A.To get facts rather than opinions.
B. To get good information so we help--rather than hurt--people. Wouldn't it be nice if everyone in the health field used the same approach? Unfortunately, they don't.
A. Description/Prediction: "What?"
B. Explanation/Understanding/Control: "Why?"
|What are people doing or thinking?||What causes (influences, makes) people to do or think what they do?|
|What is happening?||Why is it happening?|
|What do people do in a certain situation?||How can manipulating drugs, situations, or instructions change people's behavior?|
|Answered only by non-experimental/ descriptive/correlational research.||Answered only by using experiments.|
Practice on this key distinction:
Our society's emphasis on causal explanations: We like experimental methods because we want to know why things happen and how to change things. However, nonexperimental methods also tell us important things--what people do and think.
1. Can't do an experiment because you can't manipulate the predictor variable. Ex: You can't manipulate participants' gender or age.
2. Can't ethically do an experiment because you can't ethically manipulate the predictor variable (e.g., illness, poverty)
3. Want to describe or predict behavior
A. Observation:C. Case study
Useful for describing behavior and for suggesting causal hypotheses that could be tested in experiments.1. Scientific observation should beB. Surveys and Testsa. Objective: "just the facts"2. Difficulties with observation--It can sometimes be:
take special care to avoid anthropomorphism (also known as the anthropomorphic fallacy): giving animals human characteristics (warmth, genius, etc.) without objective evidence.
b. Systematically record data to avoid memory biases and errors, such as those caused by the availability heuristic)
c. Use good, fair sample if you are going to generalize your results.a. Impossible: Can we observe thoughts?3. Uses
b. Impractical: Do you want to wait to observe a robbery?
c. Unethical: How does observation differ from spying?
* What people say, not what they do for four reasons
- Social desirability bias: People may try to make themselves look better than they are. In other words, they lie.
- Obeying demand characteristics: People may say what they think we want to hear.
- Memory errors: Do people accurately remember what they do?
- As revealed by Nisbett and Wilson's findings, people don't always know why they do things.
* Only as good as questions.Two types of bad questions:
- Leading questions, which "lead" respondents to the "correct" response. These questions make it quite clear to respondents what answer the researcher wants them to give.
Ex: "You like my website, right?"
Note that some leading questions may lead to increasing the social desirability bias.
- Confusing questions: Questions that are too long, use too many " and's," "or's," "but's,""no's," "not's," or big words.
* Only as good as sample.Two factors help you have a good sample.
- Large sample. Having a large sample is a good start, but, by itself, is not enough to guarantee a good sample. A big sample can be a bad sample.
- Random sampling: a system where the sample should be similar to the larger group because every member of the larger group has an equal chance of being chosen to be in the study.
1. Learn from unique or extremea. Unique2. Problems in drawing conclusions from case studies
- Using a small, nonrandom sample means you can't generalize the results to other individuals
- Lack of control group means you can't make cause effect statements because things might have turned out the way they did even without the treatment.
A. They do tell you whether 2 variables are related. But they do not tell you which variable influences which. They may hint or suggest that one variable influences another, but they are never proof of causality. That is, they are never proof that changes in variable A cause changes in variable B. A humorous example.
That is, if variable A and variable B are correlated, you can't know which variable influences which. Why not?
Because the variables could be statistically related for any one of the following 3 reasons--and you have no way to know which of these reasons is the correct one:1. A causes (influences, affects) B
2. B causes (influences, affects,changes) A[thus, if you concluded that A-->B, you might
be confusing effects for causes]3. C causes both A and B.B. The language of correlations. Correlation coefficientsThat is, some other factor influences both A and B, but there is no direct relationship between A and B. In other words, A doesn't influence B, B doesn't influence A, but some other factor ("C") influences both of them. [Thus, if you concluded that A-->B or that B-->A, you might be ignoring the fact that you are really looking at two effects of some other cause].
can range from -1.00 to +1.00. The correlation coefficient contains two pieces of information:
- One piece is the sign (positive or negative),
- the other piece is the number itself.1. The sign of the correlation indicates the kind or type of relationship (but not the strength of the relationship)a. Positive correlations
the more ___ (fill in the blank with a variable, e.g. height), the more ____ (fill in the blank with another variable, e.g., weight)
the less ____(fill in the blank with a variable, e.g. height), the less ____(fill in the blank with a different variable, e.g., weight)b. Zero correlations: no relationshipc. Negative correlations: reversethe more ____ (fill in the blank with a variable, e.g., stress), the less _____ (fill in the blank with another variable, e.g., happiness)
the less _____(fill in the blank with a variable, e.g., stress), the more _____(fill in the blank with another variable, e.g., happiness)
2. The further away from zero, the stronger the relationship.
Practice estimating correlations.
