[Music] Welcome to the research methods unit. In this video, we'll cover the formation of testable hypotheses, types of variable, sampling methods, designing research, correlation, research procedures, planning and conducting research, ethical considerations, and we're going to cover data handling. This include quantum and qualitative data, primary and secondary data, computation, descriptive statistics, and the interpretation and display of quant data and normal distributions. I've updated the Cypoost app. Now it has over 500 GCSE flashards, over 400 multiple choice quiz questions, and over 300 key terms to test your knowledge with. Of course, all based on the information I cover in these videos. I'm so confident you'll like my app, you can use all the features for paper one for free. Only subscribe to paper 2 if you know you'll find it helpful. You can also follow along with these videos by completing my worksheets available on Patreon. Formulation of testable hypothesis. You might have a general area of psychology that you want to investigate. Say, I want to investigate how memory can be improved. But to give an aim, we need to narrow this down to a clear statement that outlines what you as a researcher intend to investigate. The aim can include why you want to investigate it. So stating the purpose for the study. Purposes could be because previous research raised interesting questions or maybe you're directly replicating someone else's work. We can phrase an aim as a question or a statement. So an example aim could be the aim of this research is to find out if color can influence recall. We can even add due to previous research showing exposure to green plants increase recall. So with an aim, we know what we're studying and perhaps why, but not how. We need to be more precise if we're going to conduct a study and importantly communicate to other researchers exactly what we're doing. So after our aim, we're going to make or formulate a hypothesis. A hypothesis is a precise statement that importantly can be tested. This means I need to state how the parts of the study interact. So in an experiment, I would state in my hypothesis exactly what the levels of independent variables are and what the dependent variable is. In this example, let's say my hypothesis is there is a difference in the number of words recalled by participants recalling in blue light compared to participants recalling in green light. I would also need to be careful to operationalize my variables, meaning I state them in a way which shows how they're measured. So instead of recall, the number of words recalled. Now, an important point about hypothesis. They're not predictions. They're statements of fact. Now, of course, we reject or accept a hypothesis depending on the evidence we collect. So, what's going on? Well, we actually have two hypotheses. As good scientists, we begin with the principle that there is no difference between the conditions or no relationship between the variables. This is called a null hypothesis or H0. Of course, scientists want to be able to show that there is a relationship between variables. But similar to a courtroom where they say innocent or proven guilty, we accept the null hypothesis unless we collect evidence powerful enough to reject it. Our other hypothesis you might be more familiar with. It's called the alternate hypothesis or H1. And this just means it's the alternative to the null and it suggests that there is a difference between the conditions. It's also known as the research hypothesis. In hypothesis testing, we start with both hypotheses, collect our data and then use statistical tests on the data. This then provides evidence and if the evidence is strong enough, we can reject the null hypothesis and accept the alternate hypothesis. Types of variable. In correlational studies, researchers measure co-variables. They take two measurements and then compare the measurements looking for relationships. These can be positive relationships. As one co-variable increases, so does the other. Or negative, as one co-variable goes up, the other one goes down. In correlational research, we can say two co-variables are associated. But we can't say with co-variables that a change in one cause a change in the other. It might be the other way round or that a third variable is responsible for the change in both. In an experimental setup, a researcher will manipulate or change one variable. This manipulated variable is called the independent variable. So for example, the researcher might alter if the participant is given a drug or a placebo is given rewards for behavior or in a control group without rewards or is in green light or blue light. These levels of independent variable are the conditions of the experiment. The researcher will then measure any change in the dependent variable that could be the reduction in symptoms, the production of behavior or recall. An important concept is operationalization. When stating the dependent variable, I need to specify exactly how it's being measured. So if I wanted to know how effective a course of say anger management therapy was. Instead of saying reduction in anger, I might measure it by a reduction in the score on a questionnaire that measures hostility or a reduction in the number of times a student has had a verbally aggressive outburst to teaching staff over say the next month. So that's the ideal. In an experiment, researchers can claim to have shown a causal relationship between the independent variable and the dependent variable. But of course, that's not all that's going on. There are other variables that can influence the dependent variable. And if these aren't controlled for, there won't have been an accurate measurement of the independent variables influence on the dependent variable. We could then suggest that the study's findings are not true because the researcher didn't set the study up in a way that controls for other explanations for the findings. Or in more scientific language, the study lacks internal validity. Extruous variables. Extruous variables are the name we give to any other variable other than the independent variable that could influence our measurement of the dependent variable. And there are lots of potential extraneuous variables. For example, demand characteristics. These are cues that the participant picks up on that suggest how they should behave. This could be verbal or non-verbal communication from the researcher and would call this experimental effects. Or maybe it's the way the experiment is set up that gives away the aim. If the participant works out the aim, it's very tempting for them to act in a way that they think is helpful or even unhelpful for the researcher. The problem is then that the researcher is measuring the effect of demand characteristics on the dependent variable, not the effect of the change in IV. Other examples are participant variables. Characteristics like age, gender, and cultural background can influence behavior as well as any prior knowledge or skills related to the task. Situational variables are environmental factors that could influence performance like any noise in the lab, maybe the temperature or visual displays. order effects. So in a repeated measure design, practice or fatigue could be an additional variable. So you can imagine how these factors, especially if they vary significantly between the conditions, can influence the performance of participants. Control of extrrenuous variables. Researchers who want to claim their results are valid, are going to want to take actions to deal with extraneous variables. To control for participant variables, we can use random allocation and match pairs. The reason independent groups design should always use random allocation between groups if possible is to control participant variables. For example, pre-existing knowledge, skills, or characteristics like age and gender. If you use randomization, you're likely to spread these variables out across the conditions, but you may still get unbalanced groups by chance. A better way to control participant variables is to run a match pairs design. This is carefully measuring a participant variable before the study and then matching the highest score in one group with the next highest in the other group. This will result in the participant variable being balanced between the groups. To control for order effects, we would use counterbalancing. As completing one condition first can change participants performance in the second condition by giving them practice or making them tired. Counterbalancing, splitting the sample of participants in half and getting each half to do one of the conditions first means any effect due to practice or fatigue is