Chapter 2
Methods of Psychology

Outline Lessons from Clever Hans
Types of Research Strategies
Statistical Methods in Psychology
Minimizing Bias in Psychological Research
Ethical Issues in Psychological Research

Lessons from Clever Hans

Mr. von Osten and Clever Hans

The Lessons Learned

Facts - objective statements based on direct observation that everyone can agree are true. It is a fact that Hans was responding in a way that gave the appearance of human-like intelligence.

Theories - a mental model designed to explain facts. Theories also make predictions about new facts that might be discovered.

Hypotheses - Any predictions about new facts that might be discovered are hypotheses. A hypothesis is an educated guess.

Skepticism - Scientists should always be skeptical about any claim and should be willing to make attempts to disprove any theory, including their own. The best explanations for phenomena are often the most parsimonious, or the most simple. Accepting that Hans could do what von Osten claimed would require re-thinking everything we know about animal intelligence and behavior. A skeptic would first consider more parsimonious explanations before accepting von Osten's hypothesis.

(A good example of the need for skepticism in science is the whole issue of "repressed memories." This issue is controversial. Read about a leading memory scientist's skeptical views of this phenomenon here.)

Observations under controlled conditions - by conducting experiments in controlled conditions scientists can be more confident about their conclusions. Creating an experiment is one way to ensure that your observations are made under controlled conditions. In the case of Clever Hans, the conditions were not controlled, so although it appeared that Hans was making intelligent answers, in fact he was receiving information about how to respond correctly from Mr. von Osten and the audience.

Observer-expectancy effects - when observations are not made in controlled conditions, the observers can sometimes affect the phenomena under observation. People observing Hans expected him to get the answers correct and (knowingly or unknowingly) communicated to him how to answer questions correctly. Experiments attempt to control for observer-expectancy effects.

LINK: The Skeptical Inquirer, the magazine for science and reason.

Types of Research Strategies

Experiments

Experiments are controlled investigations that study cause-and-effect relationships through the manipulation of variables. They allow for random assignment.

An experimenter manipulates V1 (e.g., amount of grain alcohol consumed by a person) and then measures V2 (e.g., amount of aggression displayed).

V1 is your independent variable

The experimenter (E) has total control over the independent variable (IV). That is, the E decides "how much" of the IV the subject (S) gets or the condition to which the S is assigned.

For example, one way to manipulate the IV would be:

1 oz. consumed
2 oz. consumed
4 oz. consumed
8 oz. consumed
12 oz. consumed
48 oz. consumed

V2 is your dependent variable

The dependent variable is what E measures in the experiment. You're interested in how scores on the dependent variable vary as a function of exposure to different levels of the IV.

So, the amount of aggression is our dependent variable.

Of course, we need to make sure we have a valid measure of amount of aggression. Let's assume we do. Here are our results.

Correlational Studies

Correlation = Relation between two or more measured variables. We can't always manipulate variables or hold them under control. Sometimes, we can only measure them.

Ex.: we can't force some people to go to church and others not to -- that would be unethical.

Ex.: we can't force some people to read books and others not to -- that would be unethical.

So, all we can do is measure church-going and measure book-reading to see whether the two measures are related. How do we measure relationships in the correlational method? We compute correlation coefficients using statistical techniques.

Correlation coefficients can range from -1.00 through 0 to +1.00

negative correlation (= -.01 to -1.00)

uncorrelated data - zero correlation (= 0)

positive correlation (= +.01 to +1.00)

How do we interpret relationships found using the correlational method?

The higher the correlation coefficient-- be it positive or negative -- the stronger the relationship between two variables. The closer the correlation coefficient is to ZERO -- be it positive or negative-- the less there is a relationship between the two variables

The Cardinal Rule of Correlation (repeat after me): Correlation does not imply Causation

The Cardinal Rule of Correlation Put Differently:

We can only describe a relationship, without making claims that:

Church-going CAUSES book-reading

Book-reading CAUSES church-going

Practicing Correlation

Read about the following relationships and determine if the relationship is positive or negative? Then give two explanations for each relationship.

A study of married couples showed that the longer they had been married, the more similar their opinions were on social and political issues.

In a study of American cities, a relationship was found between the number of violent crimes and the number of store selling violence-depicting pornography.

A college professor finds that the more class absences a student has, the lower their grade in the course tends to be.

All children in an orphanage received an IQ test. Results showed that the longer the children had lived in the orphanage, the lower their IQ scores.

People who are left-handed tend to die, on average, seven years earlier than people who are right-handed.

The incidence of violence (and lunacy) is greater when there is a full moon.

Students who attend class get higher grades in the course than those who attend less regularly.

As scores on a measure of depression increase, so do scores on a measure of how shame-prone they are.

The more someone watches T.V. shows with positive, prosocial content (e.g., Mr. Rogers' Neighborhood), the more helpful and empathetic they tend to be in real-life.

The more someone plays violent video games (e.g., Doom), the more aggressive they are in real-life.