- Counter-intuitive implications for comparing positive and negative correlations
- -.9 is a stronger correlation than +.7
- -.2 is a stronger correlation than 0
If you are at a Macintosh computer and have some time, you can download a computer program that will help you understand what correlation coefficients mean.
Definition of the simple experiment: a research tool that allows scientists to find out whether a treatment influences (causes) a given behavior or mental characteristic by randomly assigning some participants to get the treatment and other participants to not receive the treatment.
Brief overview of the Simple Experiment
|No treatment group||Average Score|
|Treatment group||Average Score|
|No caffeine group||15 seconds|
|Caffeine group||9 seconds|
Why do we do experiments?
To make causal statements.
Two types of hypotheses:
1. Experimental hypothesis (H1): prediction that the treatment causes an effect. It can be proven wrong.
2. Null hypothesis (Ho): prediction that the treatment does not cause an effect. It also can be proven wrong. It cannot, however, be accepted.
Joke illustrating problem of accepting the null hypothesis.What happens if we disprove the null hypothesis?
(Hint: If the statement "I did not eat the candy bar" is false, what did I do?)
Why do we have two groups?
(Hint: In the Skinner experiment with the rats running the mazes, what could we have concluded if we had only used a caffeine group. That is, what could we have concluded about the effects of caffeine if all we knew was that the rats getting caffeine ran the maze in 9 seconds?)
How can we avoid comparing apples with oranges?
(How do we know that our treatment group and control group were similar before we introduced the treatment?)
Random assignment to treatment involves using a system where
everyone who participates in the study has
an equal chance of being put into the treatment group.
What's the problem with random assignment?
Hint: Suppose we get these results:
Experimental Group = 75%
Control Group = 74%
Could these results be due to random assignment creating groups that were slightly different before we introduced the treatment?
How can this problem be solved?
Tests of statistical significance determine if the difference is to big to be due to chance alone
The tests look at two factors:
1. They look at the size of the difference.The bigger the difference between the groups, the more likely the results are to be statistically significant. For example, if the Experimental group averages 95% and the control group averages 45% on our test, that difference would probably be statistically significant. (Intuitively, you do the same thing. If your team gets beat by one point, you point out that the other team was lucky. You don't have to concede that the other team is better. However, if they beat your team by 30 points, you may have to admit that the other team is better).
2. They look at the number of participants.The more participants that are used, the more likely the results are to be statistically significant. (Why? Because if you only have a few participants, the groups might be very different at the beginning of the study. However, if you have 100 participants in each group, the groups should be pretty similar before the start of the study. If they are very similar at the start, then, if they are even slightly different at the end, that difference could be due to the treatment. Similarly, in sports, if one team beats another in a seven game series that's more convincing evidence of the team's superiority than winning a single game.)
Two possible verdicts from statistical tests1. statistically significant:
you are sure beyond a reasonable doubt (your doubt is less than 5%) that the difference between your groups is too big to be due to chance alone. If there is no real difference on the parameter, you will be wrong 5% of the time if you're using p=.05 as your level of significance. You will be committing a type I error.
So, if the difference between the treatment group and the no-treatment group is too big to be due to chance alone, then some of that difference is probably due to treatment. In other words, the treatment probably had an effect.
2. not statistically significant:
you are not sure, beyond a reasonable doubt, that the difference between the groups is due to anything more than just chance so you accept the null hypothesis of no difference. This could be due to having too small a sample or the (real) effect is small and hard to detect. You could be making an incorrect decision and thereby commit a type II error.
Some important considerations:
1. Whenever we measure anything we are only obtain an estimate of the true measure (called a parameter). Our measure is called a statistic (e.g., mean GPA); e.g.,
3. Moral of the story: we can never be sure of the "truth." We only use (somewhat) unreliable estimates. The problem comes in having to make a decision about effectiveness of an IV; e.g., is this new drug effective (or more effective than the usual treatment)? You might be wrong whatever your conclusion -- you must qualify your answer in probabilities.
4. And behavioral science never proves anything!