canceled out because it's the same in both conditions. More importantly, while we can say we've controlled for order effects, we can't say that they've been eliminated. To control for situational variables, we use standardized procedures. Environmental effects such as the temperature, lighting, or sound levels in the room can form part of a list called standardized procedures. This basically is a recipe the researcher needs to follow to give each participant the same experience. Of course, aside from the variation in the independent variable. So, the researcher would then make sure the participants in each condition had the same experience. To control for demand characteristics and investigator effects, we can use single and double blind trials. So to counter the participants changing their behavior because they figured out the aim. Well, the first thing you can do is not tell them the aim. That's a single blind trial. The participant is unaware of the true nature of the experiment. In order the researcher influences the results by say giving hints of their body language. So to limit investigator effects, we can use a double blind trial having the researcher who carries out the experiment be completely unaware of the true aims of the experiment. Carefully created standardized procedures with a script for the researcher to follow is also useful in both cases reducing the chance of giving away the aim or the researcher using different language in each condition. We can use pilot studies and peer review to identify extrrenuous variables. Sampling methods. Our first definition is for the term target population. This is every individual that forms part of the group that you plan to study. So likely it's going to be a very large number. If you're investigating pensioners or four-year-old children or six form students, we won't be able to test all of them. We need to take some of them, a sample. What we hope to do then is get our results from that sample and then apply the results back to the target population. This is called generalization. However, members of a population vary in many ways. So, ideally, we want a sample that is representative of the larger population. Random. So, let's start with random sampling. Now, it's not just grabbing anyone to take part in your study. It's mathematically random. So, everybody in the population has the same chance or probability to be selected as a member of the sample. A researcher first needs a list of all members of the population. And then you use a method of selecting them randomly. So, putting the names into a hat and drawing them out until they have a full sample or giving each name a number and using a random number generator. The strength of this method is it avoids researcher bias. The researcher can't just choose the people they want in the study which could influence the results. However, because it's random, we could randomly get an unrepresentative sample. Maybe not representing all minority groups. If the population size is large, then a random sample can be timeconuming. Systematic systematic sampling is similar to random. We still need the list of the population, but instead of picking randomly, we go down the list and choose every fifth or 10th or nth person. And you can imagine a teacher picking a sample from her class using the register and calling out every third name. This again removes a chance of researcher bias in picking who they want for their studies. And with a small population studied and list of the population already existing such as the register, this can be a quick way of getting a sample. It's unlikely but still possible to get an unrepresentive sample using a systematic approach. And with large populations, it's difficult to get a full list of members. Opportunity. An opportunity sample is the easiest sample to get and the most commonly used. The researcher simply includes anyone in the sample that they can get their hands on simply by asking them to take part. For that reason, many psychology studies are actually conducted on university students. A strength of this, of course, is it's a much faster way of getting a sample than any of the other methods. This could save money and allow the researcher to complete the study faster. But there are big problems of opportunity samples. There's a potential for researcher bias. The researcher decides who to ask and who not to ask, potentially manipulating the results. Also, the sample is likely not representative as a researcher only has access to a limited section of the population, in most cases, young university students. Volunteer. Another word for volunteer sample is self- selecting sample. This makes clearer the important factor of this sampling method that the participants select themselves. They volunteer themselves. They're not directly asked. So they might see an advert in the newspaper or online and put themselves forward. So a strength of this is by using an advertisement, especially in a popular newspaper or social network, the researcher can reach a large number of potential participants. And it's relatively easy to collect as after placing the ad, the participants are putting themselves forward. But we do have the issue of volunteer bias. People who volunteer for studies are a certain type of person. They're of course helpful and they have time to take part in psychology studies, but we want to include people who are unhelpful and people who are busy. If we don't, we may not be able to generalize our findings to the wider population. Stratified. A stratified sample is the most complex type of sample, but it tries to avoid some of the problems of the other methods. A stratified sample creates a sample that is representative of the population as a whole. So firstly the researcher will identify subgroups or strata and their proportion in the wider population. Then the sample is made by randomly selecting participants from within each strata. So they're represented in the same proportion in the final sample. So if 10% of your population were university graduates, 10% of your sample would be university graduates. And the big positive of this approach to sampling is a sample is representative of the large population. Meaning we can be confident in generalizing what we find to the population. Also, the sampling method avoids researcher bias as it randomly selects participants from within each strata. But the researcher does decide which strata are important to consider. Meaning there might be some bias in the selection of strata and you can imagine stratified sampling is timeconuming and difficult. Designing research the experimental method. I'm going to cover two experimental designs first and they are fairly simple to understand. We can either get all participants to complete both conditions of the experiment and that's a repeated measures design or we can randomly split the sample into two groups and get one group to complete condition A and the other group condition B and that's an independent groups design. Independent groups design as the name suggests the groups are independent of each other. Different sets of people in each condition. All of the data collected from the group completing condition one is compared with all of the data from the group completing condition two. The data we collect is called unrelated data. Now I probably might be able to spot where this experimental design is because the groups are not the same people. Any change we measure may not be due to the change independent variable but due to participant variables characteristics of the people taking part in each condition. For example, if your study was on temperature and reaction times, if in the random allocation to groups, more young people were randomly assigned to one of the conditions, this could influence the results. Maybe showing a difference between conditions that isn't really there. Repeated measures design. So, an alternative is to get all the participants to complete both conditions. Easy to remember its name as each participant has to repeat the experiment under each condition. So, the researcher is comparing each participant scoring condition A with their scoring condition B. Because of this, the data the researchers collected is called related data. Each data point in one condition has a related data point in the other condition. That's going to be important to remember later on when we get to the statistical testing video. Well, that's great. Now, we're comparing a participant score to their own score in the other condition. There's no possibility that participant variables are responsible for measuring a difference between the conditions. But we now have a new problem. If the participants complete both conditions, there