 Scientific progress is about pushing the boundaries of what we know about how the world works. This happens by looking at data from experiments. Experiments are designed to test the hypothesis and either support or reject it. Before you begin an experiment, you have to think through the possible outcomes and what they might mean. However, these new discoveries are only meaningful if we make sure we do a good experiment. Let's talk about how to choose a good experimental question, design a good experiment, acquire and analyze the data properly, and why repetition is so important. If we design bad experiments, we mistakenly might end up thinking that just because we are observing two things together, like one, you carry an umbrella in your bag, and two, it starts raining on your walk, that one of those things is causing the other. You are causing it to rain on your walk by carrying an umbrella in your bag. The important first step is to choose a particular question we're eager to answer. Let's pick a totally crazy question. If people do five jumping jacks as soon as they get out of bed in the morning, will their hair turn purple? There are a few key things we must think about. First, we need to clarify how we plan to test the question. We will need a group of people who will do jumping jacks, but it's also essential to have a group of people who do a different exercise, say five sit-ups. And finally, a group of people doing no exercise at all, the control group. Then we can be sure that any effect that we do see on hair color is not due to doing any type of exercise, jumping jacks or sit-ups, but is definitely due to the jumping jacks specifically. We want to make sure there's only one difference between the groups. Secondly, we need to think about which population we're testing. It's often impossible to assess the whole population. For instance, every person in the world or every person with a specific feature you're interested in, because it's too expensive and time consuming. So instead we assess a smaller sample. One problem is that in biology there's a lot of natural variability. If the experiment is not large enough, for example three people in each group, there is a high likelihood that the wrong conclusions will be drawn, because our experimental groups may not properly represent the whole population. Experiments can help by giving us tools to calculate the minimum size that our sample should be. This can often be hundreds of people. We will also now plan how to analyze the experimental data after collection. So we've got the experiment planned out. We have the question, population type and size, and data analysis plan. And we are ready to go. There are a few key things to remember while doing the experiment and acquiring the data. The first is randomization. If we put all the blonde people in one group and all the brunettes in the other, any changes we see might be due to differences in hair color at the start of the experiment, rather than the exercise. Randomization solves this issue, which means allocating whether someone should be in the test or a control group in a random way. The second is blinding. This means that both the participants and the experimenters don't know who is in which group where possible. In this case, it's pretty tricky, as you can't hide from the participants which exercise they are doing. But once the data has been acquired, it can be labeled in a way that means the groups are hidden to those doing analysis. Humans are very susceptible to bias, conscious or unconscious. So if the experimenters know which group is doing which exercise, they might assess the final hair color of the people in the jumping jacks group as more purple, even if it is not. Finally, the experiment should be repeated by the original experimenters and others around the world to confirm the results are not random chance. So remember, when designing an experiment, you need a good question, elegant experimental design with a control group, to carefully choose how many and which people you will include and what data to collect, sensible analyses and replication. There can be very serious consequences if these concepts are not implemented. If decisions on new drugs or diagnostic tests are made using experimental data that haven't properly tested the hypotheses, patients' lives can be affected. Elegantly designed, appropriately controlled experiments on the other hand can provide valuable contributions in our quest to better understand the world.