The more churches there are in a community, the more bars there are.

The greater the number of medical doctors available, the greater the death rate.

The more churches there are in a community, the less the violence rate.

Advantages and Disadvantages of Correlations

Advantages

May be more convenient to implement in some situations

May be permissible where ethical considerations prohibit random assignment of subjects;

May be conducted in a more naturalistic setting;

May give you access to a larger subject pool, since data are easier to collect than in an experiment

Biggest disadvantage:

Can't infer _ _ _ _ _ _ _ _?

Can't infer _c a u s a l i t y_!

Descriptive Studies

Descriptive studies simply describe individuals without manipulating variables or systematically investigating their relationships. Examples:

Goodall's research with chimpanzees in the wild.

Observing people behave at sporting events

Counting how often people use wastebaskets in parks

Recording pupils' interactions on playground

Research Settings

Laboratory studies - this is when a researcher sets up a controlled environment in which to conduct an experiment. Laboratory studies can be very effective at making sure that observations are made under controlled conditions.

Example: Milgram's studies of obedience. Click here.

Field studies - sometimes it is impossible or impractical to conduct research in a laboratory. For example, what if a researcher was interested in people's behaviors while in church? She could created a laboratory environment that replicates the actual church environment, but it is unlikely that people would behave in this environment exactly the same way as they would in a real church. Therefore the researcher might decide to conduct the research in the "field", which might mean going into actual churches (with permission and consent) and observing people's behavior there. This technique is called "naturalistic observation" (see below).

Other examples:

You could introduce "models" into the park. Some models throw trash away in wastebasket. Others throw trash on ground. You count how often the "subjects" use the wastebasket as a function of the model to whom they were exposed.

Sherif & the famous Robber's Cave study

Lewin's leadership studies.

Data Collection Methods

Self-report - many phenomena studied in psychology can only be investigated by asking people how they think or feel. This can be done through methods such as questionnaires, interviews and/or naturalistic observation.

Questionnaires - this is where people produce self-descriptions by checking off items on a list or writing answers to essay questions.

Interviews - in the interview, people describe their thoughts, feeling and behaviors orally to an investigator. Interviews can either be structured, where only specific questions are asked, or unstructured, where the investigator asks questions as they occur to him or her.

Naturalistic observation - sometimes the only way to find out about people's behavior is to watch them do it in the actual environment. The example above, about investigating behavior in church by actually going there and observing people is an example of naturalistic observation.

Statistical Methods in Psychology

Descriptive Statistics

Descriptive statistics are simply ways of describe a set of numbers, also called a data set. Three of the most commonly used descriptive statistics are the mean, median, and mode.

Mean - the mean is simply the arithmetic average of a set of numbers.

Median - The median is the "middle" number in a set of scores. Take a set of numbers, arrange them from lowest to highest, and the middle number is the median (a good way to remember this is to think of the median that is the middle of a highway.) If there is an odd number of data, simply pick the middle number. If there is an even number of data, the midpoint between the two middle numbers is the median.

Mode - the mode is simply the number that occurs the most frequently in a data set.

It is also useful to describe the degree to which the numbers in a data set differ from one another or from the mean. This is called variability and the most common measure of variability is the standard deviation.

Standard deviation - this is simply the average difference between each individual score and the mean. Said another way, the standard deviation is how much, on average, the scores in a data set differ from the mean.

Inferential Statistics

Whereas descriptive statistics simply describe a data set, inferential statistics attempt to make inferences about a larger population based on a data set. For example, if you're interested in studying USU student behavior, you would have a hard time collecting data from each and every student. Instead, you collect data from a sample of students, representative of the entire USU student body, and then make inferences about the student body based on the data collected from your sample.

Any time you collect data, it will contain variability due to chance. For example, by chance alone, you might collect data from more USU freshman than from USU sophomores. If you repeated your data collection several times, you would get somewhat different results each time due to this chance variability.

If this chance variability always exists in data collection, how can a researcher be confident that the inferences he or she makes about the larger population (the entire USU student body) is accurate? We use inferential statistics! Instead of making absolute conclusions about the population, researchers make statements about the population using the laws of probability and statistical significance.

Statistical Significance

In research, the inference being assessed is called the research hypothesis. In an experiment, this hypothesis usually states that the independent variable has some consistent effect on the dependent variable. In correlational research, this hypothesis is a statement about the nature of the correlation between two variables.

Inferential statistics are procedures for calculating the probability that the data could have come out like they did IF THE RESEARCH HYPOTHESIS IS WRONG. Said another way, inferential statistics tell us how likely it is that we would find the results we found in our research due to chance alone. Researchers want to be very careful (remember skepticism?) so this probability has to be very small before they confident they have supported their research hypothesis.

The components of a test of statistical significance are (a) size of the observed effect, (b) number of individual observations and (c) variability.

Statistical significance does not necessarily imply practical significance.