might be order effects. People may improve in the second condition because of practice or they might get worse in the second condition because they're bored or tired. We can try to control for order effects by counterbalancing. That's getting half the participants to do condition A then B and the other half B then A. This will then balance the influence of the order effects across the two conditions. So if tiredness or practice does affect performance, it should be the same across the two conditions. Notice counterbalancing controls order effects. It doesn't eliminate them. So looking at the evaluations for both designs, there are other issues. A problem for independent groups is this design needs more participants than repeated measures to produce the same amount of data. Another problem in repeated measures is because participants complete two conditions, they're more likely to work out the aims of the experiment and change their behavior to please the researcher. And this is known as demand characteristics. I think you've probably noticed that for these two experimental designs, the strengths of one are the weaknesses of the other. But there is another design, one that attempts to fix the problem of both participant variables in the independent groups design and order effects in the repeated measures design. Matched pairs design. In a match pair design, two separate groups of participants are used, one for each condition, but the researcher decides on a characteristic that might influence the result. So depending on the study, it might be IQ, age, or something like aggression levels. This is measured before the study begins. We can rank our participants. We would then randomly assign the highest scoring two participants to separate conditions of the experiment and then do the same for the next two and continue until all the participants are assigned. And then we have two groups that are balanced for that characteristic. The results for each pair in each condition are compared in the same way as repeated measures design. So we do treat the data again as related data. And as the participants are similar to each other, we've reduced the effect of participant variables. And as each participant only takes part in one condition, we've also removed order effects. But there are disadvantages. It takes longer to set up a match pairs design and we need more participants for the same amount of data compared to repeated measures and the participants are similar in each condition but they're not identical. So there might still be some participant variables. Designing research types of experiment laboratory experiment. The first type a laboratory or a lab experiment is probably familiar to you. A lab experiment could literally be set in a laboratory. But the important principle here is the experimentter has full control over what happens in the experiment. Environmental factors like noise, temperature, and even the instructions given to the participants are all highly controlled and they don't vary between conditions. We call these factors variables. In a lab experiment, the experimentter tries to make sure only one factor of the experiment called an independent variable changes between the conditions. All other variables are kept the same. This is so the experimental can measure how changing this independent variable changes the variable measured known as the dependent variable. For example, we might want to investigate how changing the color of light in the room but nothing else influences the recall of numbers. An advantage of using a lab experiment is by controlling for all of the variables, you can be fairly confident in suggesting a cause and effect relationship because you kept everything else constant. the variation you made to the independent variable caused the change measured in the dependent variable. This high control over the experiment means we would argue that lab experiments have high internality. What they measured is true. The observed effect is real and due to the change in the independent variable, not some other uncontrolled factor. Lab experiments are also highly replicable. You can give a list of instructions to another researcher. So a list of standardized procedures for the participants and variables to keep the same. Then if the researcher can conduct it in the same way and get similar results, your confidence in the validity of your findings increases. However, there are disadvantages. Lab experiments may lack external validity. The laboratory is not like the real world and behaviors observed in lab conditions may not generalize to environments outside of the lab like the home, school or work and this is known as lacking ecological validity. Also, the tasks used in an experiment are often unusual, not like the things we do in the real world. So, the behavior may not be the same as what we see in the real world. And this is known as the tasks lacking mundane realism. Also, being set in a lab, participants are well aware that they're in an experiment. They know the experimental wants to find something out, and they're likely to change their behavior to match what they think is expected from them. If this happens, you would say your study suffers from demand characteristics. Field experiment. Field experiments attempt to fix some of the weaknesses of lab studies. So instead of being set in a lab, they're conducted in the real world. For example, in a natural setting of a shopping center, places of work, or schools, places where people are used to behaving naturally. And of course, the strength of doing a field experiment is increased external validity. People will be expected to show more naturalistic behavior in their natural environment. So we can be confident in generalizing any behaviors measured to other situations, meaning higher ecological validity. Also, the task used are more likely to be real world tasks, meaning increased mundane realism. If the participants are unaware that they're taking part in a study, which is easy in a field experiment, demand characteristics are also not a problem. The weaknesses of field experiments are due to the lack of control that we did have in the lab experiment. The real world is chaotic and in a field experiment, we're not able to control every possible variable that might change your measurement of the dependent variable. These are known as extrrenuous variables and often in a field study, researchers are not able to randomly assign participants to each condition. This means that any effect observed may be due to a factor other than the change in the independent variable and this reduces the internal validity of the experiment. Natural experiment. In the previous two types of experiment, the researcher manipulates the independent variable themselves and then records the change in the dependent variable. However, in a natural experiment, the levels of independent variable have already happened naturally. The researcher just measures the change in the dependent variable. So, as an example, let's imagine you wanted to study what would happen to children's development if you stopped them from getting any love and affection early in life. Say you wanted to compare a group that had been deprived of love for less than 6 months, a group between 6 months and 2 years, and a group that had had no emotional care for over 2 years, and then you can put them with caring families. Well, of course, you can never actually conduct this as a lab experiment for ethical reasons. But this did happen to children in Romania in the 1980s and a researcher called Rutter followed these children up as they grew and recorded the effect of this deprivation on their development. Watch my video on the Romanian orphan studies if you want to know what R found. One strength of natural experiments is that it allows research into areas that could never be done otherwise either due to ethical reasons like rutter or cost. Also natural experiments are very high in external validity. These changes have happened naturally in real life and would have happened with or without the researcher. So any changes in behavior can't be the result of demand characteristics. However, the researcher has no control over the experiment like randomizing participants to groups or controlling for extenuous variables. There might be other factors that have influenced a dependent variable. Meaning we're not as sure of a cause and effect relationship between the IV and DV as we are in a lab study. Also, because these situations occur naturally and are often rare, these studies can't be replicated to see if we get the same results. Designing research, interviews, and questionnaires. Let's start by defining what self-report technique