Speaker 1: How does a researcher avoid bias? For example, a pet store owner's sampling method has selection bias in the form of non-response because only 5 people responded to a survey. What can she do to get a larger sample to avoid this type of bias? In this lesson, you will learn how to conduct a better statistical investigation by avoiding bias. Let's review. What is statistical bias? This is the tendency of a sample statistic to over or underestimate a population parameter and occurs when sampling or measurement systematically favors a certain outcome. Types of bias. First major type is selection bias, which is where sampling that generates a sample is not representative of the population. Three types of selection bias. Under coverage, a convenient sample, and non-response bias. The next major set of bias is what's called response bias. There are two types, social desirability and leading questions. Let's move to our main lesson. What are the implications of biased research for under coverage? Under coverage means that if you have bias there, you cannot make conclusions about the entire population. For example, if the population contains males and females, and freshmen, sophomores, juniors and seniors, but only freshmen are in the sample, conclusions on just the freshmen cannot be projected onto the sophomores, juniors and seniors. So how do you go about avoiding biased research in terms of under coverage? You can generate a large enough sample. For example, when the students were sampled, the principal might not have pulled enough students for the sample to make it representative of the population. The second thing is to make conclusions based on the characteristics of your sample. For example, we could make conclusions about the freshmen only. The third thing would be to regenerate a sample to make it more representative of the population. For example, we could regenerate a sample to make sure that there are freshmen, sophomores, juniors and seniors in the sample. Let's move to the implications of biased research for a convenient sample. Again, we cannot make conclusions about the entire population. For example, the biology teacher is standing outside her classroom in the biology building asking students what their major is. Most of the students are in the sciences. She is in the science building, so students would probably be majoring in science. Is that indicative of what's happening in the university? How do you go about avoiding biased research in terms of a convenient sample? We could regenerate a sample to make it more representative of the population using one of the two other methods for better sampling, which would be a simple random sample or a systematic sample. For example, our biology teacher could get a list of all the students at the school and generate a simple random sample. What are the implications of biased research in terms of non-response bias? Again, we cannot make conclusions about the entire population. For example, a representative sample is sent a survey, but only 10% of the respondents send back the survey. If the responses of this 10% is not representative of the population's responses, conclusions about the population cannot be made. So how do you go about avoiding this with non-response bias? You want to generate a large enough sample. For example, you know that not everyone will participate in a mailed out survey, so you could generate a larger sample than you need to make sure you actually get the number and the representation that you want. If you've already sent out your invitations, then you could resend them out to potential subjects to get them to participate in the research process. Sometimes it takes many mailings to get a representative sample. You could also regenerate another sample to make it more like the population. If the original sample chooses not to respond, then a new sample can be generated to see if they will respond. You might have to put the two samples together. What are the implications of biased research for social desirability? When we move to response bias, you're going to over or underestimate a population parameter because people want to look good or favorable in the researcher's eyes. For example, this person is given a survey on alcohol usage. However, when asked whether he uses alcohol, he chooses not to say that he uses alcohol. So how do you go about avoiding biased research in terms of social desirability? Sometimes you have to play a little bit of tricks on the mind. The first one is called a bogus pipeline. A bogus pipeline is where respondents are brought in to be interviewed and they are hooked up to a lie detector machine. However, the responses are not being monitored for truth or falsehood. They're just there. It's a psychological trick in order to get respondents to tell the truth more often. The second one is what's called self-assessment questionnaires, which are anonymous and where respondents are given surveys in private so that the researchers do not know who puts what for answers. The third one is called selection of interviewers. When the content of the research is sensitive, it is very important to choose interviewers that are sensitive to the respondents. The last one is called use of proxy subjects. Subjects do not answer the questions themselves, but another person answers the questions on their behalf. What are the implications of biased research for leading questions? Again, we will over or underestimate the population parameter. Questions are worded in such a way to get respondents to answer them a certain way. So how do you go about avoiding biased research in terms of leading questions? We need to avoid leading statements. For example, this is a leading statement, 9 out of 10 doctors recommend that you brush your teeth after every meal. What do you think? Well, this could be changed to when do you brush your teeth? The second one is at all costs avoid putting in research opinion because you want to find out what the sample is thinking. For example, don't you agree that teenagers are using cell phones too much? This could be changed to what do you think of the amount of cell phone usage by teenagers? You can change leading questions to neutral questions. For example, I assume you like the new senator, right? This could be changed to do you like the new senator? Let's go back and look at the lesson introduction. A pet store owner's sampling method has selection bias in the form of non-response because only five people responded to a survey. What can she do to get a larger sample to avoid this type of bias? A couple of things. The pet store owner could resend the surveys to those who did not respond to see if they will respond. However, since only five responded, it seems like that the pet owner's sampling method is flawed. So to overcome that, she could generate a random sample by asking customers to fill out the survey while they are in the store. In this lesson, you have learned how to conduct a better statistical investigation by avoiding bias.