling error. When you talk about bias, though, there was a systematic inclusion of some incorrect process. But really that example is really different from the population from which it was taken. When you read a randomized control trial, don't only look for proper randomization of the largest sample set. Take a large sample set and randomize them into two groups. Those two groups were done randomly or correctly but as a group, does that group represent the population in which the results we want to use and then lastly when you read an observation study remember that's very difficult not to have some form of bias in there, so clearly look for the bias that was introduced, possibly introduced in the selection of the individuals for whom data was taken to do that analysis. Now, if we want to do this we've got to strive for some sort of ideal, something that will improve our accuracy, the accuracy of getting that sample. And there are two ways to go about it. Obviously make the sample size as large as possible. When you read general articles look at that sample size. Was it as large as possible? And also do not introduce bias. The less bias there is in that sample the more representative it is of the whole population and the population that you see in the clinic, in the hospital, and have to manage. Now there are four types of sampling that, I want to discuss. The first one is going to be simple random selection. That's where we have a master list. And everyone on the list has an equal likelihood of being selected at random. This is the systematic random selection. That is where we're just going to as a system just choose every individual equally space. So every tenth individuals, every 20th individual on a list. We're going to look at clustering. That is where we don't have a master list. But at least possible participants are clustered in some way and we can choose those whole clusters as a random process. Lastly some words on certification, this way there's a mutually exclusive trait in the population. Which immediately separates and we can take participants from each of those groups.