is. This means a research technique in which the participant knowingly responds to questions revealing personal information about themselves. This could be a real-time conversation with a research directly, usually face to face, but it could be over the phone or via text message. This is what we would call an interview. Another option is to have a list of pre-prepared questions that get sent out to the participants. The participants then fill out the questions and then send them back. And this, of course, is a questionnaire. Open and close questions. Both questionnaires and interviews can use open or closed questions. An open question is one in which the participant is able to answer in any way they want. In a questionnaire, there'll be a space to write the answer. Open questions give what's known as qualitative data. This means data in the form of words. A closed question is when the participant has a limited number of options. For example, just yes or no, a series of fixed choices or points on a scale known as a liyker scale. This gives a type of data known as quantitive data. Data in the form of numbers. When the researcher uses closed questions, because of the quantitive data, they're able to easily compare responses between participants and use data analysis. This is hard with open questions, but open questions give participants the freedom to respond how they want, so may lead to more valid, meaning more truthful answers. Of course, often questionnaires and interviews will have a combination of both open and closed questions. Designing questionnaires and interviews. When deciding on questions, the researcher needs to make sure they're clear. So, for example, they might try to avoid complex scientific terminology that the participants may not understand. In an interview, the interviewer can reword the question to make it more understandable, but that of course is not possible for questionnaires. The researcher also needs to be careful not to use biased questions, otherwise known as leading questions. These questions subtly suggest how the participants should respond, often in a way that will support the researcher's ideas. The researcher might consider piloting a questionnaire interview. What this means is running it with a few participants to check for problems. For example, some of the questions may not make sense and might need to be changed. This is particularly important for questionnaires as several thousand might be sent out. Filler questions can be used. These are questions that won't be used in the data analysis, but in an interview can put the participant at ease before more difficult questions are asked. In a questionnaire, filler questions can be used to hide the true aims of the study. Structured, unstructured, and semistructured interviews. So, when considering the interview, it's a back and forth series of questions in real time. The questions can be open or closed or a mix. And if it's face toface, it's likely to be recorded by the interviewer so they can listen to it again. But the researcher has to decide on a structured, unstructured, or semistructured style. Let's consider structured interviews first. In this case, the researcher will have a full list of questions that are asked in order. Advantages of this method are you don't need a fully trained interviewer if they're just asking questions from a list. Also, it's easier to compare interviews because all interviewees have had the same experience. However, because the questions are fixed, if the interviewe says something interesting, you can't develop on that point by asking a follow-up question. An unstructured interview is when the interviewer doesn't have a set list of questions that they're going to ask. It's a free flowing informal conversation with a general topic to discuss. An advantage of this approach is you're likely to develop rapport with the participants with this more informal style, meaning they feel more comfortable and maybe more likely to give you personal information. Using an unstructured interview means that the interviewe says something interesting, you can develop that point. But with this style of interview, you do need a highly trained interviewer to do it successfully. And as every interview is different, it's going to be really hard to compare multiple interviews. A semi-structured interview is in between an unstructured and structured interview. This is a mix of prepared questions, but with the ability to ask new ones. You still need a highly trained interviewer to think of the right questions to ask at the moment, but now you also have fixed questions that you've asked to every participant. Evaluating self-report techniques. Self-report techniques are a useful method for psychologists. They're easy to replicate using the same set of questions and the use of open and closed questions give data that can be analyzed with statistics and the opportunity for participants to give detailed information about their experience. However, they do suffer from bias, especially social desiraability bias. As people want to be seen in the best light, they'll often lie in their responses to look good to the researcher. If we need to compare interviews with questionnaires, we can say questionnaires don't require trained interviewer and could just be posted online. This means they're very cheap to give to large numbers of people, making them particularly good for data analysis when using closed questions. However, problematic questions can't be dealt with in the moment, and participants might not take them seriously compared to an interview. Quite often, participants will just put yes to every question. Maybe you've done that at some point. It's such a well-known effect, it actually has a name, acquiescence bias. One way of checking for acquiescence bias is by using a similar question later in the questionnaire, but phrasing it in the opposite way. If they take yes to both, you can see they're not answering the questionnaire honestly. So we can evaluate interviews by flipping the evaluations of questionnaires around. A strength of interviews is the interviewer can rephrase hard to understand questions and build rapport with participants so they take the research seriously. But negatives include needing a highly trained interviewer to conduct the interview resulting in smaller numbers of participants and a higher cost. An additional problem of interviews is interviewer effects. We're interested in finding out the participants honest views. We can get very different responses depending on the characteristics of who conducts the interview. Things like gender, ethnicity, and personality of the interviewer can all influence the responses. For example, think about the different responses teenagers might give in interviews about sex, drugs, or opinions on old people. If the interviewer was the same age and gender, or if they were much older and the opposite gender. Designing research case studies. In a case study, a researcher gathers a range of information on an individual, a group, or an organization. Interviews are often the main source of data collected, but the researcher can include observations of behavior and even experimental findings from psychological tests and even include content analysis on records like diaries. Now, what makes case studies special is the level of detail that's collected about the individual or the group. Case studies tend to be investigations of psychologically unusual individuals, but you can do case studies of events such as looking for the reasons for football violence by reviewing footage and interviewing criminals and police. Or you could do case studies on organizations such as the hiring policies of Google or the teaching outstanding schools. You may even do a case study on a group of typical members of a demographic like a group of 15year-old working-class boys. The type of data collected is usually qualitative information in the form of words because of the use of interviews. But when the researchers include experimental techniques, quantitive data, data in the form of numbers, can be used to back up qualitative findings. The duration of a case study can be a couple of hours, a few days, or years to decades. A short case study is called a snapshot, and a long case study is called a longitudinal study. Now, longitudinal studies are particularly interesting because you can observe changes over time, but as you can imagine, it's expensive to keep longitudinal studies funded over many, many years. Case studies have been used significantly in clinical psychology. Brain damage patients often have unusual symptoms and this gives an insight on the functioning of the brain. Now, for example, Paul Brocker researched a