For example: A researcher finds statistically significant results that an educational program increases students' math performance by a half a percentage point. That is, after participating in the program, a student's math scores are raised by .05%. The cost of this program is $5000 per student. Although the researcher may have demonstrated that the .05% rise in math scores is unlikely to be due to chance, this finding is of little practical significance considering the expense of the program and the minor raise in performance.

median = 5 mode = 5

Minimizing Bias in Psychological Research

Error and Bias

Error refers to random variability in results.

Bias refers to nonrandom effects caused by some factors unrelated to the research hypothesis.

Avoiding a biased sample

When assigning participants to groups in an experiment, it is critical to ensure that the groups do not differ systematically before the experiment. If they do, then your results can be biased.

Let's look at an example: A researcher wants to determine how effective an educational program is at raising elementary school childrens' math scores. Because it would be difficult to pull children out of their classes, the experimenter decides to use Mrs. Jones' class as the experimental group, which receives the educational program. Mr. Smith's class is used as the control group, which does not receive the educational program. Why are the results of this experiment likely to be biased?

Because the experimental and control groups differ in a systematic way OTHER than the educational program. Maybe Mrs. Jones is a better math teacher than Mr. Smith, making all the children in her class score higher on math tests. The only systematic difference between the groups in an experiment should be the independent variable, which in this case is the educational program.

Avoiding measurement bias

To avoid measurement bias, researchers try to make sure their measurements are reliable and valid.

A reliable measure is a consistent measure.

A valid measure is one that actually measures what you intend it to measure. There are several different categories of validity, of which we will discuss two.

Some psychological measures lack face validity. Face validity refers to whether or not a measurements looks like it measures what it is supposed to measure. You may encounter a question on a test of schizophrenia that apparently has absolutely nothing to do with schizophrenia. In fact, this question may be quite good at contributing to a determination of whether or not you have this mental illness.

Criterion validity refers to how strongly the scores of a measurement correlate with another "valid" index of what we are measuring. For example, we have one measure of aggression (e.g., how many unprovoked fights children are observed to have on the school playground) & this measure is already known to be valid. We want to know whether another, new measure validly assesses aggression (e.g., how often children are nominated by peers to be aggressive). For the nomination measure to possess criterion validity, it must correlate strongly with observations of playground fights.

Observer-expectancy effects

Remember Mr. von Osten and his not-so-Clever Hans? Sometimes a researcher who desires a particular result can unintentionally communicate the desired result to research participants, biasing the results. Oscar Pfungst suggested that Hans was observing subtle cues from von Osten and the observing crowd to figure out when to stop counting. This is an observer-expectancy effect.

So how can we avoid such biases if they are unintentional? We can do this by keeping the observer blind. This means that the observer does not know about the research hypothesis and therefore has no expectations to communicate.

Subject expectancy effects

Suppose you went to the doctor with a stress-induced headache. Concerned about giving you unnecessary drugs, the doctor instead gives you a sugar pill but tells you that the pill is a strong pain reliever. An hour later, your headache is gone. Since you had the expectation that your headache would be gone it went away, even without the benefit of medicine. This is called a subject-expectancy effect and the sugar pill is called a placebo.

In experiments, subject expectancy effects can be controlled by using double-blind procedures. This simply means that both the person administering the procedure (e.g., the experimenter) and the participant/subject in the procedure (e.g., the patient) are kept unaware of (blind to) the condition to which the participant is being assigned. Example: The alcohol-aggression study would be "double blind" if neither the experimenter nor the participant knew exactly how much alcohol the participants had been given. In a drug treatment effectiveness study, neither the experimenter/physician nor the subject/patient would know whether the subject had received the placebo or the real drug.

educes the chance of finding statistically significant results, bias leads to false conclusions.

( Cartoons by Mark Parisi. Used by special permission. For many more, visit his site.)

Ethical Issues In Psychological Research

Research with humans

Psychological science depends largely upon willing human participants Without them, how would we ever learn anything? Therefore, respecting the rights, safety and privacy of research participants is of the utmost importance. Three are three primary issues to consider when conducting research with humans:

Rights to privacy - the privacy rights of participants are maintained by obtaining informed consent, allowing participants to quit at any time without penalty and keeping records and data confidential and secure

The use of deception - is it ethical to deceive people in psychological research? How can we avoid expectancy effects if participants are completely informed about the research and its hypotheses? Some researchers view all deception as unethical, yet others believe that it must be done in some cases, but done carefully and respectfully.

Possible discomfort or harm - participants in research must always be informed of any possible physical or psychological harm that might result from their participation in the research.

Research with animals

Most people agree that procedures that cannot ethically be done to humans can be done with animals. Others believe that animals should have the exact same rights as humans and, since they are unable to give consent, should never be used in psychological research. It is a fact that humans have benefited greatly from research conducted with animals. Fortunately there is some middle ground - the American Psychological Association has established a set of principles guiding the use of animals in research to protect them as much as is possible.

LINK: The American Psychological Association's code of ethics.