patient referred to as Tan. Now this patient could only say Tam and work while Tam was alive was combined with postmortem dissection of his brain after his death and this work led to the identification of Brock's area a part of the brain responsible for speech production. Freud used a number of case studies to develop psychoynamics and the most famous was little hands. This young boy had a phobia of horses that Freud identified as being symbolic of his fear as father. And Freud conducted this case study by sending a series of letters to his father and use this case study to support his theory of childhood development. Case studies of children of abnormal upbringings can be used to test the ideas of childhood development. And one famous example is a case of a child called Genie who was severely deprived of care from infancy until 13. Now even with the help of trained psychologists she was unable to develop beyond simple communication and struggle to behave appropriately. Genie demonstrated the importance of early years and critical periods of childhood and learning language and social skills. Evaluations of case studies. We can positively evaluate case studies easily. There just isn't another method that collects as much in-depth and rich information about individuals. This leads to a high level of realism that can be argued to be highly valid. Psychologists call this approach holistic, understanding a behavior or individual, not just from one perspective, but from a range. It's the preferred approach of a group of psychologists called the humanists. As I've said, case studies often look at the behavior of very rare individuals. This behavior and the participant situations can't be replicated in the lab. Genie situation, for example, can never be ethically replicated. And this means that case studies are often the only way to study certain behaviors. And another strength is just one unusual case study can show a pre-existing psychological theory is incorrect or maybe just not yet complete. But there are critical evaluations of case studies. Most of these critical evaluations question the scientific nature of case studies. For example, case studies are often completed long after events and depend heavily on memory. So what's recorded in interviews is often inaccurate. And you have the other problems of interviews like social desiraability bias being studies of often one unique individual. The findings from case studies cannot be generalized to wider populations. It might be a range of factors of that individual that cause a certain behavior rather than for example damage to a particular part of the brain. Now linked to this point is the exact replication of case studies is impossible. And while the amount of data collected is large, this presents another problem. Sometimes more data is collected than can actually be used and the researcher ultimately decides what to include in the report and might only include data that supports their ideas. Now this is one example of researcher bias. Another way case studies suffer from researcher bias is as the researchers work so closely with a case often over many years they may lose objectivity when interpreting behavior. And while case studies shouldn't be generalized and lack the scientific credibility of experimental methods, they can generate hypotheses that can be tested empirically and then ultimately accepted. Well over a 100red years after Paul Brocker and Tan's death, we now can use fMRI scans to confirm the existence of the region of the brain associated with speech production. Designing research observation studies. So we should probably start by defining an observation. This is researchers watching and recording behavior as it happens. As simple as this sounds, the researcher has choices to make about the type of observation they want to conduct. And this might depend on the research question that they're investigating. One choice they need to make is between a controlled and a naturalistic observation. A controlled observation is when we control the situation the participants experience and record their behaviors. This is done in a lab which helps to control as many variables as possible. Give the participants the same experience. Think of Mgrim, Bora, Ainsworth. While an advantage of this approach is we can reduce the effects of extrrenous variables on the participants behavior and we're able to repeat the observation and get reliable results. However, we can argue a big weakness is the environment itself is artificial and we may not see the same behavior repeated in the participants natural environment. Our other option is a naturalistic observation. The participants are observed in their normal environment and this has the advantage of high realism. the participants should behave as they normally would and we can claim that our findings have external validity in this case ecological validity. However, the lack of control means that there might be unknown extrrenous variables causing the behavior. Another choice is between overt or covert observation. In an overt observation, the participants can see you and critically they know they're being observed. This is important when you consider that one of the ethical guidelines is participants needing to agree to take part in research. they need to give their informed consent. But of course, the weakness with this is as soon as someone knows they're being observed, they're going to change their behavior. Maybe they want to look good in front of the observer or act in a way that they give the researcher results the participant thinks the researcher wants. This is demand characteristics. So, a covert observation solves that problem. The participant doesn't know that they're being observed and they can't see. The researcher is now observing natural behavior, giving their research more validity. But of course, a weakness is the research can now be argued to be unethical because the participant hasn't given their informed consent. And one more choice we should consider is if the researchers can conduct a participant or nonparticipant observation. In a participant observation, the researcher become involved in the group they're studying, maybe doing the same activities as the participants. The advantage of this is the researcher has firsthand knowledge of the participant situation and the researcher may build a rapport with participants, meaning they may open up more in what they tell and be more naturalistic in how they behave around the researcher. But in a participant observation, the researcher does run the risk of losing objectivity, becoming biased because they can only see the situation from the participant's point of view. A nonparticipant observation would involve the researcher standing back and recording the group without becoming part of it. This had the advantage of increasing objectivity but has the weakness of losing some important findings because they're too far removed from the experiences of the participants. So they are the types of observations easy to remember and each of the evaluations as you can see are just reflections of each other. Observational design. When designing our observation, we do need to consider exactly what behavior we're looking for and how we're going to record it. First, let's talk about operationalized behavioral categories. When I say something is operationalized, what I mean is I'm clearly defining a variable. This is so I can objectively measure it. If I say, for example, I'm observing aggressive behavior in children, that's a little vague and open to some interpretation. But if I say I'm recording every punch, push, and kick, it's very clear how I'm defining aggression. So a behavioral category is exactly that. I have a target behavior that I want to observe, in this case, aggression. And from that I'm going to create a list of behavioral categories that can be easily observed and countered. This can be turned into a frequency chart to help me record assessing observational reliability. Now after the data is collected, it's a good idea to show or assess its reliability. For this researchers would conduct a test of interrator reliability. This is simply using two researchers in the same observation. The researchers would be given the same list of operationalized behavior categories and conduct the observations separately. After the observation, the researchers would bring their data sets together and see if they're similar. They would do this by conducting a test of correlation such as a Spearman's row. And we're going to learn more about this particular test much later in the research method section. But for now, all you need to know is it's used to test the strength of a relationship between two sets of data. In this case, observation results. and most researchers would accept a correlation of 0.8 or stronger as showing the results are reliable. Correlation. Let's start by considering how different a correlational study is from an experimental study. Now, as a reminder, in an experiment, the researcher manipulates an independent variable and then measures how the variation in the dependent variable influences the measurement of the dependent variable. But in a correlation, the researcher doesn't manipulate any variables. They just measure. A correlation has two co-variables and these are simply variables that the researcher has measured and then compares. We can measure and then compare for example age, IQ, reaction time, bank account balance, number of pets, height, hostility level. When the data is being collected by the researcher, they can then display this data on a graph called a scatteragramgram. Let's use an example. A researcher might be interested in the relationship between the percentage score on a final class test against the student attendance over the year. So at the end of the year they can use the data collected for each student. Let me show you how to draw a scatterraph. Let's plot an X and a Y axis. When plotting a correlation because we're using co-variables in the scatterraph either variable can be on the X or the Y-axis. We'll pick our reasonable scale and then add labels to each axis and then give the graph a title. Now as we enter our data, each dot is a data pair for each measurement of one student. So let me take the data points for each student and then plot them from the table. Now what we see with this data is a positive correlation. This means that when one coariable increases the other coariable increases. So as you might have guessed a negative correlation is when one coariable increases and the other decreases. We can also have zero correlation and this is when there's no relationship between the coariables. Evaluation of the use of correlations. When considering evaluations, one of the most important critical evaluations is a phrase that you might have heard before. correlation does not show causation. And what this means is that even though there may be a relationship between two variables, we're not sure which variable is altering the other coariable or potentially is there a third unknown variable. Even in my example, you may assume the low attendance caused low exam performance. But an altered possibility is a students who struggled with the difficulty of the work avoided class more than the more able students. And with correlations, there can be a third factor that changes both co variables. One famous example is a positive correlation between ice cream sales and death by drowning. And it really looks from the set of data that the more ice cream is sold, the more kids drown. So what's going on here? Should we ban ice cream? Well, of course, it's a third variable, the temperature. As the weather gets warmer, more people buy ice cream, and more people go swimming at the beach. So linked to that point, just measuring two pre-existing coariables gives the researcher no control over potential extraneous variables. However, we can say a positive evaluation of using correlations in research is it can highlight potential causal relationships that researchers can then use to make predictions and investigate further with experimentation. Also, correlational research often has few ethical problems when researchers are often measuring pre-existing variables. Also, the correlation coefficient is a useful tool in describing the strength of correlation. Research procedures. These concepts are covered elsewhere in this video. So, I won't be including a research procedure section. A planning and conducting research. These concepts are also covered elsewhere in this video, so I won't be including a planning and conducting research section. Ethical considerations. It's fair to say that some psychological studies have a bit of a reputation for mistreating participants. In fact, they're some of the most well-known studies. They also tend to be older studies. These days, psychologists are more likely to follow a set of guidelines. In America, the American Psychological Association writes the ethical guidelines, and in the UK is the British Psychological Society, known as the BPS. In this video, I'll focus on the BPS's guidelines. If you want to read the latest version of the BPS's guidelines, you can find them here. It's actually only a few pages long and it's pretty easy to read. But it's important to keep in mind these ethical guidelines are just advice. Psychologists can and do bend and even break these rules. But serious mistreatment of participants would likely result in being expelled as a member of the BPS. Ethical issues. Informed consent. Participants should be made aware of the aims, purpose, and consequences of taking part in research and provide their informed consent before the study begins. If participants are not able to give consent, such as children or mentally incapable, consent can be given by a parent or guardian. Right to withdraw before taking part in the research, participants should be told they have the ability to end their participation in the study at any stage and this includes destroying any personal data collected on them like interview recordings. Protection from harm. The researcher needs to consider the study from the perspective of the participant and consider any risk to the participant's psychological well-being, physical health, personal values, and dignity. Confidentiality. Personal record should be kept securely. And when it comes to publishing results, not give away personally identifiable information, but in some situations, confidentiality needs to be broken. So if the researcher feels a participant or somebody else is in danger, debriefing. After the participant has completed their role in the study, the researcher should give them a debriefing. This is a conversation that tells the participants the reasons for the research, any outcomes, and the existence of other groups. This is also the point to check for any harm caused by taking part and offer assistance. Ethical issues. Psychologists can be torn between the participants ethical rights and wanting to gain valid data free from the demand characteristics or maybe investigating interesting but controversial, even harmful topics. Consider Mgrim. Yeah, the researcher who used an authority figure to pressure participants into giving what they thought were real electric shocks to learners. From the perspective of the guidelines above, he breached pretty much everyone. Informed consent, no. The participants were deceived. Right to withdraw, no. The participants were pressured to continue for as long as possible. Protection from harm, no. There's pretty good reason to think many participants suffered during and after the experiment. Confidentiality, no. The participants were recorded and the recordings were released. Some people even criticized Mgrim for his debriefing. Participants leaving not fully understanding the purpose of the study and having concerns about their ability to harm others. Dealing with ethical issues. So there are ethical considerations but also researchers need to produce valid data. By now you're aware of demand characteristics. The participant altering their behavior because they're aware of the research aim usually shaping the behavior in the belief it'll help out the researcher. There are three alternatives to informed consent that avoid giving away the aim and they are prior general consent. This is having a long list of things that could happen in an experiment and get a participant to agree to all of it and not be told which parts will actually be included in the study. Retroactive consent. This is where you'd get consent for the participants data to be used, but only after they've taken part in the research. Presumptive consent. You ask a group similar to the participants if they would agree to take part in a study. And if they agree, you assume the experimental group would consent. Now, none of these ways of getting consent are perfect, but they are ways of avoiding participants changing their behavior due to demand characteristics. There may be times that a research design requires deception or maybe even risking harming participants. In this case, a costbenefit analysis can be conducted. This is considering and comparing all potential costs to the participants with the potential benefits to wider society of the research. Considering Mgrim again, a number of participants did suffer emotional harm from taking part in the experiment. But Mgrim has been one of the most famous and influential studies ever conducted, taught to millions of students, and helps all of us think more deeply about what our limits would be if we were asked to do something wrong by an authority figure. From a costbenefit analysis perspective, you may judge Mgrim was worth the cost. An ethics committee is a group of people who consider if research should be carried out based on ethical principles and they may use a costbenefit analysis in their decision-making process. If you're at university to do a psychology degree, the decision on if you can collect your data for your final project will be decided by the university's ethics committee. The debriefing can also be a time to deal with any ethical issues such as revealing to the participants if they've been deceived, revealing the existence of any other groups, reminding them they still have the right to withdraw their data, checking for harm, and offering support if they have been harmed. This happens at the end of a study and sometimes in the exam, you can be asked to design a debriefing form or an informed consent form. Quantitive and qualitative data. Firstly, quantitive and qualitative. Nice and easy to remember which way around this goes as quantive data is data in the form of numbers and of course the word quantity refers to the number or amount of something. So when I say data in the form of numbers I mean data taken by the recording of variables. So I could gather reaction times an emotional rating on a like scale or record the number of times someone performs an action in an observation. This data can then be described with descriptive statistics. So averages and range but also summarized in tables and graphs. Qualitive data on the other hand is data in the form of words meaning descriptions of behavior, thoughts and feelings. Qualitative data can of course be turned into quantitive data. So recordings of observations and interviews can be coded and behavioral categories tallied. And as discussed in previous video, content analysis is another way to turn large amounts of qualitive data into quantitive data. Quantive data is often collected in experiments and closed questionnaires while quantive data is suited to interviews, observations, open questionnaires, and case studies. But of course, many studies will collect a combination of both quantitive and qualitative data. The strengths and weaknesses of using quantitive or qualitative data tend to be opposites of each other. So while one of quantive data's advantages is it's seen as objectively measured and with less chance of bias, so more scientific, qualitive data is more subjective, open to interpretation by the researcher and potentially biased. Quantitive data is easy to summarize using graphs and tables to show patterns in data while qualitive data is more difficult to summarize. Quantive data techniques tend to be more reliable. If the stud is replicated, we tend to get similar or consistent results. This is compared to replicating qualitative studies where the data tends to be more variable meaning less reliable findings. But a weakness of quantive data is it seen as lacking depth and detail focusing only on individual features of behavior and only on what can be measured. So the advantage of qualitative data is it seen as richer with far more detail and could be argued to be a truer reflection. So a more valid measurement of human experience. A really good example showing the difference in the information gained between quantitive and qualitative data comes from Mgrim. Now you're probably aware that the researcher in Milham study uses authority to convince participants to give stronger and stronger electric shocks to another person using a machine to the point of potentially killing that person. And you may remember Mgrim's quantitive findings. Just seeing these results with all the participants going to 300 volts and 65% going all the way to 450 volts, you probably think that these participants are unfeilling monsters and there's no way that you would deliver these shocks. But if we look at some qualitative data, for example, the observational recordings of the original participants, we can see how emotionally difficult it was for the participants to give the sharks. This qualitative data provides context. It's not that these people were happy to give such strong shocks or unfeilling, but the power and authority figure is so strong that it will make people do terrible acts against their better judgment, even when they find what they've been asked to do very emotionally difficult. Primary and secondary data. Primary data is collected when the researchers themselves are responsible for generating the data. We can say this is firstirhand or original data. Common ways to collect primary data are that the researcher conducts experiments, observations, interviews, questionnaires, and case studies. Also, the reason data is collected is to answer a particular research question. The data is focused on the demands of the hypothesis. Secondary data is when the researcher uses data to answer their research question that's already been collected and published, often to answer a very different research question. So secondary data can be government or business statistics and records or previously published data from other studies. When it comes to evaluations, of course, a downside to primary data is going out into the real world and collecting all of your data means a large investment in cost and research time. And when it comes to secondary data, if that data is already collected, freely accessible and ready to analyze, well, that's a cost saving. One advantage that primary data has over secondary data is it's likely to be more valid because its collection is shaped to the demands of the research question. Meaning you designed a study to collect data that explicitly answers that question. Researchers using secondary data have to take whatever data is available even if it was collected for very different reasons. Connected to this point, if researchers collect their own primary data, they can control their study to avoid extraneous variables. If however they collect secondary data, they have to trust that the original researchers collected valid data. Computation. This video does not include the full computation section which deals with relatively straightforward mathematical principles. I have however included in the next section how to calculate the mean, how to calculate percentages and some examples of how to convert between fractions, decimals and percentages. descriptive statistics to understand the difference with measures of central tendency. Let's imagine you're going for a job interview at a small company and they ask if you have any questions and reasonably you ask what the average wage of the people who work at that company is the CEO is interviewing you and here's a list of all the salaries of the company employees. Let's say it's this set of data. 10,000 15,000 15,000 15,000 20,000 24,000 24,000 26,000 36,000 and 195,000. The mode. Let's start calculating measures of central tendency with the simplest, the mode. The mode is the most frequent score in a quantitive data set. So you can see in this data set, the mode is 15,000. There can in some sets of data be more than one mode. If there are two modes, we'll call that data biodal. And if there are more than two, multimodal positive, the mode or isn't affected by outliers, meaning it's particularly useful for discrete numbers. For example, it can make more sense to say the average family has two children than 1.89 children. And if the data is in categories like average pet choice, giving the modal group is the only way of giving an average. However, a criticism of the mode is in small data sets, there are likely to be multiple modes or even no mode if every score is different. So, not given a clear average value. It also doesn't include all the values in its calculation. So, it's not as sensitive as the mean. Median. The median value is calculated by ordering the values from the lowest to the highest and using the number in the middle as the average. If there are an even number of data points, then take the two middle scores and it's halfway between the two. So in this case, the median value would be 22,000. Positives of the median are it's not affected by extreme outline scores. Unlike the mode, it's very easy to calculate. However, criticism of the median are not all the raw scores are used in its calculation. So we'd say that it's not as sensitive as the mean. The mean. So to work out the mean we add all the scores together which gives us 380,000 and divide by the number of scores in this case 10. So in this data set our mean average salary at this company will be £38,000. Positives of the mean are we have used and represented all of the raw data in this calculation. So we would say it's sensitive. However the main criticism of the mean is because of its sensitivity it's easily shifted towards extreme values. Now, going back to our scenario, the CEO would be perfectly accurate in giving any of those figures as the average wage of the company's employees. I might be tempted to state the mean to show his company in the best light and to get you to join his company. And you've probably noticed this careful selection of descriptive statistics is often used by politicians and companies to spin and misrepresent their policies. So because of their contrasting strengths and weaknesses, the correct measure of central tendency should be used for the right situations. Measures of dispersion, the range. As measures of dispersion are about the spread of a set of data, the range may be the first you think of. It's easy to calculate. You just take the smallest value from the largest to get the range. Using our example, 195,000US 10,000 gives a range of 185,000. Some of you may have been told to add one when working out the range. That's fine. It's also an acceptable way of stating the range. Well, a positive is the range is clearly very easy to calculate. However, the main criticism of the range is extreme scores easily distort its value and the range doesn't show if the scores are more spread out or are clustered around the mean. Computation percentages. Percentages are also a way of describing data and we can be asked in the exam how to calculate percentages. You may already know how to calculate percentages and in that case keep in mind that when you see a percentage question in your exam, you may need to explain what a percentage means rather than just stating what the percentage is. But you can skip this next bit if you know how to work out a percentage. Grab your calculator. You're allowed one in your A level psychology exam. Do not forget it. What follows is how I calculate percentages. There are a few ways and you may be shown a different way that makes more sense to you and that is fine. Firstly, let's cover how to calculate the percentage of any number that we're given. In this example, the psychologist is conducting a stratified sample for a questionnaire. Her total population is all of the students at a large further education college. She knows the full population of the college is 2,480 students. 595 students are studying for A levels. 1,389 students are studying for BTech qualifications. 288 students are only attending to retake GCSE and 28 students are adults taking access to university courses. What is the percentage of each subgroup and how many of each group should she include in a sample of 250 students? You may know that percentages can also be expressed as a fraction or a decimal. In this case, we have both parts of the fraction that we need because for our percentage, we want to know 595 out of 2480. So let's write our fraction as 595 over 2480. We can convert this fraction into a decimal by dividing. And here is my calculation. Now we have the fraction as a decimal. To turn this decimal into percentage, we times by 100. Or more simply, we just move the decimal place over to the right twice. I'm going to round this up and we get to 24%. Pause this video and try the same calculations with the other groups. Here are the percentages. Secondly, let's see what we need to do if you need to find the percent of a number. So, you now know that the sample in the study is 250 and we need to draw that 24%, 56%, 12%, and 8% of this number. Well, the easiest way to work it out is to turn the percentage into a decimal and then times the decimal by the number of participants. So, we can work out 24% of 250 by multiplying 0.24 24 by 250 giving 60 participants who are in our A level strata and here are the answers. I may want to talk about the percentage difference between two numbers. Let's say next year the questionnaire is repeat at the college and the number of Alevel students at the college grew to 724. What's the percentage change in the number of A-level students from one year to the next? Well, to work out the percentage change between two numbers, subtract the old number from the new number. So 724 minus 595 and then divide this by the old number. Then times this by 100. That gives us an increase of 21.7%. Let's say the number of B tech students reduced from 1,389 to 974. Using the same calculation, we get minus 29.9. Of course, the minus means there's a decrease in the number of B tech students. Okay, I hope that helped with calculating percentages, interpretation and display of quantitive data table. This is a raw data table. When first collected before summarizing, researchers will gather quantitive results together into a raw data table. Researchers, especially when conducting observations, may also collect their raw data in the form of a frequency table. A frequency table is also known as a tally chart. It generally has three columns. The first with behaval category or value of a variable, the second column for a tally mark for each observation or record, and the final column for a total of the tally. But even without the total column, reading the frequency table is as easy as counting the tally marks. Bar charts. A bar chart is used to summarize the frequency of categorical data which is also known as nomal data. This means data in distinct categories like favorite pet. The categorical variable is usually placed on the x or horizontal side to side axis with the frequency or value on the y-axis, the vertical up and down axis. The height of the bar showing the value or frequency of that category. A very important point to make about bar charts is as each bar is a distinct category and could be arranged in any order, the bars do not touch. If they touch, this suggests continuous data and that's a histogram, not a bar chart. So be very careful when you draw a bar chart as it's an easy mistake to make. Now it might be that a bar chart shows more than one variable at the same time. If this is the case, then the chart will have a legend or will be appropriately labeled. Scattergrams. Scattergrams are used to show the relationship between two coariables in correlational research. As a coariables, it doesn't matter which way around each variable goes on the X and Y axis. Each point on the scatteragramgram represents two measurements of one participant. Common correlational relationships are positive and negative. Check out my video on correlations for more on using scatteragramgrams to show correlations. Histograms. Histograms show frequency and at first glance appear to be similar to bar charts, but you can see the bars are touching. This is to show that the scale on the x-axis is continuous. Common examples are test scores, age group, and time scales like months and years. In this example, each a group follows on from the next. And we can see the bars have to be displayed in order to make sense of the data. Normal distributions. So after collecting data about the frequency of some continuous factor, it can be displayed on a graph that shows its frequency. This is known as a histogram with the up and down y-axis representing frequency and the side to side x-axis showing the value of the scores. Sometimes the spread of scores can show no pattern, but sometimes the spread can be more scores on one side or the other. Sets of data that show this distinctive bell curve are what's known as a normal distribution. And you can tell it's normal distribution because the two sides of the curve are symmetrical, the same on both sides. And the mean, the median, and the mode are all at the top in the center of the curve. The sides of the curve don't touch the x-axis as there always theoretically could be extremely small or large scores. The mode will always be at the top of the curve as the mode is the most common or frequent score. The median is the middle score. So as the curve is symmetrical, you'll have 50% of the scores on each side of the highest point with the same amount of area under each side. And the extreme scores of each side balance out keeping the mean in the center. Now let's consider the normal distribution in terms of IQ again. So the most frequent score is 100 and as each side of the curve is symmetrical, someone's as likely to have a score of 99 as 101 or as likely to score 85 as 115. Most of the scores are close to the center. I want to thank all my patrons for their support. They allow me to work part-time and make sight boost and in return they get extra resources. For example, GCSE patrons get digital worksheets, exam question walkthroughs, and these videos as a set of teaching slides.